orange-hyper v1.0.1

Release Notes

v1.0.1

README after-intro onboarding patch for Orange Hyper.

• Package version is 1.0.1.
• README version is 1.0-doc.2.
• Release date is 2026-06-18.
• Adapter JSON contract_version remains "0.1".
• This patch does not add runtime features. It keeps the v1 stable command
  surface and focuses on how users should actually start using Orange Hyper
  with an AI.
• The README sections after "Orange Hyper 소개" now lead with the AI-first
  usage model: the user talks normally, and the AI or adapter calls
  orange ... --json kernel commands when Orange Hyper is useful.
• Added a Starter Prompt that tells the AI to use Orange Hyper without making
  the user manage CLI commands, to avoid turning small explanations into
  Quests, to record intent and verification evidence for real work, to propose
  memory without auto-accepting it, to avoid MCP auto-installation, to keep
  Hook/Growth/Eval advisory, and to use Orange Kernel commands instead of
  direct .orange-hyper edits.
• Added conversation-first examples for Quest capture, Memory Proposal,
  Identity HTML plus Growth/Eval review, and MCP Advisor suggestions.
• Reframed the feature list as artifacts Orange Hyper leaves behind: Quest,
  Evidence, Memory Proposal, Accepted Memory, Knowledge Graph, Identity HTML,
  Hook Warning, MCP Suggestion, Growth Signal, and Eval Report.
• Clarified that the Knowledge Graph is an accepted project memory graph, not a
  code dependency graph. It shows accepted decision, constraint, risk,
  verification, and component memory, excludes pending/rejected proposals, and
  is currently a read-only Identity HTML preview rather than a full graph
  editor.
• Moved CLI examples to lower reference sections: "For AI / Adapter authors",
  "Manual fallback", and "Kernel command reference".
• Synced the same structure across Korean, English, Simplified Chinese, and
  Japanese README files while keeping Korean README as the source of truth.
• docs/22_V1_STABILIZATION.md now records the AI-first usage model as part of
  the v1 stable line.
• Package metadata, package lock metadata, citation metadata, and
  src/core/origin.js now target 1.0.1.

Explicitly not included

• new CLI command
• runtime behavior change
• Knowledge Graph renderer or full graph editor
• MCP automatic installation or execution
• hook automatic mutation or installation
• adapter runtime implementation
• project memory/config automatic mutation

v1.0.0

First stable boundary release for Orange Hyper.

• Package version is 1.0.0.
• README version is 1.0-doc.2.
• Release date is 2026-06-18.
• Adapter JSON contract_version remains "0.1".
• v1.0.0 is not a new feature release. It promotes the Boundary Audit and
  Stabilization Polish work from v1.0.0-alpha.0 and v1.0.0-alpha.1 into the
  first stable Orange Hyper surface.
• Stable surface fixed for v1:
  Seed Kernel, Memory Delta Proposal, Memory Graph Usability, Minimal Hook
  Preview, MCP Advisor, Growth Signal Preview, Adapter Invocation Contract, and
  Eval and Reports.
• The stable command surface remains: init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Stable readiness evidence carries forward the alpha smoke surface and is
  re-run during release prep: orange --help, doctor --json,
  eval report --json, adapter dry-run project-status --json,
  growth status --json, hook run stop --json,
  mcp suggest --query ... --json, and graph list --json.
• Hook warnings such as HOOK_CAPSULE_STALE are read-only warning evidence,
  not release failures, when the command exits successfully and preserves
  no-mutation boundaries.
• Package surface policy remains clean: include bin, src, docs, README
  files, release notes, license, provenance/security/citation metadata, README
  assets, and scripts/check-readme-sync.js; exclude tests, .orange-hyper,
  node_modules, coverage, temporary output, and local generated artifacts.

Explicitly not included

• telemetry or network upload
• LLM judge
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• adapter runtime implementation
• auto planner or auto execution loop
• project memory/config automatic mutation
• Adapter JSON Contract shape change

v1.0.0-alpha.1

v1 Stabilization Polish alpha.

• Package version is 1.0.0-alpha.1.
• README version is 1.0-doc.1.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization polish after the v1.0.0-alpha.0 Boundary Audit
  and npm alpha smoke. It does not add a new CLI command, runtime adapter, MCP
  runner, hook installer, role system, planner, LLM judge, telemetry path, or
  automatic project memory/config mutation.
• docs/22_V1_STABILIZATION.md now records the published alpha.0 smoke surface:
  orange --help, doctor --json, eval report --json,
  adapter dry-run project-status --json, growth status --json,
  hook run stop --json, mcp suggest --query ... --json, and
  graph list --json.
• The alpha.0 smoke warning policy is now explicit: local generated-state hook
  warnings such as missing or stale capsule/identity summaries are manual
  follow-up evidence, not failures, when the command exits successfully and
  preserves read-only/no-mutation flags.
• CLI diagnostic text is clearer for v1 users without changing runtime
  behavior: doctor now distinguishes warnings from blocking problems, hook
  unsupported-event errors list supported events, adapter unknown-recipe errors
  list available recipe ids, and JSON failure hints for doctor/hook/eval/adapter
  point to the safest manual next command.
• Package surface policy remains clean: npm pack --dry-run is expected to
  include bin, src, docs, README files, release notes, license, public
  metadata, assets, and scripts/check-readme-sync.js, while excluding tests,
  .orange-hyper local/generated artifacts, node_modules, coverage, and temp
  output.
• CITATION.cff, package metadata, package lock metadata, README metadata,
  roadmap notes, adapter contract wording, v1 boundary tests, and
  src/core/origin.js now target 1.0.0-alpha.1 / 1.0-doc.1 as appropriate.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v1.0.0-alpha.0

v1.0 Stabilization Candidate alpha.

• Package version is 1.0.0-alpha.0.
• README version is 1.0-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization work, not a new feature release. It re-audits the
  v0.1 through v0.8 Seed Kernel, Memory Proposal, Memory Graph, Hook, MCP
  Advisor, Growth, Adapter, and Eval boundaries before a future v1.0 stable.
• Added docs/22_V1_STABILIZATION.md as the v1 readiness report covering
  v0.1-v0.8 summaries, boundary audit results, known limitations, stable
  readiness work, non-goals, Trusted Publishing, and local-only/no-telemetry
  principles.
• README Korean, English, Simplified Chinese, and Japanese docs now mark the
  README version as 1.0-doc.0, describe v1 as a stabilization candidate, and
  expose an audited command surface marker.
• docs/10_DEVELOPMENT_ROADMAP.md now marks v1.0 Stabilization Candidate as
  the current roadmap stage.
• src/core/origin.js generator metadata now reports 1.0.0-alpha.0.
• Package metadata now targets 1.0.0-alpha.0.
• The Adapter JSON envelope remains unchanged:
  { ok, contract_version: "0.1", command, data } for success and
  { ok, contract_version: "0.1", command, error } for failure.
• The audited command surface is init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Shared .orange-hyper state remains limited to config, completed Quests,
  accepted memory proposals, and graph provenance. Capsules, traces, identity,
  hook reports, eval reports, pending/rejected proposals, and local/ remain
  local/generated state.
• Package dry-run policy remains: include bin, src, docs, README files,
  release notes, license, and public metadata; exclude tests, .orange-hyper
  local/generated artifacts, node_modules, temporary output, and coverage.

Boundary summary

• Seed Kernel stays lightweight: no graph DB, vector DB, branch workflow, or
  external API requirement.
• Memory Proposal stays proposal-first: no automatic memory write and no
  accept without explicit user command.
• Graph reads current-project accepted memory only; pending/rejected proposals
  are not graph nodes.
• Hook stays read-only / warning-first and does not repair, rebuild, generate,
  install, or mutate state automatically.
• MCP remains Advisor only and does not install, run, configure, persist keys,
  call networks, or write project memory/config.
• Growth remains preview only and keeps auto_unlock: false plus
  requires_user_approval: true.
• Adapter remains an invocation contract and does not execute recipes or
  mutate .orange-hyper directly.
• Eval remains local-only and does not upload telemetry, call LLM judges,
  estimate token savings, claim success-rate improvement, or mutate state.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v0.8.0

Local-only Eval and Reports stable release.

• Package version is 0.8.0.
• README version is 0.8-doc.2.
• Adapter JSON `contract_version...

orange-hyper v1.0.0

Release Notes

v1.0.0

First stable boundary release for Orange Hyper.

• Package version is 1.0.0.
• README version is 1.0-doc.2.
• Release date is 2026-06-18.
• Adapter JSON contract_version remains "0.1".
• v1.0.0 is not a new feature release. It promotes the Boundary Audit and
  Stabilization Polish work from v1.0.0-alpha.0 and v1.0.0-alpha.1 into the
  first stable Orange Hyper surface.
• Stable surface fixed for v1:
  Seed Kernel, Memory Delta Proposal, Memory Graph Usability, Minimal Hook
  Preview, MCP Advisor, Growth Signal Preview, Adapter Invocation Contract, and
  Eval and Reports.
• The stable command surface remains: init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Stable readiness evidence carries forward the alpha smoke surface and is
  re-run during release prep: orange --help, doctor --json,
  eval report --json, adapter dry-run project-status --json,
  growth status --json, hook run stop --json,
  mcp suggest --query ... --json, and graph list --json.
• Hook warnings such as HOOK_CAPSULE_STALE are read-only warning evidence,
  not release failures, when the command exits successfully and preserves
  no-mutation boundaries.
• Package surface policy remains clean: include bin, src, docs, README
  files, release notes, license, provenance/security/citation metadata, README
  assets, and scripts/check-readme-sync.js; exclude tests, .orange-hyper,
  node_modules, coverage, temporary output, and local generated artifacts.

Explicitly not included

• telemetry or network upload
• LLM judge
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• adapter runtime implementation
• auto planner or auto execution loop
• project memory/config automatic mutation
• Adapter JSON Contract shape change

v1.0.0-alpha.1

v1 Stabilization Polish alpha.

• Package version is 1.0.0-alpha.1.
• README version is 1.0-doc.1.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization polish after the v1.0.0-alpha.0 Boundary Audit
  and npm alpha smoke. It does not add a new CLI command, runtime adapter, MCP
  runner, hook installer, role system, planner, LLM judge, telemetry path, or
  automatic project memory/config mutation.
• docs/22_V1_STABILIZATION.md now records the published alpha.0 smoke surface:
  orange --help, doctor --json, eval report --json,
  adapter dry-run project-status --json, growth status --json,
  hook run stop --json, mcp suggest --query ... --json, and
  graph list --json.
• The alpha.0 smoke warning policy is now explicit: local generated-state hook
  warnings such as missing or stale capsule/identity summaries are manual
  follow-up evidence, not failures, when the command exits successfully and
  preserves read-only/no-mutation flags.
• CLI diagnostic text is clearer for v1 users without changing runtime
  behavior: doctor now distinguishes warnings from blocking problems, hook
  unsupported-event errors list supported events, adapter unknown-recipe errors
  list available recipe ids, and JSON failure hints for doctor/hook/eval/adapter
  point to the safest manual next command.
• Package surface policy remains clean: npm pack --dry-run is expected to
  include bin, src, docs, README files, release notes, license, public
  metadata, assets, and scripts/check-readme-sync.js, while excluding tests,
  .orange-hyper local/generated artifacts, node_modules, coverage, and temp
  output.
• CITATION.cff, package metadata, package lock metadata, README metadata,
  roadmap notes, adapter contract wording, v1 boundary tests, and
  src/core/origin.js now target 1.0.0-alpha.1 / 1.0-doc.1 as appropriate.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v1.0.0-alpha.0

v1.0 Stabilization Candidate alpha.

• Package version is 1.0.0-alpha.0.
• README version is 1.0-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization work, not a new feature release. It re-audits the
  v0.1 through v0.8 Seed Kernel, Memory Proposal, Memory Graph, Hook, MCP
  Advisor, Growth, Adapter, and Eval boundaries before a future v1.0 stable.
• Added docs/22_V1_STABILIZATION.md as the v1 readiness report covering
  v0.1-v0.8 summaries, boundary audit results, known limitations, stable
  readiness work, non-goals, Trusted Publishing, and local-only/no-telemetry
  principles.
• README Korean, English, Simplified Chinese, and Japanese docs now mark the
  README version as 1.0-doc.0, describe v1 as a stabilization candidate, and
  expose an audited command surface marker.
• docs/10_DEVELOPMENT_ROADMAP.md now marks v1.0 Stabilization Candidate as
  the current roadmap stage.
• src/core/origin.js generator metadata now reports 1.0.0-alpha.0.
• Package metadata now targets 1.0.0-alpha.0.
• The Adapter JSON envelope remains unchanged:
  { ok, contract_version: "0.1", command, data } for success and
  { ok, contract_version: "0.1", command, error } for failure.
• The audited command surface is init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Shared .orange-hyper state remains limited to config, completed Quests,
  accepted memory proposals, and graph provenance. Capsules, traces, identity,
  hook reports, eval reports, pending/rejected proposals, and local/ remain
  local/generated state.
• Package dry-run policy remains: include bin, src, docs, README files,
  release notes, license, and public metadata; exclude tests, .orange-hyper
  local/generated artifacts, node_modules, temporary output, and coverage.

Boundary summary

• Seed Kernel stays lightweight: no graph DB, vector DB, branch workflow, or
  external API requirement.
• Memory Proposal stays proposal-first: no automatic memory write and no
  accept without explicit user command.
• Graph reads current-project accepted memory only; pending/rejected proposals
  are not graph nodes.
• Hook stays read-only / warning-first and does not repair, rebuild, generate,
  install, or mutate state automatically.
• MCP remains Advisor only and does not install, run, configure, persist keys,
  call networks, or write project memory/config.
• Growth remains preview only and keeps auto_unlock: false plus
  requires_user_approval: true.
• Adapter remains an invocation contract and does not execute recipes or
  mutate .orange-hyper directly.
• Eval remains local-only and does not upload telemetry, call LLM judges,
  estimate token savings, claim success-rate improvement, or mutate state.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v0.8.0

Local-only Eval and Reports stable release.

• Package version is 0.8.0.
• README version is 0.8-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.8.0 promotes the validated v0.8.0-alpha.0 and v0.8.0-alpha.1 Eval
  and Reports surface to stable without adding telemetry, upload, judging, or
  automatic state mutation.
• Stable command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest verification, Memory Delta Proposal flow,
  accepted graph nodes, doctor diagnostics, existing local hook report
  warnings, MCP Advisor signals, Growth Signal Preview, Adapter Invocation
  Contract readiness, and identity report existence.
• eval report --json exposes report_id, schema_version: 2,
  report_kind: "eval-report", project_id, project_name,
  local_only: true, telemetry: false, network_upload: false,
  llm_judge: false, summary, sections, known_gaps, and
  unavailable_metrics.
• Report sections use only good, needs-attention, and
  insufficient-data. Every section includes a human-readable reason and
  numeric evidence_count; v0.8.0 does not produce a score, rank, or grade.
• known_gaps and unavailable_metrics keep unsupported measurements
  explicit. token.savings is unavailable, and
  success_rate.improvement is unavailable.
• token.savings stays unavailable because token counts are not collected.
  v0.8.0 does not estimate token savings.
• success_rate.improvement stays unavailable because no comparative
  task-pack outcome dataset is collected. v0.8.0 does not claim success-rate
  improvement.
• eval explain documents metric source, limitation, and unavailable reason
  for local metrics, including token.savings,
  success_rate.improvement, hook.warning.usefulness, and
  memory.acceptance_rate.
• --write-report remains the only eval write path. It writes a Markdown
  local-only report under .orange-hyper/evals/reports/ only when explicitly
  requested and rejects path/value arguments.
• Eval reports use local state only. They are generated artifacts, not project
  memory, and they do not update config, Quest, Proposal, Graph, Identity,
  hook, MCP, adapter, or growth state.
• identity build does not automatically embed eval summaries; eval reports
  remain separate explicit commands.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate impr...

orange-hyper v1.0.0-alpha.1

Release Notes

v1.0.0-alpha.1

v1 Stabilization Polish alpha.

• Package version is 1.0.0-alpha.1.
• README version is 1.0-doc.1.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization polish after the v1.0.0-alpha.0 Boundary Audit
  and npm alpha smoke. It does not add a new CLI command, runtime adapter, MCP
  runner, hook installer, role system, planner, LLM judge, telemetry path, or
  automatic project memory/config mutation.
• docs/22_V1_STABILIZATION.md now records the published alpha.0 smoke surface:
  orange --help, doctor --json, eval report --json,
  adapter dry-run project-status --json, growth status --json,
  hook run stop --json, mcp suggest --query ... --json, and
  graph list --json.
• The alpha.0 smoke warning policy is now explicit: local generated-state hook
  warnings such as missing or stale capsule/identity summaries are manual
  follow-up evidence, not failures, when the command exits successfully and
  preserves read-only/no-mutation flags.
• CLI diagnostic text is clearer for v1 users without changing runtime
  behavior: doctor now distinguishes warnings from blocking problems, hook
  unsupported-event errors list supported events, adapter unknown-recipe errors
  list available recipe ids, and JSON failure hints for doctor/hook/eval/adapter
  point to the safest manual next command.
• Package surface policy remains clean: npm pack --dry-run is expected to
  include bin, src, docs, README files, release notes, license, public
  metadata, assets, and scripts/check-readme-sync.js, while excluding tests,
  .orange-hyper local/generated artifacts, node_modules, coverage, and temp
  output.
• CITATION.cff, package metadata, package lock metadata, README metadata,
  roadmap notes, adapter contract wording, v1 boundary tests, and
  src/core/origin.js now target 1.0.0-alpha.1 / 1.0-doc.1 as appropriate.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v1.0.0-alpha.0

v1.0 Stabilization Candidate alpha.

• Package version is 1.0.0-alpha.0.
• README version is 1.0-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization work, not a new feature release. It re-audits the
  v0.1 through v0.8 Seed Kernel, Memory Proposal, Memory Graph, Hook, MCP
  Advisor, Growth, Adapter, and Eval boundaries before a future v1.0 stable.
• Added docs/22_V1_STABILIZATION.md as the v1 readiness report covering
  v0.1-v0.8 summaries, boundary audit results, known limitations, stable
  readiness work, non-goals, Trusted Publishing, and local-only/no-telemetry
  principles.
• README Korean, English, Simplified Chinese, and Japanese docs now mark the
  README version as 1.0-doc.0, describe v1 as a stabilization candidate, and
  expose an audited command surface marker.
• docs/10_DEVELOPMENT_ROADMAP.md now marks v1.0 Stabilization Candidate as
  the current roadmap stage.
• src/core/origin.js generator metadata now reports 1.0.0-alpha.0.
• Package metadata now targets 1.0.0-alpha.0.
• The Adapter JSON envelope remains unchanged:
  { ok, contract_version: "0.1", command, data } for success and
  { ok, contract_version: "0.1", command, error } for failure.
• The audited command surface is init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Shared .orange-hyper state remains limited to config, completed Quests,
  accepted memory proposals, and graph provenance. Capsules, traces, identity,
  hook reports, eval reports, pending/rejected proposals, and local/ remain
  local/generated state.
• Package dry-run policy remains: include bin, src, docs, README files,
  release notes, license, and public metadata; exclude tests, .orange-hyper
  local/generated artifacts, node_modules, temporary output, and coverage.

Boundary summary

• Seed Kernel stays lightweight: no graph DB, vector DB, branch workflow, or
  external API requirement.
• Memory Proposal stays proposal-first: no automatic memory write and no
  accept without explicit user command.
• Graph reads current-project accepted memory only; pending/rejected proposals
  are not graph nodes.
• Hook stays read-only / warning-first and does not repair, rebuild, generate,
  install, or mutate state automatically.
• MCP remains Advisor only and does not install, run, configure, persist keys,
  call networks, or write project memory/config.
• Growth remains preview only and keeps auto_unlock: false plus
  requires_user_approval: true.
• Adapter remains an invocation contract and does not execute recipes or
  mutate .orange-hyper directly.
• Eval remains local-only and does not upload telemetry, call LLM judges,
  estimate token savings, claim success-rate improvement, or mutate state.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v0.8.0

Local-only Eval and Reports stable release.

• Package version is 0.8.0.
• README version is 0.8-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.8.0 promotes the validated v0.8.0-alpha.0 and v0.8.0-alpha.1 Eval
  and Reports surface to stable without adding telemetry, upload, judging, or
  automatic state mutation.
• Stable command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest verification, Memory Delta Proposal flow,
  accepted graph nodes, doctor diagnostics, existing local hook report
  warnings, MCP Advisor signals, Growth Signal Preview, Adapter Invocation
  Contract readiness, and identity report existence.
• eval report --json exposes report_id, schema_version: 2,
  report_kind: "eval-report", project_id, project_name,
  local_only: true, telemetry: false, network_upload: false,
  llm_judge: false, summary, sections, known_gaps, and
  unavailable_metrics.
• Report sections use only good, needs-attention, and
  insufficient-data. Every section includes a human-readable reason and
  numeric evidence_count; v0.8.0 does not produce a score, rank, or grade.
• known_gaps and unavailable_metrics keep unsupported measurements
  explicit. token.savings is unavailable, and
  success_rate.improvement is unavailable.
• token.savings stays unavailable because token counts are not collected.
  v0.8.0 does not estimate token savings.
• success_rate.improvement stays unavailable because no comparative
  task-pack outcome dataset is collected. v0.8.0 does not claim success-rate
  improvement.
• eval explain documents metric source, limitation, and unavailable reason
  for local metrics, including token.savings,
  success_rate.improvement, hook.warning.usefulness, and
  memory.acceptance_rate.
• --write-report remains the only eval write path. It writes a Markdown
  local-only report under .orange-hyper/evals/reports/ only when explicitly
  requested and rejects path/value arguments.
• Eval reports use local state only. They are generated artifacts, not project
  memory, and they do not update config, Quest, Proposal, Graph, Identity,
  hook, MCP, adapter, or growth state.
• identity build does not automatically embed eval summaries; eval reports
  remain separate explicit commands.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop

v0.8.0-alpha.1

Eval Report Quality Hardening alpha.

• Package version is 0.8.0-alpha.1.
• README version is 0.8-doc.1.
• Adapter JSON contract_version remains "0.1".
• eval report now includes a top summary with project identity,
  generated_at, local-only report mode, total section count,
  needs-attention count, insufficient-data count, and explicit no
  telemetry / no network / no LLM judge flags.
• eval report --json now exposes report_id, schema_version: 2,
  project_id, project_name, local_only: true, telemetry: false,
  network_upload: false, llm_judge: false, sections, known_gaps,
  and unavailable_metrics.
• Report sections now include status, reason, and evidence_count.
  Allowed statuses remain only good, needs-attention, and
  insufficient-data; no score or grade is produced.
• eval explain now includes a limitation for each metric and documents
  token.savings, success_rate.improvement,
  hook.warning.usefulness, and memory.acceptance_rate without estimating
  missing data.
• --write-report remains the only eval write path. It writes Markdown only
  under .orange-hyper/evals/reports/ with an eval-report- timestamped
  filename and still rejects path/value arguments.
• identity build does not automatically include eval summaries in this
  alpha; eval reports remain separate explicit commands.
• Tests cover report summary, section status metadata, unavailable token and
  success metrics, JSON report schema, local-only report writing,
  explain limitations, and identity/eval separation.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• toke...

orange-hyper v1.0.0-alpha.0

Release Notes

v1.0.0-alpha.0

v1.0 Stabilization Candidate alpha.

• Package version is 1.0.0-alpha.0.
• README version is 1.0-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha is stabilization work, not a new feature release. It re-audits the
  v0.1 through v0.8 Seed Kernel, Memory Proposal, Memory Graph, Hook, MCP
  Advisor, Growth, Adapter, and Eval boundaries before a future v1.0 stable.
• Added docs/22_V1_STABILIZATION.md as the v1 readiness report covering
  v0.1-v0.8 summaries, boundary audit results, known limitations, stable
  readiness work, non-goals, Trusted Publishing, and local-only/no-telemetry
  principles.
• README Korean, English, Simplified Chinese, and Japanese docs now mark the
  README version as 1.0-doc.0, describe v1 as a stabilization candidate, and
  expose an audited command surface marker.
• docs/10_DEVELOPMENT_ROADMAP.md now marks v1.0 Stabilization Candidate as
  the current roadmap stage.
• src/core/origin.js generator metadata now reports 1.0.0-alpha.0.
• Package metadata now targets 1.0.0-alpha.0.
• The Adapter JSON envelope remains unchanged:
  { ok, contract_version: "0.1", command, data } for success and
  { ok, contract_version: "0.1", command, error } for failure.
• The audited command surface is init, quest, route, capsule,
  remember, graph, hook, mcp, growth, adapter, eval, doctor,
  and identity.
• Shared .orange-hyper state remains limited to config, completed Quests,
  accepted memory proposals, and graph provenance. Capsules, traces, identity,
  hook reports, eval reports, pending/rejected proposals, and local/ remain
  local/generated state.
• Package dry-run policy remains: include bin, src, docs, README files,
  release notes, license, and public metadata; exclude tests, .orange-hyper
  local/generated artifacts, node_modules, temporary output, and coverage.

Boundary summary

• Seed Kernel stays lightweight: no graph DB, vector DB, branch workflow, or
  external API requirement.
• Memory Proposal stays proposal-first: no automatic memory write and no
  accept without explicit user command.
• Graph reads current-project accepted memory only; pending/rejected proposals
  are not graph nodes.
• Hook stays read-only / warning-first and does not repair, rebuild, generate,
  install, or mutate state automatically.
• MCP remains Advisor only and does not install, run, configure, persist keys,
  call networks, or write project memory/config.
• Growth remains preview only and keeps auto_unlock: false plus
  requires_user_approval: true.
• Adapter remains an invocation contract and does not execute recipes or
  mutate .orange-hyper directly.
• Eval remains local-only and does not upload telemetry, call LLM judges,
  estimate token savings, claim success-rate improvement, or mutate state.

Explicitly not included

• new CLI feature
• MCP automatic installation or execution
• hook automatic mutation or installation
• role automatic creation
• subagent orchestration
• auto planner or auto execution loop
• LLM judge
• telemetry or network upload
• adapter runtime implementation
• project memory/config automatic mutation

v0.8.0

Local-only Eval and Reports stable release.

• Package version is 0.8.0.
• README version is 0.8-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.8.0 promotes the validated v0.8.0-alpha.0 and v0.8.0-alpha.1 Eval
  and Reports surface to stable without adding telemetry, upload, judging, or
  automatic state mutation.
• Stable command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest verification, Memory Delta Proposal flow,
  accepted graph nodes, doctor diagnostics, existing local hook report
  warnings, MCP Advisor signals, Growth Signal Preview, Adapter Invocation
  Contract readiness, and identity report existence.
• eval report --json exposes report_id, schema_version: 2,
  report_kind: "eval-report", project_id, project_name,
  local_only: true, telemetry: false, network_upload: false,
  llm_judge: false, summary, sections, known_gaps, and
  unavailable_metrics.
• Report sections use only good, needs-attention, and
  insufficient-data. Every section includes a human-readable reason and
  numeric evidence_count; v0.8.0 does not produce a score, rank, or grade.
• known_gaps and unavailable_metrics keep unsupported measurements
  explicit. token.savings is unavailable, and
  success_rate.improvement is unavailable.
• token.savings stays unavailable because token counts are not collected.
  v0.8.0 does not estimate token savings.
• success_rate.improvement stays unavailable because no comparative
  task-pack outcome dataset is collected. v0.8.0 does not claim success-rate
  improvement.
• eval explain documents metric source, limitation, and unavailable reason
  for local metrics, including token.savings,
  success_rate.improvement, hook.warning.usefulness, and
  memory.acceptance_rate.
• --write-report remains the only eval write path. It writes a Markdown
  local-only report under .orange-hyper/evals/reports/ only when explicitly
  requested and rejects path/value arguments.
• Eval reports use local state only. They are generated artifacts, not project
  memory, and they do not update config, Quest, Proposal, Graph, Identity,
  hook, MCP, adapter, or growth state.
• identity build does not automatically embed eval summaries; eval reports
  remain separate explicit commands.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop

v0.8.0-alpha.1

Eval Report Quality Hardening alpha.

• Package version is 0.8.0-alpha.1.
• README version is 0.8-doc.1.
• Adapter JSON contract_version remains "0.1".
• eval report now includes a top summary with project identity,
  generated_at, local-only report mode, total section count,
  needs-attention count, insufficient-data count, and explicit no
  telemetry / no network / no LLM judge flags.
• eval report --json now exposes report_id, schema_version: 2,
  project_id, project_name, local_only: true, telemetry: false,
  network_upload: false, llm_judge: false, sections, known_gaps,
  and unavailable_metrics.
• Report sections now include status, reason, and evidence_count.
  Allowed statuses remain only good, needs-attention, and
  insufficient-data; no score or grade is produced.
• eval explain now includes a limitation for each metric and documents
  token.savings, success_rate.improvement,
  hook.warning.usefulness, and memory.acceptance_rate without estimating
  missing data.
• --write-report remains the only eval write path. It writes Markdown only
  under .orange-hyper/evals/reports/ with an eval-report- timestamped
  filename and still rejects path/value arguments.
• identity build does not automatically include eval summaries in this
  alpha; eval reports remain separate explicit commands.
• Tests cover report summary, section status metadata, unavailable token and
  success metrics, JSON report schema, local-only report writing,
  explain limitations, and identity/eval separation.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop

v0.8.0-alpha.0

Eval and Reports Preview alpha.

• Package version is 0.8.0-alpha.0.
• README version is 0.8-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the local-only Eval and Reports Preview command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• New JSON command ids are eval.snapshot, eval.report, and
  eval.explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest counts, completed/verified/unverified
  Quest counts, Memory Delta Proposal flow, accepted graph node count, doctor
  errors/warnings, latest local hook report warnings when available, MCP
  Advisor availability, growth candidate count, adapter recipe count, and
  identity report existence.
• eval report emits a Markdown local report to stdout by default. It writes
  only when --write-report is explicit, and the write path is limited to
  .orange-hyper/evals/reports/.
• eval explain shows where each metric came from and keeps unavailable
  metrics such as token savings and success-rate improvement unestimated.
• Eval sections use conservative signal summaries: good,
  needs-attention, and insufficient-data. v0.8 does not create an overall
  grade.
• .orange-hyper/evals/reports/ is now treated as local/generated ignored
  state.
• Tests cover eval human/JSON output, no default file creation, opt-in report
  writes, path traversal rejection, no project memory/config mutation,
  unavailable metrics, and JSON envelope stability.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• model capability improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution ...

orange-hyper v0.8.0-alpha.1

Release Notes

v0.8.0-alpha.1

Eval Report Quality Hardening alpha.

• Package version is 0.8.0-alpha.1.
• README version is 0.8-doc.1.
• Adapter JSON contract_version remains "0.1".
• eval report now includes a top summary with project identity,
  generated_at, local-only report mode, total section count,
  needs-attention count, insufficient-data count, and explicit no
  telemetry / no network / no LLM judge flags.
• eval report --json now exposes report_id, schema_version: 2,
  project_id, project_name, local_only: true, telemetry: false,
  network_upload: false, llm_judge: false, sections, known_gaps,
  and unavailable_metrics.
• Report sections now include status, reason, and evidence_count.
  Allowed statuses remain only good, needs-attention, and
  insufficient-data; no score or grade is produced.
• eval explain now includes a limitation for each metric and documents
  token.savings, success_rate.improvement,
  hook.warning.usefulness, and memory.acceptance_rate without estimating
  missing data.
• --write-report remains the only eval write path. It writes Markdown only
  under .orange-hyper/evals/reports/ with an eval-report- timestamped
  filename and still rejects path/value arguments.
• identity build does not automatically include eval summaries in this
  alpha; eval reports remain separate explicit commands.
• Tests cover report summary, section status metadata, unavailable token and
  success metrics, JSON report schema, local-only report writing,
  explain limitations, and identity/eval separation.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop

v0.8.0-alpha.0

Eval and Reports Preview alpha.

• Package version is 0.8.0-alpha.0.
• README version is 0.8-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the local-only Eval and Reports Preview command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• New JSON command ids are eval.snapshot, eval.report, and
  eval.explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest counts, completed/verified/unverified
  Quest counts, Memory Delta Proposal flow, accepted graph node count, doctor
  errors/warnings, latest local hook report warnings when available, MCP
  Advisor availability, growth candidate count, adapter recipe count, and
  identity report existence.
• eval report emits a Markdown local report to stdout by default. It writes
  only when --write-report is explicit, and the write path is limited to
  .orange-hyper/evals/reports/.
• eval explain shows where each metric came from and keeps unavailable
  metrics such as token savings and success-rate improvement unestimated.
• Eval sections use conservative signal summaries: good,
  needs-attention, and insufficient-data. v0.8 does not create an overall
  grade.
• .orange-hyper/evals/reports/ is now treated as local/generated ignored
  state.
• Tests cover eval human/JSON output, no default file creation, opt-in report
  writes, path traversal rejection, no project memory/config mutation,
  unavailable metrics, and JSON envelope stability.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• model capability improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop
• HTML dashboard

v0.7.0

Adapter Invocation Contract stable release.

• Package version is 0.7.0.
• README version is 0.7-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.7.0 promotes the validated v0.7.0-alpha.0 and v0.7.0-alpha.1 Adapter
  Invocation Contract surface to stable without adding adapter runtime
  execution.
• Stable command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• Built-in adapter recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• adapter dry-run returns a dry-run execution plan with recipe_id,
  dry_run: true, executed: false, steps, commands,
  required_inputs, missing_inputs, safety_flags,
  expected_contract_version, next_user_decision, mutation_policy, and
  adapter_rules.
• Recipe inputs declare whether values come from user, previous_step, or
  project_state, so adapters can distinguish user prompts from prior JSON
  outputs and existing kernel state.
• Every recipe step declares mutates_project_state and
  requires_user_approval; mutating steps require approval, while read-only
  status/advice/hook checks do not.
• Safety flags remain fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• The stable Adapter Layer remains an invocation contract: adapters call Orange
  CLI --json, parse JSON envelopes only, avoid .orange-hyper direct file
  mutation, and do not duplicate kernel state logic.

Explicitly not included

• actual adapter runtime
• Codex/Claude-specific adapter automatic installation
• direct .orange-hyper file mutation
• human output parsing
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• auto planner or auto execution loop
• recipe automatic execution

v0.7.0-alpha.1

Adapter Recipe Quality Hardening.

• Package version is 0.7.0-alpha.1.
• README version is 0.7-doc.1.
• Adapter JSON contract_version remains "0.1".
• Recipe steps now carry step_index and structured input_requirements.
• Placeholder inputs such as <request>, <quest-id>, <proposal-id>, and
  <query> now declare input_source: user, previous_step, or
  project_state.
• adapter dry-run now returns recipe_id, dry_run: true,
  executed: false, steps, required_inputs, missing_inputs,
  safety_flags, expected_contract_version, and next_user_decision.
• Dry-run keeps a commands alias for compatibility with the initial v0.7
  alpha while making steps the clearer execution-plan field.
• Recipe metadata now pins mutating commands to requires_user_approval: true
  and non-mutating read commands to requires_user_approval: false.
• hook run session-start and hook run stop remain read-only when
  --write-report is omitted, so their recipe steps are
  mutates_project_state: false and requires_user_approval: false.
• Tests now verify command ids, placeholder input sources, missing dry-run
  inputs, mutation approval flags, read-only approval flags, no dry-run
  .orange-hyper mutation, unknown recipe failures, and the no-human-output
  parsing guard.
• docs/20_ADAPTER_LAYER.md documents the recipe quality standard.
• docs/16_ADAPTER_CONTRACT.md includes a richer dry-run JSON example.

Explicitly still not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.7.0-alpha.0

Adapter Invocation Contract alpha.

• Package version is 0.7.0-alpha.0.
• README version is 0.7-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the Adapter Layer command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• New JSON command ids are adapter.list, adapter.show, and
  adapter.dryRun.
• Built-in recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• Each recipe declares id, title, purpose, when_to_use, commands,
  required_inputs, outputs, safety_rules, forbidden_actions,
  expected_contract_version, and safety_flags.
• Each recipe command step declares command, why, required_input,
  expected_json_command_id, mutates_project_state, and
  requires_user_approval.
• Safety flags are fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• adapter dry-run does not execute recipe commands and does not modify
  .orange-hyper.
• docs/20_ADAPTER_LAYER.md documents that natural-language and skill layers
  must call the kernel through Orange CLI --json commands, must not parse
  human output, and must not duplicate kernel state logic.

Explicitly not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signa...

orange-hyper v0.8.0-alpha.0

Release Notes

v0.8.0-alpha.0

Eval and Reports Preview alpha.

• Package version is 0.8.0-alpha.0.
• README version is 0.8-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the local-only Eval and Reports Preview command surface:
  orange eval snapshot, orange eval report, and orange eval explain.
• New JSON command ids are eval.snapshot, eval.report, and
  eval.explain.
• eval snapshot reads current local .orange-hyper project state and
  summarizes project identity, Quest counts, completed/verified/unverified
  Quest counts, Memory Delta Proposal flow, accepted graph node count, doctor
  errors/warnings, latest local hook report warnings when available, MCP
  Advisor availability, growth candidate count, adapter recipe count, and
  identity report existence.
• eval report emits a Markdown local report to stdout by default. It writes
  only when --write-report is explicit, and the write path is limited to
  .orange-hyper/evals/reports/.
• eval explain shows where each metric came from and keeps unavailable
  metrics such as token savings and success-rate improvement unestimated.
• Eval sections use conservative signal summaries: good,
  needs-attention, and insufficient-data. v0.8 does not create an overall
  grade.
• .orange-hyper/evals/reports/ is now treated as local/generated ignored
  state.
• Tests cover eval human/JSON output, no default file creation, opt-in report
  writes, path traversal rejection, no project memory/config mutation,
  unavailable metrics, and JSON envelope stability.

Explicitly not included

• external telemetry
• network upload or API calls
• LLM judge calls
• token savings estimation
• success-rate improvement claims
• model capability improvement claims
• project memory or config automatic mutation
• automatic Quest, Proposal, Graph, or Identity creation
• MCP automatic execution
• hook automatic execution
• subagent execution
• auto planner or auto execution loop
• HTML dashboard

v0.7.0

Adapter Invocation Contract stable release.

• Package version is 0.7.0.
• README version is 0.7-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.7.0 promotes the validated v0.7.0-alpha.0 and v0.7.0-alpha.1 Adapter
  Invocation Contract surface to stable without adding adapter runtime
  execution.
• Stable command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• Built-in adapter recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• adapter dry-run returns a dry-run execution plan with recipe_id,
  dry_run: true, executed: false, steps, commands,
  required_inputs, missing_inputs, safety_flags,
  expected_contract_version, next_user_decision, mutation_policy, and
  adapter_rules.
• Recipe inputs declare whether values come from user, previous_step, or
  project_state, so adapters can distinguish user prompts from prior JSON
  outputs and existing kernel state.
• Every recipe step declares mutates_project_state and
  requires_user_approval; mutating steps require approval, while read-only
  status/advice/hook checks do not.
• Safety flags remain fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• The stable Adapter Layer remains an invocation contract: adapters call Orange
  CLI --json, parse JSON envelopes only, avoid .orange-hyper direct file
  mutation, and do not duplicate kernel state logic.

Explicitly not included

• actual adapter runtime
• Codex/Claude-specific adapter automatic installation
• direct .orange-hyper file mutation
• human output parsing
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• auto planner or auto execution loop
• recipe automatic execution

v0.7.0-alpha.1

Adapter Recipe Quality Hardening.

• Package version is 0.7.0-alpha.1.
• README version is 0.7-doc.1.
• Adapter JSON contract_version remains "0.1".
• Recipe steps now carry step_index and structured input_requirements.
• Placeholder inputs such as <request>, <quest-id>, <proposal-id>, and
  <query> now declare input_source: user, previous_step, or
  project_state.
• adapter dry-run now returns recipe_id, dry_run: true,
  executed: false, steps, required_inputs, missing_inputs,
  safety_flags, expected_contract_version, and next_user_decision.
• Dry-run keeps a commands alias for compatibility with the initial v0.7
  alpha while making steps the clearer execution-plan field.
• Recipe metadata now pins mutating commands to requires_user_approval: true
  and non-mutating read commands to requires_user_approval: false.
• hook run session-start and hook run stop remain read-only when
  --write-report is omitted, so their recipe steps are
  mutates_project_state: false and requires_user_approval: false.
• Tests now verify command ids, placeholder input sources, missing dry-run
  inputs, mutation approval flags, read-only approval flags, no dry-run
  .orange-hyper mutation, unknown recipe failures, and the no-human-output
  parsing guard.
• docs/20_ADAPTER_LAYER.md documents the recipe quality standard.
• docs/16_ADAPTER_CONTRACT.md includes a richer dry-run JSON example.

Explicitly still not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.7.0-alpha.0

Adapter Invocation Contract alpha.

• Package version is 0.7.0-alpha.0.
• README version is 0.7-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the Adapter Layer command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• New JSON command ids are adapter.list, adapter.show, and
  adapter.dryRun.
• Built-in recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• Each recipe declares id, title, purpose, when_to_use, commands,
  required_inputs, outputs, safety_rules, forbidden_actions,
  expected_contract_version, and safety_flags.
• Each recipe command step declares command, why, required_input,
  expected_json_command_id, mutates_project_state, and
  requires_user_approval.
• Safety flags are fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• adapter dry-run does not execute recipe commands and does not modify
  .orange-hyper.
• docs/20_ADAPTER_LAYER.md documents that natural-language and skill layers
  must call the kernel through Orange CLI --json commands, must not parse
  human output, and must not duplicate kernel state logic.

Explicitly not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signal Preview surface to stable without adding automatic role, tool, hook,
  subagent, planner, workflow, config, graph, or project-memory mutation.
• Stable command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• growth status reports conservative project growth state from Quest, Route,
  accepted Memory Graph, Hook warning, Doctor, and MCP-advisor-shaped
  documentation/API freshness signals.
• growth suggest creates candidates only from repeated source-backed
  evidence. Each candidate includes score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• growth explain reports the same deterministic rule path with source-backed
  evidence instead of LLM calls, network calls, MCP calls, or planner loops.
• No-candidate negative cases are expected when evidence is too thin; a single
  Quest or one-off generic signal should not create a candidate.
• Positive repo smoke showed repeated documentation, verification, and memory
  hygiene evidence producing verification-discipline, documentation-focus,
  and memory-hygiene candidates while keeping growthLevel conservatively at
  sprout.
• identity build includes a Growth Signal Preview summary with growth level
  reason, candidate count, top candidates, growth confidence summary, and
  No automatic unlocks.
• At the v0.6.0 cut, v0.7 Adapter Layer was the next roadmap step after this
  stable Growth Signal Preview boundary.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.1

Growth Signal Preview quality hardening.

• Package version is 0.6.0-alpha.1.
• README version i...

orange-hyper v0.7.0

Release Notes

v0.7.0

Adapter Invocation Contract stable release.

• Package version is 0.7.0.
• README version is 0.7-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.7.0 promotes the validated v0.7.0-alpha.0 and v0.7.0-alpha.1 Adapter
  Invocation Contract surface to stable without adding adapter runtime
  execution.
• Stable command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• Built-in adapter recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• adapter dry-run returns a dry-run execution plan with recipe_id,
  dry_run: true, executed: false, steps, commands,
  required_inputs, missing_inputs, safety_flags,
  expected_contract_version, next_user_decision, mutation_policy, and
  adapter_rules.
• Recipe inputs declare whether values come from user, previous_step, or
  project_state, so adapters can distinguish user prompts from prior JSON
  outputs and existing kernel state.
• Every recipe step declares mutates_project_state and
  requires_user_approval; mutating steps require approval, while read-only
  status/advice/hook checks do not.
• Safety flags remain fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• The stable Adapter Layer remains an invocation contract: adapters call Orange
  CLI --json, parse JSON envelopes only, avoid .orange-hyper direct file
  mutation, and do not duplicate kernel state logic.

Explicitly not included

• actual adapter runtime
• Codex/Claude-specific adapter automatic installation
• direct .orange-hyper file mutation
• human output parsing
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• auto planner or auto execution loop
• recipe automatic execution

v0.7.0-alpha.1

Adapter Recipe Quality Hardening.

• Package version is 0.7.0-alpha.1.
• README version is 0.7-doc.1.
• Adapter JSON contract_version remains "0.1".
• Recipe steps now carry step_index and structured input_requirements.
• Placeholder inputs such as <request>, <quest-id>, <proposal-id>, and
  <query> now declare input_source: user, previous_step, or
  project_state.
• adapter dry-run now returns recipe_id, dry_run: true,
  executed: false, steps, required_inputs, missing_inputs,
  safety_flags, expected_contract_version, and next_user_decision.
• Dry-run keeps a commands alias for compatibility with the initial v0.7
  alpha while making steps the clearer execution-plan field.
• Recipe metadata now pins mutating commands to requires_user_approval: true
  and non-mutating read commands to requires_user_approval: false.
• hook run session-start and hook run stop remain read-only when
  --write-report is omitted, so their recipe steps are
  mutates_project_state: false and requires_user_approval: false.
• Tests now verify command ids, placeholder input sources, missing dry-run
  inputs, mutation approval flags, read-only approval flags, no dry-run
  .orange-hyper mutation, unknown recipe failures, and the no-human-output
  parsing guard.
• docs/20_ADAPTER_LAYER.md documents the recipe quality standard.
• docs/16_ADAPTER_CONTRACT.md includes a richer dry-run JSON example.

Explicitly still not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.7.0-alpha.0

Adapter Invocation Contract alpha.

• Package version is 0.7.0-alpha.0.
• README version is 0.7-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the Adapter Layer command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• New JSON command ids are adapter.list, adapter.show, and
  adapter.dryRun.
• Built-in recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• Each recipe declares id, title, purpose, when_to_use, commands,
  required_inputs, outputs, safety_rules, forbidden_actions,
  expected_contract_version, and safety_flags.
• Each recipe command step declares command, why, required_input,
  expected_json_command_id, mutates_project_state, and
  requires_user_approval.
• Safety flags are fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• adapter dry-run does not execute recipe commands and does not modify
  .orange-hyper.
• docs/20_ADAPTER_LAYER.md documents that natural-language and skill layers
  must call the kernel through Orange CLI --json commands, must not parse
  human output, and must not duplicate kernel state logic.

Explicitly not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signal Preview surface to stable without adding automatic role, tool, hook,
  subagent, planner, workflow, config, graph, or project-memory mutation.
• Stable command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• growth status reports conservative project growth state from Quest, Route,
  accepted Memory Graph, Hook warning, Doctor, and MCP-advisor-shaped
  documentation/API freshness signals.
• growth suggest creates candidates only from repeated source-backed
  evidence. Each candidate includes score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• growth explain reports the same deterministic rule path with source-backed
  evidence instead of LLM calls, network calls, MCP calls, or planner loops.
• No-candidate negative cases are expected when evidence is too thin; a single
  Quest or one-off generic signal should not create a candidate.
• Positive repo smoke showed repeated documentation, verification, and memory
  hygiene evidence producing verification-discipline, documentation-focus,
  and memory-hygiene candidates while keeping growthLevel conservatively at
  sprout.
• identity build includes a Growth Signal Preview summary with growth level
  reason, candidate count, top candidates, growth confidence summary, and
  No automatic unlocks.
• At the v0.6.0 cut, v0.7 Adapter Layer was the next roadmap step after this
  stable Growth Signal Preview boundary.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.1

Growth Signal Preview quality hardening.

• Package version is 0.6.0-alpha.1.
• README version is 0.6-doc.1.
• Adapter JSON contract_version remains "0.1".
• growthLevel calibration is more conservative and no longer advances from
  accepted node count alone.
• Level calculation now considers node type diversity, verified Quest ratio,
  repeated evidence count, pending proposal load, doctor ok state, and active
  project boundary.
• Growth candidates now include deterministic score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• Candidate thresholds now require repeated source-backed evidence; a single
  generic string match is not enough to create a candidate.
• Backend/API false positives from generic API/Route Contract text are pinned
  by regression tests.
• growth explain now reports source-backed evidence for each candidate,
  including Quest, node, route layer, Hook warning, and MCP-shaped signal
  source fields where available.
• identity build now shows growth level reason, candidate count, top
  candidates, growth confidence summary, and No automatic unlocks.

Explicitly still not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.0

Growth Signal Preview alpha.

• Package version is 0.6.0-alpha.0.
• README version is 0.6-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds a read-only Growth Signal Preview command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• New JSON command ids are growth.status, growth.suggest, and
  growth.explain.
• growth status summarizes project identity, accepted memory nodes, node
  type distribution, route/layer distribution, verified/unverified Quest
  ratio, pending memory...

orange-hyper v0.7.0-alpha.1

Release Notes

v0.7.0-alpha.1

Adapter Recipe Quality Hardening.

• Package version is 0.7.0-alpha.1.
• README version is 0.7-doc.1.
• Adapter JSON contract_version remains "0.1".
• Recipe steps now carry step_index and structured input_requirements.
• Placeholder inputs such as <request>, <quest-id>, <proposal-id>, and
  <query> now declare input_source: user, previous_step, or
  project_state.
• adapter dry-run now returns recipe_id, dry_run: true,
  executed: false, steps, required_inputs, missing_inputs,
  safety_flags, expected_contract_version, and next_user_decision.
• Dry-run keeps a commands alias for compatibility with the initial v0.7
  alpha while making steps the clearer execution-plan field.
• Recipe metadata now pins mutating commands to requires_user_approval: true
  and non-mutating read commands to requires_user_approval: false.
• hook run session-start and hook run stop remain read-only when
  --write-report is omitted, so their recipe steps are
  mutates_project_state: false and requires_user_approval: false.
• Tests now verify command ids, placeholder input sources, missing dry-run
  inputs, mutation approval flags, read-only approval flags, no dry-run
  .orange-hyper mutation, unknown recipe failures, and the no-human-output
  parsing guard.
• docs/20_ADAPTER_LAYER.md documents the recipe quality standard.
• docs/16_ADAPTER_CONTRACT.md includes a richer dry-run JSON example.

Explicitly still not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.7.0-alpha.0

Adapter Invocation Contract alpha.

• Package version is 0.7.0-alpha.0.
• README version is 0.7-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the Adapter Layer command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• New JSON command ids are adapter.list, adapter.show, and
  adapter.dryRun.
• Built-in recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• Each recipe declares id, title, purpose, when_to_use, commands,
  required_inputs, outputs, safety_rules, forbidden_actions,
  expected_contract_version, and safety_flags.
• Each recipe command step declares command, why, required_input,
  expected_json_command_id, mutates_project_state, and
  requires_user_approval.
• Safety flags are fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• adapter dry-run does not execute recipe commands and does not modify
  .orange-hyper.
• docs/20_ADAPTER_LAYER.md documents that natural-language and skill layers
  must call the kernel through Orange CLI --json commands, must not parse
  human output, and must not duplicate kernel state logic.

Explicitly not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signal Preview surface to stable without adding automatic role, tool, hook,
  subagent, planner, workflow, config, graph, or project-memory mutation.
• Stable command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• growth status reports conservative project growth state from Quest, Route,
  accepted Memory Graph, Hook warning, Doctor, and MCP-advisor-shaped
  documentation/API freshness signals.
• growth suggest creates candidates only from repeated source-backed
  evidence. Each candidate includes score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• growth explain reports the same deterministic rule path with source-backed
  evidence instead of LLM calls, network calls, MCP calls, or planner loops.
• No-candidate negative cases are expected when evidence is too thin; a single
  Quest or one-off generic signal should not create a candidate.
• Positive repo smoke showed repeated documentation, verification, and memory
  hygiene evidence producing verification-discipline, documentation-focus,
  and memory-hygiene candidates while keeping growthLevel conservatively at
  sprout.
• identity build includes a Growth Signal Preview summary with growth level
  reason, candidate count, top candidates, growth confidence summary, and
  No automatic unlocks.
• At the v0.6.0 cut, v0.7 Adapter Layer was the next roadmap step after this
  stable Growth Signal Preview boundary.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.1

Growth Signal Preview quality hardening.

• Package version is 0.6.0-alpha.1.
• README version is 0.6-doc.1.
• Adapter JSON contract_version remains "0.1".
• growthLevel calibration is more conservative and no longer advances from
  accepted node count alone.
• Level calculation now considers node type diversity, verified Quest ratio,
  repeated evidence count, pending proposal load, doctor ok state, and active
  project boundary.
• Growth candidates now include deterministic score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• Candidate thresholds now require repeated source-backed evidence; a single
  generic string match is not enough to create a candidate.
• Backend/API false positives from generic API/Route Contract text are pinned
  by regression tests.
• growth explain now reports source-backed evidence for each candidate,
  including Quest, node, route layer, Hook warning, and MCP-shaped signal
  source fields where available.
• identity build now shows growth level reason, candidate count, top
  candidates, growth confidence summary, and No automatic unlocks.

Explicitly still not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.0

Growth Signal Preview alpha.

• Package version is 0.6.0-alpha.0.
• README version is 0.6-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds a read-only Growth Signal Preview command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• New JSON command ids are growth.status, growth.suggest, and
  growth.explain.
• growth status summarizes project identity, accepted memory nodes, node
  type distribution, route/layer distribution, verified/unverified Quest
  ratio, pending memory proposals, Hook warning summary, MCP advisor signal
  summary, and decorative growthLevel.
• growth suggest returns deterministic candidates such as verification
  discipline, memory hygiene, backend/API focus, documentation focus, MCP
  documentation advisor readiness, and hook hygiene.
• Every growth candidate keeps auto_unlock: false and
  requires_user_approval: true.
• growth explain reports the deterministic evidence/rule path for each
  candidate without LLM calls, network calls, MCP calls, or planner loops.
• identity build now includes a compact Growth Signal Preview summary in the
  generated HTML and summary JSON.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.5.0

MCP Advisor stable release.

• Package version is 0.5.0.
• README version is 0.5-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.5.0 promotes the validated MCP Advisor alpha surface to stable without
  adding MCP integration or automatic execution behavior.
• Stable command surface:
  orange mcp list, orange mcp show <mcp-id>, and
  orange mcp suggest [--quest <quest-id>] [--query <text>].
• The built-in MCP catalog remains local and deterministic, starting with
  context7, github, sentry, and linear.
• mcp suggest returns proposal cards with score-backed recommendation
  metadata: score, confidence, matched_signals, why_now, and
  requires_user_approval: true.
• Proposal cards remain recommendation cards, not install or execution results;
  they include not_executed: true and config_mutation: false.
• Weak or unrelated inputs return the explicit no-suggestion state with
  suggestions: [], proposal_cards: [], no_suggestion_reason, and
  suggested_next_step.
• Ranking is deterministic by descending score, built-in catalog order, and
  MCP id ...

orange-hyper v0.7.0-alpha.0

Release Notes

v0.7.0-alpha.0

Adapter Invocation Contract alpha.

• Package version is 0.7.0-alpha.0.
• README version is 0.7-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds the Adapter Layer command surface:
  orange adapter list, orange adapter show <recipe-id>, and
  orange adapter dry-run <recipe-id>.
• New JSON command ids are adapter.list, adapter.show, and
  adapter.dryRun.
• Built-in recipes are quest-capture, work-complete-to-memory,
  project-status, hook-check, and mcp-advice.
• Each recipe declares id, title, purpose, when_to_use, commands,
  required_inputs, outputs, safety_rules, forbidden_actions,
  expected_contract_version, and safety_flags.
• Each recipe command step declares command, why, required_input,
  expected_json_command_id, mutates_project_state, and
  requires_user_approval.
• Safety flags are fixed as direct_file_mutation: false,
  parses_human_output: false, requires_json_mode: true,
  auto_accept: false, auto_install: false, and auto_unlock: false.
• adapter dry-run does not execute recipe commands and does not modify
  .orange-hyper.
• docs/20_ADAPTER_LAYER.md documents that natural-language and skill layers
  must call the kernel through Orange CLI --json commands, must not parse
  human output, and must not duplicate kernel state logic.

Explicitly not included

• Codex/Claude-specific adapter automatic installation
• actual adapter runtime
• automatic Quest creation
• automatic memory proposal creation
• automatic accept or reject
• automatic graph rebuild
• automatic hook execution
• MCP automatic installation or execution
• subagent orchestration
• direct .orange-hyper file mutation
• auto planner or auto execution loop

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signal Preview surface to stable without adding automatic role, tool, hook,
  subagent, planner, workflow, config, graph, or project-memory mutation.
• Stable command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• growth status reports conservative project growth state from Quest, Route,
  accepted Memory Graph, Hook warning, Doctor, and MCP-advisor-shaped
  documentation/API freshness signals.
• growth suggest creates candidates only from repeated source-backed
  evidence. Each candidate includes score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• growth explain reports the same deterministic rule path with source-backed
  evidence instead of LLM calls, network calls, MCP calls, or planner loops.
• No-candidate negative cases are expected when evidence is too thin; a single
  Quest or one-off generic signal should not create a candidate.
• Positive repo smoke showed repeated documentation, verification, and memory
  hygiene evidence producing verification-discipline, documentation-focus,
  and memory-hygiene candidates while keeping growthLevel conservatively at
  sprout.
• identity build includes a Growth Signal Preview summary with growth level
  reason, candidate count, top candidates, growth confidence summary, and
  No automatic unlocks.
• At the v0.6.0 cut, v0.7 Adapter Layer was the next roadmap step after this
  stable Growth Signal Preview boundary.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.1

Growth Signal Preview quality hardening.

• Package version is 0.6.0-alpha.1.
• README version is 0.6-doc.1.
• Adapter JSON contract_version remains "0.1".
• growthLevel calibration is more conservative and no longer advances from
  accepted node count alone.
• Level calculation now considers node type diversity, verified Quest ratio,
  repeated evidence count, pending proposal load, doctor ok state, and active
  project boundary.
• Growth candidates now include deterministic score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• Candidate thresholds now require repeated source-backed evidence; a single
  generic string match is not enough to create a candidate.
• Backend/API false positives from generic API/Route Contract text are pinned
  by regression tests.
• growth explain now reports source-backed evidence for each candidate,
  including Quest, node, route layer, Hook warning, and MCP-shaped signal
  source fields where available.
• identity build now shows growth level reason, candidate count, top
  candidates, growth confidence summary, and No automatic unlocks.

Explicitly still not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.0

Growth Signal Preview alpha.

• Package version is 0.6.0-alpha.0.
• README version is 0.6-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds a read-only Growth Signal Preview command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• New JSON command ids are growth.status, growth.suggest, and
  growth.explain.
• growth status summarizes project identity, accepted memory nodes, node
  type distribution, route/layer distribution, verified/unverified Quest
  ratio, pending memory proposals, Hook warning summary, MCP advisor signal
  summary, and decorative growthLevel.
• growth suggest returns deterministic candidates such as verification
  discipline, memory hygiene, backend/API focus, documentation focus, MCP
  documentation advisor readiness, and hook hygiene.
• Every growth candidate keeps auto_unlock: false and
  requires_user_approval: true.
• growth explain reports the deterministic evidence/rule path for each
  candidate without LLM calls, network calls, MCP calls, or planner loops.
• identity build now includes a compact Growth Signal Preview summary in the
  generated HTML and summary JSON.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.5.0

MCP Advisor stable release.

• Package version is 0.5.0.
• README version is 0.5-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.5.0 promotes the validated MCP Advisor alpha surface to stable without
  adding MCP integration or automatic execution behavior.
• Stable command surface:
  orange mcp list, orange mcp show <mcp-id>, and
  orange mcp suggest [--quest <quest-id>] [--query <text>].
• The built-in MCP catalog remains local and deterministic, starting with
  context7, github, sentry, and linear.
• mcp suggest returns proposal cards with score-backed recommendation
  metadata: score, confidence, matched_signals, why_now, and
  requires_user_approval: true.
• Proposal cards remain recommendation cards, not install or execution results;
  they include not_executed: true and config_mutation: false.
• Weak or unrelated inputs return the explicit no-suggestion state with
  suggestions: [], proposal_cards: [], no_suggestion_reason, and
  suggested_next_step.
• Ranking is deterministic by descending score, built-in catalog order, and
  MCP id tie-breaker.
• Korean and English catalog signal matching is stable for docs/version/API,
  repository issue/PR, runtime incident/error, and product ticket/task contexts.
• Quest-based suggestions remain read-only and can use active or completed
  Quest title, request, constraints, and notes without mutating the Quest or
  project memory.

Explicitly not included

• MCP automatic installation
• MCP automatic execution
• MCP config automatic modification
• API key storage
• external network calls
• hook-triggered MCP suggestion, installation, or execution
• subagent execution
• role evolution
• auto planner or auto execution loop
• automatic Quest, Proposal, Graph, config, or project memory mutation

v0.5.0-alpha.1

MCP Advisor recommendation-quality hardening.

• Package version is 0.5.0-alpha.1.
• README version is 0.5-doc.1.
• Adapter JSON contract_version remains "0.1".
• mcp suggest suggestions now include deterministic score,
  confidence, matched_signals, why_now, and
  requires_user_approval: true.
• Weak or unrelated requests now return an explicit no-suggestion state with
  suggestions: [], no_suggestion_reason, and suggested_next_step.
• Multiple matching MCPs are ranked by score, with deterministic catalog-order
  tie-breaking.
• Korean and English catalog signals were expanded for docs/version/API,
  repository issue/PR, runtime incident/error, and product ticket/task contexts.
• Quest-based suggestions include source_quest_id, support active and
  completed Quests, and read Quest request/title/constraints/notes without
  mutating Quest or project memory.
• Proposal cards now make the execution boundary explicit with
  not_executed: true and config_mutation: false.

Explicitly still not included

• MCP automatic installation
• MCP automatic execution
• MCP config automatic modification
• API key stora...

orange-hyper v0.6.0

Release Notes

v0.6.0

Growth Signal Preview stable release.

• Package version is 0.6.0.
• README version is 0.6-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.6.0 promotes the validated v0.6.0-alpha.0 and v0.6.0-alpha.1 Growth
  Signal Preview surface to stable without adding automatic role, tool, hook,
  subagent, planner, workflow, config, graph, or project-memory mutation.
• Stable command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• growth status reports conservative project growth state from Quest, Route,
  accepted Memory Graph, Hook warning, Doctor, and MCP-advisor-shaped
  documentation/API freshness signals.
• growth suggest creates candidates only from repeated source-backed
  evidence. Each candidate includes score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• growth explain reports the same deterministic rule path with source-backed
  evidence instead of LLM calls, network calls, MCP calls, or planner loops.
• No-candidate negative cases are expected when evidence is too thin; a single
  Quest or one-off generic signal should not create a candidate.
• Positive repo smoke showed repeated documentation, verification, and memory
  hygiene evidence producing verification-discipline, documentation-focus,
  and memory-hygiene candidates while keeping growthLevel conservatively at
  sprout.
• identity build includes a Growth Signal Preview summary with growth level
  reason, candidate count, top candidates, growth confidence summary, and
  No automatic unlocks.
• v0.7 Adapter Layer is the next roadmap step after this stable Growth Signal
  Preview boundary.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.1

Growth Signal Preview quality hardening.

• Package version is 0.6.0-alpha.1.
• README version is 0.6-doc.1.
• Adapter JSON contract_version remains "0.1".
• growthLevel calibration is more conservative and no longer advances from
  accepted node count alone.
• Level calculation now considers node type diversity, verified Quest ratio,
  repeated evidence count, pending proposal load, doctor ok state, and active
  project boundary.
• Growth candidates now include deterministic score, evidence_count,
  matched_signals, confidence, reason, suggested_next_step,
  auto_unlock: false, and requires_user_approval: true.
• Candidate thresholds now require repeated source-backed evidence; a single
  generic string match is not enough to create a candidate.
• Backend/API false positives from generic API/Route Contract text are pinned
  by regression tests.
• growth explain now reports source-backed evidence for each candidate,
  including Quest, node, route layer, Hook warning, and MCP-shaped signal
  source fields where available.
• identity build now shows growth level reason, candidate count, top
  candidates, growth confidence summary, and No automatic unlocks.

Explicitly still not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.6.0-alpha.0

Growth Signal Preview alpha.

• Package version is 0.6.0-alpha.0.
• README version is 0.6-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds a read-only Growth Signal Preview command surface:
  orange growth status, orange growth suggest, and
  orange growth explain.
• New JSON command ids are growth.status, growth.suggest, and
  growth.explain.
• growth status summarizes project identity, accepted memory nodes, node
  type distribution, route/layer distribution, verified/unverified Quest
  ratio, pending memory proposals, Hook warning summary, MCP advisor signal
  summary, and decorative growthLevel.
• growth suggest returns deterministic candidates such as verification
  discipline, memory hygiene, backend/API focus, documentation focus, MCP
  documentation advisor readiness, and hook hygiene.
• Every growth candidate keeps auto_unlock: false and
  requires_user_approval: true.
• growth explain reports the deterministic evidence/rule path for each
  candidate without LLM calls, network calls, MCP calls, or planner loops.
• identity build now includes a compact Growth Signal Preview summary in the
  generated HTML and summary JSON.

Explicitly not included

• role automatic creation
• MCP automatic installation or execution
• hook policy automatic changes
• subagent execution or recommendation automation
• auto planner or auto execution loop
• project memory automatic mutation
• graph node automatic creation
• workflow enforcement

v0.5.0

MCP Advisor stable release.

• Package version is 0.5.0.
• README version is 0.5-doc.2.
• Adapter JSON contract_version remains "0.1".
• v0.5.0 promotes the validated MCP Advisor alpha surface to stable without
  adding MCP integration or automatic execution behavior.
• Stable command surface:
  orange mcp list, orange mcp show <mcp-id>, and
  orange mcp suggest [--quest <quest-id>] [--query <text>].
• The built-in MCP catalog remains local and deterministic, starting with
  context7, github, sentry, and linear.
• mcp suggest returns proposal cards with score-backed recommendation
  metadata: score, confidence, matched_signals, why_now, and
  requires_user_approval: true.
• Proposal cards remain recommendation cards, not install or execution results;
  they include not_executed: true and config_mutation: false.
• Weak or unrelated inputs return the explicit no-suggestion state with
  suggestions: [], proposal_cards: [], no_suggestion_reason, and
  suggested_next_step.
• Ranking is deterministic by descending score, built-in catalog order, and
  MCP id tie-breaker.
• Korean and English catalog signal matching is stable for docs/version/API,
  repository issue/PR, runtime incident/error, and product ticket/task contexts.
• Quest-based suggestions remain read-only and can use active or completed
  Quest title, request, constraints, and notes without mutating the Quest or
  project memory.

Explicitly not included

• MCP automatic installation
• MCP automatic execution
• MCP config automatic modification
• API key storage
• external network calls
• hook-triggered MCP suggestion, installation, or execution
• subagent execution
• role evolution
• auto planner or auto execution loop
• automatic Quest, Proposal, Graph, config, or project memory mutation

v0.5.0-alpha.1

MCP Advisor recommendation-quality hardening.

• Package version is 0.5.0-alpha.1.
• README version is 0.5-doc.1.
• Adapter JSON contract_version remains "0.1".
• mcp suggest suggestions now include deterministic score,
  confidence, matched_signals, why_now, and
  requires_user_approval: true.
• Weak or unrelated requests now return an explicit no-suggestion state with
  suggestions: [], no_suggestion_reason, and suggested_next_step.
• Multiple matching MCPs are ranked by score, with deterministic catalog-order
  tie-breaking.
• Korean and English catalog signals were expanded for docs/version/API,
  repository issue/PR, runtime incident/error, and product ticket/task contexts.
• Quest-based suggestions include source_quest_id, support active and
  completed Quests, and read Quest request/title/constraints/notes without
  mutating Quest or project memory.
• Proposal cards now make the execution boundary explicit with
  not_executed: true and config_mutation: false.

Explicitly still not included

• MCP automatic installation
• MCP automatic execution
• MCP config automatic modification
• API key storage
• external network calls
• hook-triggered MCP suggestion, installation, or execution
• subagent execution
• role evolution
• auto planner or auto execution loop
• automatic Quest, Proposal, Graph, config, or project memory mutation

v0.5.0-alpha.0

MCP Advisor alpha.

• Package version is 0.5.0-alpha.0.
• README version is 0.5-doc.0.
• Adapter JSON contract_version remains "0.1".
• This alpha adds a read-only MCP Advisor command surface:
  orange mcp list, orange mcp show <mcp-id>, and
  orange mcp suggest [--quest <quest-id>] [--query <text>].
• New JSON command ids are mcp.list, mcp.show, and mcp.suggest.
• The built-in catalog starts with context7, github, sentry, and
  linear, each with use cases, useful-when guidance, risks, token impact,
  install hints, and persistent-use policy.
• mcp suggest returns proposal cards with tool, why_now,
  expected_benefit, scope, risk, token_impact, install_command,
  use_once_or_persist, and requires_user_approval: true.
• MCP suggestions read current project state only. They do not install MCPs,
  run MCPs, write MCP config, create Quest/Proposal/Graph state, write hook
  reports, or save project memory.
• Hooks remain separate from MCP Advisor. Hook preview/status/run output does
  not automatically include MCP advisor summaries.

Explicitly not included

• MCP automatic installation
• MCP automatic execution
• MCP config automatic modification
• API key storage
• external network calls
• hook-triggered MCP suggestion, installation, or execution
• subagent execution
• role evolution
• auto planner or auto execution loop
• automatic Quest, Proposal, Graph, config, or project memory mutation

v0.4.0

Minimal Hook Preview stable release.

• Package version is 0.4.0.
• README ve...