Skip to content

orange-hyper v0.8.0-alpha.0

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Nam-Cheol Nam-Cheol released this 17 Jun 14:53
· 7 commits to main since this release
v0.8.0-alpha.0
9da31f1

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 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 version is 0.4-doc.2.
  • Adapter JSON contract_version remains "0.1".
  • v0.4.0 promotes the validated v0.4.0-alpha.0 and v0.4.0-alpha.1
    Minimal Hook Preview surface to stable without adding new CLI behavior.
  • orange hook preview, orange hook status,
    orange hook run session-start, and orange hook run stop are the stable
    read-only / warning-first hook preview commands.
  • session-start and stop perform read-only checks only. They observe
    project identity, Project Boundary, doctor quick status, graph/index state,
    pending proposals, and capsule/identity freshness without repairing or
    generating project memory.
  • Hook warnings use stable adapter-facing { code, message, hint } objects
    with HOOK_* warning codes such as HOOK_PROJECT_ID_MISSING,
    HOOK_IDENTITY_SUMMARY_MISSING, HOOK_CAPSULE_STALE,
    HOOK_PENDING_PROPOSALS, HOOK_DOCTOR_NOT_OK, and
    HOOK_GRAPH_PROVENANCE_WARNING.
  • Hook report files use the stable report schema with generator metadata,
    schema_version, report_kind, generated_at, project_id,
    project_name, event, readOnly: true, autoMutation: false,
    warnings, summaries for doctor/graph/identity/capsule, and
    recommended_commands.
  • The no-mutation invariant is stable: default hook preview/status/run
    commands do not modify Quest, Proposal, Graph, Capsule, Identity, Project
    Boundary, or config state.
  • --write-report is the only hook option that writes a file. It writes a
    local-only diagnostic report under .orange-hyper/hooks/reports/, which is
    ignored by default and is not project memory.
  • v0.5 MCP Advisor is the next roadmap step after this stable hook preview
    boundary.

Explicitly not included

  • MCP implementation or installation
  • subagent execution
  • role evolution
  • auto planner or auto execution loop
  • automatic Quest creation
  • automatic memory proposal creation
  • automatic proposal accept/reject
  • automatic graph rebuild
  • automatic doctor repair
  • automatic Quest, Proposal, or Graph mutation

v0.4.0-alpha.1

Minimal Hook Preview hardening alpha.

  • Package version is 0.4.0-alpha.1.
  • README version is 0.4-doc.1.
  • Adapter JSON contract_version remains "0.1".
  • Hook warnings now use stable adapter-facing { code, message, hint }
    objects with HOOK_* codes such as HOOK_PROJECT_ID_MISSING,
    HOOK_IDENTITY_SUMMARY_MISSING, HOOK_CAPSULE_STALE,
    HOOK_PENDING_PROPOSALS, HOOK_DOCTOR_NOT_OK, and
    HOOK_GRAPH_PROVENANCE_WARNING.
  • --write-report now writes a stable hook report schema with
    schema_version, generated_at, project_id, project_name, event,
    readOnly: true, autoMutation: false, warnings, summaries for
    doctor/graph/identity/capsule, and recommended_commands.
  • Hook reports remain off by default and write only under
    .orange-hyper/hooks/reports/; reports are local diagnostics, not project
    memory.
  • Hook preview/status/run no-mutation coverage now verifies that default hook
    execution does not modify Quest, Proposal, Graph, Capsule, Identity, or config
    state.
  • Capsule and identity stale warnings now use a documented mtime comparison
    against config, Quest, proposal, graph node, and graph index source files,
    with no graph rebuild, capsule generation, or identity build.
  • Tests cover warning code shape, report schema, report-only writes, stale
    capsule/identity warnings, doctor-not-ok warnings, pending proposal warnings,
    and the existing v0.4.0-alpha.0 hook command surface.

Explicitly not included

  • automatic Quest creation
  • automatic memory proposal creation
  • automatic proposal accept/reject
  • automatic graph rebuild
  • automatic doctor repair
  • hook installation
  • Codex-specific or Claude-specific adapter implementation
  • MCP implementation or installation
  • subagent execution
  • role evolution
  • auto planner or auto execution loop

v0.4.0-alpha.0

Minimal Hook Preview alpha.

  • Package version is 0.4.0-alpha.0.
  • README version is 0.4-doc.0.
  • Adapter JSON contract_version remains "0.1".
  • This alpha adds a read-only, warning-first hook command surface only:
    orange hook preview, orange hook status,
    orange hook run session-start, and orange hook run stop.
  • New JSON command ids are hook.preview, hook.status,
    hook.runSessionStart, and hook.runStop.
  • hook preview shows project_id, doctor quick check, capsule freshness,
    identity summary, graph/index, local report, readOnly: true, and
    autoMutation: false targets without installing hooks.
  • hook status reports previewAvailable: true, installed: false,
    readOnly: true, autoMutation: false, supported events
    session-start and stop, and future unsupported events such as
    user-prompt-submit and post-tool-use.
  • hook run session-start observes .orange-hyper, config.project_id,
    Project Boundary, identity summary presence, accepted memory node count, and
    doctor quick status.
  • hook run stop observes doctor quick status, completed Quest verification
    anomalies, accepted graph node provenance anomalies, pending memory proposal
    count, stale/missing capsule or identity warnings, and project boundary
    warnings.
  • Hook run warnings are returned as warnings and hints; missing project_id or
    doctor diagnostics do not trigger automatic repair.
  • Hook reports are off by default. --write-report writes only under
    .orange-hyper/hooks/reports/, which is ignored by default, and the option
    does not accept a path or filename.

Explicitly not included

  • automatic Quest creation
  • automatic memory proposal creation
  • automatic proposal accept/reject
  • automatic graph rebuild
  • automatic doctor repair
  • hook installation
  • MCP implementation or installation
  • subagent execution
  • role evolution
  • auto planner or auto execution loop
  • branch/PR/SPEC workflow enforcement

v0.3.0

Memory Graph Usability stable release.

  • Package version is 0.3.0.
  • README version remains 0.3-doc.3.
  • Adapter JSON contract_version remains "0.1".
  • v0.3.0 promotes the validated v0.3.0-alpha.0 through v0.3.0-alpha.3
    surface to stable without adding new CLI behavior.
  • This is a Memory Graph Usability stable release, not a Hook, MCP, Subagent,
    role evolution, auto planner, auto execution loop, graph editing, or
    semantic/vector search release.

Stable surface

  • orange graph list, orange graph show, orange graph search, and
    orange graph rebuild-index provide read-only exploration of current-project
    accepted memory nodes.
  • Graph filters cover node type, source Quest, and source Proposal where
    supported; search remains deterministic plain-text search, not fuzzy,
    semantic, or vector search.
  • orange identity build includes a read-only Identity Graph Preview with
    accepted memory node count, node type distribution, source columns,
    project-boundary state, and node details.
  • Project Boundary Guard keeps graph results scoped to the current
    project_id and keeps cross-project memory out of current-project graph
    views.
  • Shared Memory State Policy keeps public .orange-hyper/ state bounded to
    shareable config, completed Quests, accepted Memory Delta Proposals, accepted
    graph nodes/read models, and graph provenance.
  • README identity rewrite and synchronized Korean, English, Simplified Chinese,
    and Japanese READMEs are the stable public project narrative.
  • Type Safety Foundation remains a development-only contract-checking layer for
    Adapter JSON envelopes and Quest/Proposal/Graph/Doctor/Identity shapes while
    the package is still distributed as JavaScript.
  • Trusted Publishing is the official npm release path. Tags matching
    v*-alpha.* publish with npm dist-tag alpha; stable tags matching vX.Y.Z,
    including v0.3.0, publish with the default latest dist-tag.

v0.3.0-alpha.3

Publish Pipeline Trusted Publishing release.

  • Package version is 0.3.0-alpha.3.
  • README version remains 0.3-doc.3.
  • Adapter JSON contract_version remains "0.1".
  • Publish pipeline moved to GitHub Actions trusted publishing.
  • Official npm publish now runs from the push tag v* workflow with
    id-token: write, npm registry setup, full release checks, package dry-run,
    and npm publish without NODE_AUTH_TOKEN as the default path.
  • Alpha tags matching v*-alpha.* publish with npm dist-tag alpha; stable
    tags matching vX.Y.Z publish with npm dist-tag latest.
  • Local npm publish is not the official release path because local terminals
    may not have an npm provenance OIDC provider. Emergency local publish can use
    NPM_CONFIG_PROVENANCE=false, but that bypass is not the official release
    path.
  • Public Memory State Audit now keeps .orange-hyper/ dogfooding evidence
    bounded to shared project memory: config, completed Quests, accepted Memory
    Delta Proposals, and graph provenance.
  • orange doctor now warns if local/generated/private .orange-hyper state is
    tracked by Git, warns on private-looking local paths in public memory, and
    errors on token/secret/env/auth-like strings before commit.
  • Existing dogfooding memory evidence no longer stores local absolute temporary
    paths; accepted proposal hashes were refreshed to preserve graph provenance.
  • No MCP, hooks, subagents, role evolution, auto planner, or auto execution loop
    behavior is introduced.

v0.3.0-alpha.1

Graph Quality Hardening and real-repo dogfooding release.

  • Package version is 0.3.0-alpha.1.
  • README version is 0.3-doc.3.
  • Adapter JSON contract_version remains "0.1".
  • No MCP, hooks, subagents, role evolution, auto planner, or auto execution loop
    behavior is introduced.

Graph Quality Hardening

  • orange init now idempotently backfills missing v0.2/v0.3 storage directories
    and graph read-model files without overwriting existing config/project data.
  • .orange-hyper/.gitignore now keeps accepted Memory Delta Proposals shareable
    with accepted graph nodes while pending/rejected proposals and generated local
    state remain ignored by default.
  • orange graph list now supports --type, --source-quest, and
    --source-proposal filters in human and JSON modes.
  • orange graph search now supports --type and --source-quest filters and
    returns deterministic plain-text scores. It remains non-fuzzy, non-semantic,
    and non-vector search.
  • graph show human output now highlights node id, type, title, Candidate
    Memory, source Quest/Proposal, accepted time, and provenance.
  • graph rebuild-index separates deterministic updated_at from generation
    metadata and preserves semantic read-model output across repeated rebuilds.

Doctor and Identity

  • doctor --json now includes structured graph/project-boundary diagnostics
    with code, message, and hint while preserving existing string arrays.
  • Diagnostic codes include CONFIG_PROJECT_ID_MISSING,
    GRAPH_INDEX_ORPHAN_ENTRY, ACCEPTED_PROPOSAL_MISSING_NODE,
    ACCEPTED_NODE_SOURCE_PROPOSAL_MISSING, GRAPH_NODE_PROJECT_MISMATCH, and
    LEGACY_PROJECT_ID_MISSING.
  • Identity preview now shows accepted memory node count, node type distribution,
    project-boundary active state, source columns, and anchor-based node details.
  • Identity remains read-only and does not include graph editing controls or
    heavy graph dependencies.

Documentation and Checks

  • Added npm run check:readme-sync to verify the README version metadata across
    Korean, English, Simplified Chinese, and Japanese READMEs.
  • Updated README, Memory Graph spec, Identity Dashboard spec, and Adapter
    Contract examples for graph filters, deterministic ranking, idempotent index
    principles, and alpha.1 identity preview hardening.

Type Safety Foundation

  • Added TypeScript as a development-only contract checker while keeping Orange
    Hyper distributed as JavaScript.
  • Added tsconfig.json with allowJs, checkJs, and noEmit for gradual JS
    typechecking without a dist build pipeline.
  • Added domain contract types for Adapter JSON envelopes, command ids, project
    identity, Quest frontmatter, route contracts, Memory Delta Proposals,
    accepted graph nodes, graph index entries, doctor diagnostics, and identity
    summaries.
  • Added focused JSDoc return-shape links for JSON envelopes and core
    Quest/Memory/Graph/Doctor/Identity result surfaces.
  • Adapter JSON Contract remains contract_version "0.1".

v0.3.0-alpha.0

Memory Graph Usability, Identity Graph Preview, and README Identity Rewrite
alpha.

  • Package version is 0.3.0-alpha.0.
  • README version is 0.3-doc.2 for this documentation-only narrative restore
    patch.
  • Adapter JSON contract_version remains "0.1".

Memory Graph Usability

  • New command: orange graph list [--json].
  • New command: orange graph show <node-id> [--json].
  • New command: orange graph search <query> [--json].
  • New command: orange graph rebuild-index [--json].
  • New JSON command ids are graph.list, graph.show, graph.search, and
    graph.rebuildIndex.
  • Graph commands return only accepted memory nodes whose project_id matches
    the current .orange-hyper/config.json.
  • Pending and rejected Memory Delta Proposals are not graph nodes.
  • graph/index.json is a read model regenerated from graph node Markdown.
  • graph search is simple text search over node id, title, Candidate
    Memory/summary, node type, source quest/proposal, tags, and keywords.
  • doctor now checks graph index parseability, source/index mismatch, orphan
    index entries, accepted proposals missing graph nodes, graph provenance
    consistency, project boundary mismatches, and graph selector/path safety.

Identity Graph Preview

  • identity build remains a single self-contained HTML file and now includes a
    read-only graph preview with accepted memory node count, node type
    distribution, source quest/proposal table, simple SVG node-link preview, and
    selected node detail.
  • Identity output explicitly displays Graph preview is read-only. and
    Graph editing is not supported.

Documentation

  • README Narrative Restore / Visual-Narrative Reconciliation is a documentation
    patch, not a core feature release.
  • Restored the canonical README narrative after the 3-card summary: problem
    definition, harness reflection, chosen direction, and Orange Hyper identity.
  • Kept the existing visual structure assets in place: hero image,
    assets/readme/core-flow.png, and assets/readme/memory-lifecycle.png.
  • Kept package version and Adapter JSON contract_version unchanged.
  • No CLI, core runtime, Memory Graph behavior, Identity runtime, MCP, hooks,
    subagents, or role evolution changes are included.

README Identity Rewrite

  • Internal documentation work name: v0.3-doc-prep.
  • Korean is the base README language.
  • Added synchronized English, Simplified Chinese, and Japanese README files.
  • Clarified that README version, package version, and Adapter JSON contract
    version are separate version axes.
  • Reframed the README around problem definition, harness reflection, project
    direction, Orange Hyper identity, current features, usage, roadmap,
    non-goals, and docs links.
  • Kept current recommended npx --package orange-hyper@latest orange ...
    usage, alpha channel usage, source checkout usage, and npm link usage.

README Visual Structure

  • This is a README visual structure update, not a core feature release.
  • README version for the visual structure pass was 0.3-doc.1; the narrative
    restore patch reconciles that visual rhythm with the longer README narrative
    at 0.3-doc.2.
  • No CLI, Memory Graph, Identity runtime, Adapter JSON Contract, or package
    runtime behavior changes are included.
  • Moved the user-provided core-flow and memory-lifecycle README images under
    assets/readme/.
  • Synchronized README.md, README.en.md, README.zh-CN.md, and
    README.ja.md around the same visual rhythm: hero, project identity, language
    links, badges and version metadata, 3-card problem/reflection/direction,
    Core Flow image, Memory Lifecycle image, feature summary, Quickstart, Identity
    Dashboard / Graph Preview, roadmap, non-goals, and docs links.
  • Added assets to the package file list so the README images are included in
    npm package contents.

Explicitly not included

  • automatic memory write
  • graph state editing
  • Obsidian-grade editor
  • fuzzy/semantic/vector search
  • D3/Cytoscape/heavy graph dependency
  • external source import
  • MCP/hooks/subagents/role evolution
  • auto planner or auto execution loop

Verification checklist

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

README package inclusion check: confirm the npm pack --dry-run file list
includes README.md, README.en.md, README.zh-CN.md, README.ja.md, and
the assets/readme/ images.

Fresh temp smoke

node bin/orange.js init
node bin/orange.js quest new "remember graph usability" --layer L2 --json
node bin/orange.js quest done <quest-id> --evidence "manual smoke passed" --json
node bin/orange.js remember propose --quest <quest-id> --json
node bin/orange.js remember accept <proposal-id> --json
node bin/orange.js graph rebuild-index --json
node bin/orange.js graph list --json
node bin/orange.js graph search "graph usability" --json
node bin/orange.js doctor --json
node bin/orange.js identity build --json

v0.2.1

This patch release includes two bounded release-surface fixes before v0.3 Memory
Graph usability work:

  • Project Boundary Guard
  • project_id/project_name propagation
  • capsule Project Boundary header
  • doctor project boundary checks
  • doctor --repair-project-id
  • external source import explicitly blocked
  • CLI/npx execution surface fix
  • orange-hyper bin alias
  • README explicit npx --package usage

Project Boundary Guard release.

This patch release keeps v0.3 Memory Graph usability work from indexing memory
that belongs to another project or came from unrelated external context. The
latest published npm version is 0.2.0, so this release bumps the package
version to 0.2.1.

  • Package version is 0.2.1.
  • Adapter JSON contract_version remains "0.1".
  • orange init now writes stable random project_id and project_name fields
    to .orange-hyper/config.json without storing absolute local paths.
  • New Quest, Memory Delta Proposal, Accepted Memory Node, Capsule boundary, and
    Identity summary JSON output record the current project identity.
  • orange capsule adds a Project Boundary header explaining that only
    Quest/Proposal/Accepted Node artifacts with the current project_id are
    project memory.
  • doctor warns on legacy artifacts missing project_id and errors on explicit
    cross-project mismatches.
  • doctor verifies proposal-to-source-Quest and accepted-node-to-source-proposal
    project identity consistency.
  • New option: orange doctor --repair-project-id. It fills missing legacy
    project identity fields with the current config values, but does not overwrite
    a different existing project_id.
  • doctor --json keeps the existing JSON envelope and includes boundary
    diagnostics under data.project_boundary.
  • External source memory import remains a future feature. v0.2.1 does not add
    remember propose --from-file, external report import, clipboard/pasted
    report automatic proposal, Memory Graph rendering, MCP, hooks, subagents, role
    evolution, auto planner, or auto execution loop.

v0.2.0

Memory Delta Proposal stable release.

This stable release does not add new core features after v0.2.0-alpha.2.
It promotes the v0.2 proposal review lifecycle to the stable channel: completed
Quests can produce pending Memory Delta Proposals, users can inspect, validate,
revise, accept, or reject them, and doctor/identity build can summarize the
resulting proposal state. This is not a Memory Graph rendering release.

  • Package version is 0.2.0.
  • Adapter JSON contract_version remains "0.1".
  • Stable usage is npx orange-hyper ... after publish.
  • Alpha usage remains npx orange-hyper@alpha ... for the alpha channel.
  • Expected post-publish dist-tags: latest: 0.2.0, alpha: 0.2.0-alpha.2.
  • Stabilized from v0.2.0-alpha.0: remember propose, remember list,
    remember show, remember accept, and remember reject.
  • Stabilized from v0.2.0-alpha.1: proposal quality validation, pending
    duplicate prevention, list filters, accepted node provenance, doctor
    provenance checks, and identity proposal/node counts.
  • Stabilized from v0.2.0-alpha.2: remember validate, remember revise,
    accepted/rejected revise protection, pending duplicate revise protection,
    proposal timestamp warning, and identity review-stage messaging.
  • Accepted proposals create candidate graph node files with proposal/source
    Quest provenance. They do not activate graph rendering or automatic retrieval.
  • There is no automatic memory write. Only a user-approved remember accept
    can create an accepted graph node candidate.

Explicitly not included:

  • Memory Graph rendering
  • Obsidian-style dashboard graph
  • MCP/hooks/subagents/role evolution
  • auto planner or auto execution loop
  • raw prompt archive
  • automatic memory write

Stable Publish Method

0.2.0 is stable, so publish it to npm without an explicit prerelease tag. The
npm publish command uses the tag config to decide which dist-tag to apply,
and its default is latest; alpha releases should continue to use
npm publish --tag alpha.

Do not run the publishing commands until the release is explicitly approved.

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

git add -A
git commit -m "chore: release orange-hyper v0.2.0 memory proposal stable"

git tag -a v0.2.0 -m "orange-hyper v0.2.0"
git push origin main
git push origin v0.2.0

gh release create v0.2.0 \
  --title "orange-hyper v0.2.0" \
  --notes-file RELEASE_NOTES.md

npm publish

After publish:

npm dist-tag ls orange-hyper
npx orange-hyper --help

Expected dist-tags:

latest: 0.2.0
alpha: 0.2.0-alpha.2

Alpha.2 Smoke Evidence

The v0.2.0-alpha.2 npm package was used to harden the stable proposal review
contract before 0.2.0:

  • node bin/orange.js --help exposed the full v0.2 remember command surface.
  • npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache checked the
    package contents before stable promotion.
  • Fresh temp smoke covered init, quest new --json, quest done --json,
    remember propose --json, remember validate --json,
    remember revise --json, remember accept --json, doctor --json, and
    identity build --json.
  • The smoke JSON outputs kept contract_version: "0.1" and command ids
    quest.new, quest.done, remember.propose, remember.validate,
    remember.revise, remember.accept, doctor.run, and identity.build.

v0.2.0-alpha.2

Memory Proposal Review UX release.

This release lets users and adapters safely inspect, validate, revise, and
re-validate pending Memory Delta Proposals before accepting them. It does not
enable Memory Graph rendering, MCP, hooks, subagents, role evolution, auto
planning, automatic execution loops, or automatic memory writes.

  • Package version is 0.2.0-alpha.2.
  • Adapter JSON contract_version remains "0.1".
  • New command: orange remember validate <proposal-id> [--json].
  • New command: orange remember revise <proposal-id> --candidate "..." [--json].
  • New command: orange remember revise <proposal-id> --why "..." [--json].
  • New command: orange remember revise <proposal-id> --confidence low|medium|high [--json].
  • JSON command ids are remember.validate and remember.revise.
  • remember validate works for pending, accepted, and rejected proposals.
  • remember revise is limited to pending proposals. Accepted/rejected proposals
    remain protected user decision records.
  • Successful revise writes the proposal through the kernel, updates
    updated_at, and reruns quality validation.
  • Revising Candidate Memory to duplicate another pending proposal fails with a
    clear JSON/human error.
  • doctor now warns when proposal updated_at is earlier than created_at.
  • identity build remains a placeholder. It now reports the v0.2 proposal
    review stage and includes pendingMemoryProposalsWithWarnings in JSON
    summary.

Explicitly not included:

  • Memory Graph rendering
  • Obsidian-style dashboard graph
  • MCP/hooks/subagents/role evolution
  • auto planner or auto execution loop
  • raw prompt archive
  • automatic memory write

Verification Checklist

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

Fresh temp smoke:

node bin/orange.js init
node bin/orange.js quest new "remember a durable decision" --layer L2 --json
node bin/orange.js quest done <quest-id> --evidence "manual smoke passed" --json
node bin/orange.js remember propose --quest <quest-id> --json
node bin/orange.js remember validate <proposal-id> --json
node bin/orange.js remember revise <proposal-id> --candidate "Durable project memory candidate." --json
node bin/orange.js remember validate <proposal-id> --json
node bin/orange.js remember accept <proposal-id> --json
node bin/orange.js doctor --json
node bin/orange.js identity build --json

v0.2.0-alpha.1

Memory Proposal Quality Hardening release.

This release hardens the v0.2 Memory Delta Proposal workflow without enabling
Memory Graph rendering, MCP, hooks, subagents, role evolution, auto planning, or
automatic execution loops.

  • Package version is 0.2.0-alpha.1.
  • Adapter JSON contract_version remains "0.1".
  • remember propose is idempotent for matching pending proposals with the same
    source_quest, node_type, and Candidate Memory.
  • Duplicate remember propose --json calls return the existing proposal and set
    data.duplicated: true.
  • remember list now supports --status pending|accepted|rejected,
    --type decision|constraint|component|risk|verification, and
    --quest <quest-id> in human and JSON modes.
  • Proposal quality validation now checks required content sections, source
    quest or verification evidence references, suggested node type consistency,
    and low|medium|high confidence.
  • Very short or generic Candidate Memory text is reported as a warning rather
    than a hard error.
  • Accepted graph node candidates now include source_proposal, source_quest,
    accepted_at, node_type, origin: memory-delta-proposal, and
    source_proposal_hash provenance.
  • doctor now reports proposal quality warnings/errors and catches accepted
    proposal to graph node provenance mismatches.
  • identity build still does not render a graph. It now reports:
    Memory proposals are active., Graph rendering is not active yet., and
    Accepted memory nodes are candidate project memory.

Accepted/rejected policy:

  • v0.2 only de-duplicates pending proposals.
  • accepted/rejected proposals remain historical user decisions.
  • Re-proposing after accept/reject may create a new pending proposal id instead
    of rewriting history.

Explicitly not included:

  • Memory Graph rendering
  • Obsidian-style dashboard graph
  • MCP/hooks/subagents/role evolution
  • auto planner or auto execution loop
  • raw prompt archive
  • automatic memory write

Verification Checklist

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

Fresh temp smoke:

node bin/orange.js init
node bin/orange.js quest new "remember a durable decision" --layer L2 --json
node bin/orange.js quest done <quest-id> --evidence "manual smoke passed" --json
node bin/orange.js remember propose --quest <quest-id> --json
node bin/orange.js remember propose --quest <quest-id> --json
node bin/orange.js remember list --status pending --json
node bin/orange.js remember list --type decision --json
node bin/orange.js remember accept <proposal-id> --json
node bin/orange.js doctor --json
node bin/orange.js identity build --json

v0.2.0-alpha.0

Memory Delta Proposal alpha release.

This release adds the proposal-first memory workflow without enabling full
Memory Graph rendering or automatic memory writes.

  • Package version is 0.2.0-alpha.0.
  • Adapter JSON contract_version remains "0.1".
  • New CLI commands:
    • orange remember propose --quest <quest-id>
    • orange remember list
    • orange remember show <proposal-id>
    • orange remember accept <proposal-id>
    • orange remember reject <proposal-id>
  • Completed L2+ Quests can produce pending Memory Delta Proposals when they have
    verification evidence or an unverified reason.
  • L0/L1 Quests are blocked from default memory proposal generation.
  • accept moves a pending proposal to accepted and creates a graph node
    candidate with proposal/source Quest provenance.
  • reject moves a pending proposal to rejected and does not create graph nodes.
  • doctor validates proposal schema, source Quest existence, status/location
    consistency, accepted graph provenance, rejected no-node state, graph JSON,
    and traversal-safe ids.
  • identity build remains a placeholder and adds memory proposal/node counts
    plus top proposal node types. It does not render a graph.

Explicitly not included:

  • Memory Graph rendering
  • Obsidian-style dashboard graph
  • MCP/hooks/subagents/role evolution
  • auto planner or auto execution loop
  • raw prompt archive
  • unapproved graph node creation
  • default L0/L1 memory proposal generation

Verification Checklist

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

Fresh temp smoke:

node bin/orange.js init
node bin/orange.js quest new "remember a durable decision" --layer L2 --json
node bin/orange.js quest done <quest-id> --evidence "manual smoke passed" --json
node bin/orange.js remember propose --quest <quest-id> --json
node bin/orange.js remember list --json
node bin/orange.js remember show <proposal-id> --json
node bin/orange.js remember accept <proposal-id> --json
node bin/orange.js doctor --json
node bin/orange.js identity build --json

v0.1.0

Seed Kernel stable release.

This stable release does not add new core features after v0.1.0-alpha.4.
It promotes the frozen Seed Kernel CLI and Adapter Contract to 0.1.0.

  • Package version is 0.1.0.
  • Adapter JSON contract_version remains "0.1" and is the stable v0.1 contract.
  • Stable usage is npx orange-hyper ... after publish.
  • Alpha usage remains npx orange-hyper@alpha ... for the alpha channel.
  • This release keeps the Seed Kernel boundary: no Memory Graph, MCP, hooks,
    subagents, role system, auto planner, or auto execution loop.

Stable Publish Method

0.1.0 is stable, so publish it to npm without an explicit prerelease tag.
The npm publish command uses the
tag config to
decide which dist-tag to apply, and its default is latest; alpha releases
should continue to use npm publish --tag alpha.

Do not run the publishing commands until the release is explicitly approved.

npm version 0.1.0 --no-git-tag-version

npm test
git diff --check
node bin/orange.js --help
npm pack --dry-run --cache /private/tmp/orange-hyper-npm-cache

git add -A
git commit -m "chore: release orange-hyper v0.1.0 seed kernel stable"

git tag -a v0.1.0 -m "orange-hyper v0.1.0"
git push origin main
git push origin v0.1.0

gh release create v0.1.0 \
  --title "orange-hyper v0.1.0" \
  --notes-file RELEASE_NOTES.md

npm publish

After publish:

npm dist-tag ls orange-hyper
npx orange-hyper --help

Expected dist-tags:

latest: 0.1.0
alpha: 0.1.0-alpha.4

Alpha Dogfooding Evidence

The alpha channel was used to harden the stable Seed Kernel contract before
0.1.0:

  • v0.1.0-alpha.2 added adapter-friendly JSON output for quest new and
    route --quest.
  • v0.1.0-alpha.3 added JSON output for capsule, quest done, doctor, and
    identity build.
  • v0.1.0-alpha.4 froze the adapter envelope with contract_version: "0.1",
    dot-notation command ids, and explicit stdout/stderr behavior.
  • npx orange-hyper@alpha --help returned the Seed Kernel command list.
  • npx orange-hyper@alpha dogfooding in a fresh temp directory verified:
    init, quest new --json, route --json, capsule --json,
    quest done --json, doctor --json, and identity build --json.
  • The alpha dogfood JSON outputs returned contract_version: "0.1" with
    command ids quest.new, route.show, capsule.build, quest.done,
    doctor.run, and identity.build.
  • Pre-stable package dry-run produced orange-hyper@0.1.0; after publish,
    0.1.0 should become latest while alpha remains 0.1.0-alpha.4.

v0.1.0-alpha.4

Adapter Contract Freeze release.

  • JSON success and failure envelopes now include "contract_version": "0.1".
  • JSON envelope command values now use dot notation such as quest.new, route.show, capsule.build, quest.done, doctor.run, and identity.build.
  • docs/16_ADAPTER_CONTRACT.md now fixes stdout/stderr behavior for JSON and human modes.
  • README points to the adapter contract instead of repeating JSON envelope details.
  • This release keeps the v0.1 Seed Kernel scope: no Memory Graph, MCP, hooks, subagents, role system, auto planner, or auto execution loop.

v0.1.0-alpha.3

Seed Kernel Adapter Contract release.

  • orange capsule --quest <id> --json, orange quest done ... --json, orange doctor --json, and orange identity build --json now provide machine-readable output.
  • JSON success payloads use the { ok, command, data } envelope.
  • JSON failures use the { ok, command, error } envelope when the executable catches an error in --json mode.
  • docs/16_ADAPTER_CONTRACT.md defines adapter rules, command examples, error shape, and exit codes.
  • README clarifies that skills/adapters should call JSON mode and that users are not expected to manually drive every command long-term.

v0.1.0-alpha.2

Seed Kernel UX/API hardening release.

  • orange quest new now prints a copy-friendly quest id line, file path, and next kernel commands.
  • orange quest new --json and orange route --quest <id> --json provide machine-readable output for skills and adapters without mixing human text.
  • orange quest done accepts --evidence-file <path> for longer UTF-8 verification evidence.
  • Completion now rejects evidence and --unverified together.
  • README clarifies that the CLI is the Seed Kernel interface, not the final user experience.

v0.1.0-alpha.0

Orange Hyper Seed Kernel의 첫 alpha 릴리즈입니다. 이 릴리즈는 강한 자동화 하네스가 아니라 repo-local Quest, Route, Capsule 기록을 남기는 최소 커널입니다.

Included CLI

  • orange init
  • orange quest new
  • orange quest list
  • orange quest show
  • orange quest done
  • orange route
  • orange capsule
  • orange doctor
  • orange identity build

Core Concepts

  • Quest as editable intent capsule
  • Route contract as public work contract
  • verified/unverified completion
  • path traversal protection
  • identity placeholder

Explicitly Not Included

  • hooks
  • MCP
  • subagents
  • role evolution
  • auto planner
  • auto loop
  • branch/PR/spec workflow enforcement
  • forced Quest creation for every request
  • runtime automation
  • telemetry/network behavior
  • postinstall mutation
  • provider/model bridge
  • Memory Graph rendering

Verification Results

  • npm test: passed, 19 tests
  • git diff --check: passed
  • CLI smoke: node bin/orange.js --help passed
  • npm pack --dry-run: passed package contents check

Manual Release Commands

Do not run these until the release is explicitly approved.

npm test
git diff --check
npm pack --dry-run
git add -A
git commit -m "chore: prepare v0.1.0-alpha.0 seed kernel release"
git tag v0.1.0-alpha.0
git push origin main
git push origin v0.1.0-alpha.0
npm publish --tag alpha
Assets 2
Loading