TomorrowEdge 1.6.1 Canopus

TomorrowEdge 1.6.1 Canopus is a hardening release for the Convergence Runtime.

Highlights:

• tedge canopus init now generates the public Canopus schema: objective/acceptance/convergence, target_state, blocking_checks, advisory_checks, and reviewer_role.
• Legacy goal/evaluation/loop specs remain supported for v1.6 compatibility, with tedge control now emitting a compatibility warning on stderr.
• Fixed the default init template warning by aligning verification_passes between success conditions and blocking checks.
• Added src/core/canopus/* public re-exports for Canopus naming while keeping src/core/controlPlane compatible.
• Added examples/canopus/simple_bugfix_runtime/objective.yaml as the public hard-gate runtime demo.
• Hardened package:smoke with tedge client --smoke-once, so installed-package GUI smoke exits reliably.
• Package files now include README GUI screenshots and the public Canopus example.
• Split timeout-bounded CLI coverage into tests/unit/canopusCli.test.ts.

Validation:

• npm ci
• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run audit:check
• npm run test:core
• npm run test:council
• npm run test:control
• npm run package:smoke
• npm run pack:dry:strict (full test: 66 files / 578 tests)

Package hygiene:

• Release zip excludes .env, .runs, and assignments local artifacts.
• npm pack --json --ignore-scripts confirms GUI screenshots and the Canopus public demo are included.

GitHub triage:

• Open PRs #481 and #487 were inspected. Both are still draft GUI PRs, so they remain open for the next GUI-specific merge pass rather than being folded into this Canopus hardening release.

TomorrowEdge 1.6.0 Canopus

TomorrowEdge 1.6.0 Canopus is the Convergence Runtime Release.

Canopus adds a convergence layer inside TomorrowEdge's existing local orchestration, governance, and strategy-evolution runtime for heterogeneous Coding Agents. Agents no longer finish because they claim completion; a run must satisfy an ObjectiveContract, pass an AcceptanceMatrix, leave evidence, update RunState, and respect a bounded ConvergencePolicy.

Highlights:

• Added Canopus Runtime with ObjectiveContract-compatible target loading, AcceptanceMatrix checks, ConvergencePolicy bounds, RunState persistence, EvidenceGate integration, Trace Ledger artifacts, and ConvergenceEngine loop control.
• Added tedge canopus plus tedge control compatibility commands for init, validate, run, status, and report.
• Added mock, noop, shell, and Sirius Council AgentBridge paths.
• Added real simple_bugfix_runtime acceptance demo and hard-gate/blocking-check veto demo.
• Added pre/post-action acceptance snapshots, run-baseline diff semantics, allowed/denied path enforcement, and required evidence checks.
• Preserved TomorrowEdge's project-level positioning as a local orchestration, governance, and strategy-evolution runtime for heterogeneous Coding Agents.

Validation:

• npm run verify:release-gates
• npm run test:control
• npx vitest run tests/unit/cliContract.test.ts --fileParallelism=false
• npx tsx src/cli/index.ts canopus validate examples/control_plane/simple_bugfix_runtime/goal.yaml

Package hygiene:

• Release zip excludes .env, .runs, and assignments local artifacts.
• npm package scope is narrowed so example runtime state is not packed.

TomorrowEdge 1.5.1 Sirius

TomorrowEdge 1.5.1 hardens the Sirius Agent Council real-workflow path.\n\nHighlights:\n- Preserves explicit user verification boundaries with a bounded file verifier for generated-file / HTML / SVG deliverables.\n- Improves OpenRouter/OpenAI-compatible structured JSON calls by suppressing reasoning output where supported and adding per-request timeout/retry controls.\n- Adds same-model structured repair for planner and scenario profiler JSON instead of local semantic fallback.\n- Adds multi-file live patch output support and safer risk normalization for generated document artifacts.\n- Adds conservative Windows CLI text normalization for Chinese/UTF-8 task text.\n- Absorbs community GUI fixes for focused cockpit shell verification and Markdown answer rendering with copyable code blocks.\n\nVerified with npm test, npm run build, npm run web:build, npm run docs:status, npm run secrets:scan, npm run audit:check, package smoke, and npm run pack:dry.

TomorrowEdge v1.5.0 Sirius

TomorrowEdge v1.5.0 Sirius

Sirius turns TomorrowEdge into the local governance and policy-evolution runtime for heterogeneous coding agents.

This release moves the main story away from benchmark/dashboard expansion and back to the core product thesis: strong-agent governance, budget-bounded heterogeneous execution, Agent Council planning, owned TaskGraph delegation, bounded strategy mutation, and Chief Agent final review.

Highlights

• Chief Agent Router: high-level engineering goals enter a Chief Agent first.
• AgentCapabilityProfile: Codex, Claude Code, DeepSeek, MiMo, Ollama, mock, fixture, command agents, and custom adapters are replaceable capability nodes rather than hardcoded defaults.
• Agent Council Planning: council members produce structured critique, gap fill, alternative plans, task claims, and consensus moves.
• Task Ownership Assignment: core TaskGraph nodes carry ownerAgentId, assigned provider/model, claim mode, fallback candidates, and human-readable assignment reasons.
• Delegated Execution Runtime: owned nodes execute under Objective Contract, EvidenceGate, BudgetGate, Debate v2, Strategy Memory, and Trace Ledger.
• Bounded Strategy Mutation: execution failures can trigger owner switch, reviewer/judge insertion, debate increase, task split, or council replan without mutating safety boundaries.
• Chief Final Review / Judge: deliverables return to the Chief Agent before final completion.
• Model semantic routing alignment: task/scenario/workflow classification is routed through model-backed semantic analysis instead of brittle hardcoded regex paths.
• README/product docs: first screen now defines TomorrowEdge as a local governance and policy-evolution runtime for heterogeneous coding agents; benchmark and dashboard utilities are explicitly secondary.

Commands

tedge council run "refactor this app with a chief agent and council review"
tedge run "fix this failing workflow" --agent-council
tedge serve --open

Verification

Release gates used for the current v1.5.0 tag alignment:

• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run test:council

Earlier 1.5.0 pack gate also passed full package verification before publication.

TomorrowEdge 1.4.3

TomorrowEdge 1.4.3 clears the older experiment and fixture issue queue.\n\nHighlights:\n- error-loop experiments now support explicit baseline aliases: direct, reflection_only, preference_feedback, and error_memory.\n- deterministic experiment fixtures now cover wrong-API, wrong-file, hidden-invariant, cross-language, React/UI, state-machine, and flaky-validator scenarios.\n- tedge experiment dashboard builds an offline HTML cohort dashboard and dashboard_summary.json from error-loop experiment bundles.\n\nValidation:\n- npx vitest run tests/unit/errorLoopExperiment.test.ts tests/unit/cliContract.test.ts --fileParallelism=false\n- npm run build\n- npm run web:build\n- npm run docs:status\n- npm run secrets:scan\n- npm run audit:check\n- npm run pack:dry:strict

TomorrowEdge v1.4.2

TomorrowEdge v1.4.2 polishes the 1.4.1 alpha candidate.

Highlights:

• Fixture mode is now clearly bounded as a deterministic sample workflow provider, with CLI warnings outside bundled fixture/sample repos.
• Strategy memory normalizes failing-test repair tasks as bugfix plus secondary test_failure signal.
• Cockpit RoleGraph state now reads authoritative execution status, attempts, and timing.
• Summarize TaskGraph nodes now link final summary, trace completeness, and objective trace artifacts.
• Added targeted test groups for integration, UI, memory, core, and adaptive runtime checks.

Validation:

• npm run release:verify passed.
• npm test passed: 60 files / 484 tests.
• Build, web build, docs status, secrets scan, audit check, package smoke, and strict pack dry-run passed.

Assets:

• tomorrowedge-1.4.2-latest-20260611-013052.zip: source/release bundle with cockpit web assets.
• tomorrowedge-1.4.2.tgz: npm package tarball.

TomorrowEdge v1.4.0

v1.4.0 Adaptive Orchestration Runtime

TomorrowEdge 1.4.0 moves the Objective Contract / Policy Genome work deeper into runtime execution instead of leaving it as trace-only metadata.

Highlights

• TaskGraph Planner: plans now carry validated task nodes, dependencies, roles, evidence requirements, and verification links.
• RoleGraphScheduler v0: role graph execution drives candidate, review, judge, runner, and verification/repair phases and records role-node results in the event ledger.
• Evidence Dependency: reviewer, judge, runner, and repair phases emit evidence-gap events; blocking gaps now hard-stop governed actions with workflow_stop_reason.
• Debate Protocol v2: debate is represented as structured claims, rebuttals, evidence refs, unresolved issues, and judge resolutions.
• Policy Evolution v2: policy variants support counterfactual replay and tournament summaries over objective traces.
• External Agent Adapter Layer: generic, Codex, and Claude Code profiles normalize external outputs before they enter the workflow.
• Strict external-agent semantics: strict normalization failures abort without native fallback and remain visible as external_agent_normalization / external_agent_error / role_node_result events.
• Planner cache isolation: access mode, workflow intent, allowed phases, roles, and tools are part of the planner cache key.
• Approval semantics: patch/shell approval requirements remain waiting-for-user states and no longer poison required RoleGraph nodes as failed executions.
• Unified BudgetGate: model planner, live advisory/patch, and pre-judge debate invocations pass through the same budget decision path.

Validation

• npm test
• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run audit:check
• npm run pack:dry

Note: Vite still prints the existing Tailwind content warning during web builds; build succeeds.

PRs: #442, #444
Issues: #441, #443

TomorrowEdge v1.3.10

TomorrowEdge v1.3.10

This patch corrects the 1.3.8 answer-first path.

• Read-only/chat answers must now be generated by the configured summarizer model route or explicitly blocked.
• Provider fallback is disabled for user-facing answer generation, so unavailable answer models no longer silently fall back to local templates or mock routes.
• Removed local hardcoded answer fallbacks for greetings, mathematical proofs, repository summaries, and evidence-derived completed answers.
• Mock provider user replies no longer contain task-specific knowledge; they identify themselves as offline test-provider output.
• Cockpit no longer synthesizes a completed answer when no model-generated reply exists.

Validation:

• npm test
• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run audit:check
• npm run pack:dry:strict

TomorrowEdge v1.3.9

TomorrowEdge v1.3.9 fixes a Cockpit status projection bug for patch-like tasks that produce no files.

Highlights:

• Patch-like workflows that finish with no changed files and no patch applied now appear as failed/needs revision instead of green Done.
• The main panel now shows No patch generated with a clear retry/revision explanation.
• Workflow evidence remains available in supporting details.

Issue addressed:

• #431 GUI no longer presents missing patch deliverables as successful Done tasks.

Verified:

• npx vitest run tests/unit/cockpitViewModel.test.ts --fileParallelism=false: 15 tests passed
• npm test: 414 tests passed
• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run audit:check
• npm run pack:dry:strict

TomorrowEdge v1.3.8

TomorrowEdge v1.3.8 is an answer-first Cockpit hotfix for read-only, chat-style, and governed advisory tasks.

Highlights:

• Completed Cockpit tasks now carry a user-facing userReply in final summaries.
• Read-only and ask-user workflows can use the configured summarizer provider for direct natural-language answers, with deterministic local fallback for fixture/no-provider runs.
• Governed read-only reasoning tasks now record reviewer/judge advisory evidence plus an evidence packet, so strict policy thresholds do not downgrade correct answer-only workflows just because there is no patch.
• The GUI main panel now shows the answer as primary content and moves workflow evidence into collapsible supporting details.

Issues addressed:

• #429 completed Cockpit tasks now produce a visible user-facing reply.
• #430 the Cockpit main view now prioritizes answer/result content over internal orchestration telemetry.

Verified:

• npx vitest run tests/unit/agentGraph.test.ts tests/unit/cockpitViewModel.test.ts tests/unit/cockpitWeb.test.ts --fileParallelism=false: 77 tests passed
• npm test: 413 tests passed
• npm run build
• npm run web:build
• npm run docs:status
• npm run secrets:scan
• npm run audit:check
• npm run pack:dry:strict