Changelog

This page tracks the public docs-facing history of Claude Code Orchestrator Skill.

v0.7.1 - Manual retry invalidates workflow success

• Fixed GitHub issue #24.
• workflow-retry-node now changes a manually invalidated workflow from succeeded to needs_rerun.
• Invalidated nodes now clear current handoff, validation, gate, run id, token, and cost evidence, while keeping a compact stale_evidence marker.
• Workflow reports now show a visible manual-invalidation warning so Codex and dashboards cannot accept stale success.
• Selftest now covers manual retry status, stale evidence, and report warning behavior.

v0.7.0 - Workflow DAG, handoff contracts, and node gates

• Added the first workflow DAG controller layer for GitHub issues #20, #21, and #22.
• Added workflow-validate, workflow-dry-run, workflow-run --mock, workflow-status, workflow-retry-node, workflow-stop, and workflow-report. Real DAG worker execution is intentionally disabled in v0.7.0.
• Added handoff-template, handoff-validate, handoff-read, and handoff-repair-prompt.
• Added MCP tools for workflow and handoff control.
• Added mock controller gates for retry decisions, max-retry blocking, missing handoff blocking, loop guard, and decision-trail reports.
• Added examples/workflows/safe-refactor.yaml.

v0.6.4 - Data-proven worker supervision fixes

• Fixed GitHub issues #16, #17, and #18.
• --final-only now filters raw stream noise before applying the persisted stdout budget and writes compact final text.
• run, run-streaming, and run-visible now use the --cwd scoped artifact root, with a run index so controller commands can still find runs by id.
• Actual token aggregates are computed from raw Claude modelUsage before redaction.
• Windows PID checks now use a system API instead of slow tasklist calls.
• mock-stream-test now includes gates for final-only noise filtering, cwd artifact routing, and token aggregate preservation.

v0.6.3 - Docs deploy stability

• Fixed the GitHub Actions docs deploy secret-scan false positive caused by a selftest placeholder token.
• Kept placeholder-secret regression coverage while splitting the sample token string so repository scans do not flag it as a real key.
• Synced README, docs changelog, package metadata, and version metadata for the hotfix.

v0.6.2 - Actual model attribution

• Fixed #15 by recording Claude stream modelUsage as actual_model_usage.
• Added actual_model, actual_cost_usd, actual_total_tokens, and route_mismatch to run status and metadata.
• detect_failure_modes now flags route mismatches as high-severity controller risks.
• usage-summary, dashboard, and controller reports now distinguish declared route from actual billed model.
• Added supervise-decision as a compatibility alias for decision-review.

v0.6.1 - Issue audit completion

• Expanded controller-report / pressure-report Markdown with by-model usage totals, total duration, token estimates, output bytes, event bytes, budget stops, warning counts, blocking counts, and max severity.
• Added richer per-run report rows: duration, token estimate, stdout/events bytes, warning/blocking counts, budget state, and source/artifact counts.
• Added token estimates to the dashboard output-budget panel.
• Added warning/blocking risk counts to usage-summary and its by-model breakdown.
• Routed remaining legacy metadata writes through the shared UTF-8/control-character sanitizer.

v0.6.0 - Controller operations hardening

• Fixed GitHub issues #3-#12.
• Added transactional spawn-role-team capacity checks and rollback for partial launches.
• Added hard output/event budgets, final-only mode, and output-budget stop reasons for streaming workers.
• send-instruction now preserves the prior profile/model by default and reports route drift.
• Added UTF-8 JSON/control-character safeguards for Chinese paths, prompts, metadata, and dashboard output.
• Split risk semantics into blocking state, warnings, max severity, and compatibility ok.
• Classified secret scan findings without printing raw secret values.
• Split project source changes from .agent-workspace agent artifacts.
• Upgraded the dashboard into a controller operations panel.
• Added controller-report / pressure-report plus MCP report tools.
• Added supervisor-style decision-review / cc_decision_review.

v0.5.1 - Portable assets and safer cleanup

• Fixed GitHub issues #1 and #2.
• Portable tools/cc-orchestrator copies now discover version.json and Prompt Pack assets.
• CC_ORCHESTRATOR_SKILL_ROOT is now honored when users want to point the tool at a full Skill package.
• healthcheck now reports skill_root, version_path, prompt_pack_path, and Prompt Pack availability.
• clean-workspace now preserves freshly initialized scaffold folders instead of suggesting that they be deleted.
• Selftest now covers Prompt Pack availability and scaffold-preserving cleanup.

v0.5.0 - Workspace governance

• Added .agent-workspace/claude-code-orchestrator as the default home for agent-generated artifacts.
• Added init-workspace, workspace-status, migrate-data, clean-workspace, archive-runs, repair-mcp-paths, and folder-policy.
• Added matching MCP tools for every workspace governance command.
• Updated worker prompts and generated CLAUDE.md so logs, reports, temp files, and rollback notes stay under the managed artifact root.
• Updated install examples and docs so artifact routing works across machines.

v0.4.1 - Controller checkpoints and queue polish

• Added rolling checkpoint-###.md summaries for long-running workers.
• Added deduplicated tool-call summaries, such as Grep x7 and Read x3.
• Made controller-mode poll-run write compact controller artifacts by default.
• Added queued, running, done, failed, timed_out, and cancelled queue states.
• Improved the local dashboard layout for model routing, timeline, logs, diff, risk, and controls.

v0.4.0 - Codex controller system

• Added references/codex-controller-playbook.md.
• Added Prompt Pack templates for repo audit, bugfix, security audit, frontend polish, test generation, refactor planning, and release checks.
• Added compact run polling, cc_summarize_run, and cc_compact_events.
• Added real queue policy, model registry, benchmark history, local override preservation, worker quality history, and failure-mode detection.
• Added timeline dashboard and daily update-check guidance.

v0.3.0 - Verification and safer operations

• Added one-click cc_verify_run.
• Chained diff summary, write-scope checks, secret scanning, optional tests, and Markdown reports.
• Added mock streaming tests that do not spend model quota.
• Added usage summaries, upgrade checks, benchmark suite, and MCP auto-registration.
• Added version.json as the version metadata source.

v0.2.0 - Live worker control

• Added run-streaming / cc_run_streaming_agent.
• Started Claude Code with --output-format stream-json --include-partial-messages.
• Added poll-run, run-status, and stop-run.
• Added role team spawning, result collection, cross review, dashboard, report export, cost guard, and visible worker windows.

v0.1.0 - Skill, MCP, CLI, and CCSwitch foundation

• Created the Codex Skill entrypoint.
• Added the bundled MCP server and CLI orchestrator.
• Added CCSwitch profile discovery and Claude Code binary discovery.
• Added local model scoring by role and role-based model routing.
• Added safe defaults, run metadata, logs, CLAUDE.md generation, UTF-8 output handling, and secret redaction.