Claw-Empire v2.0.4 Release Notes

• Release date: 2026年03月12日
• Scope: first-class Docker deployment support, stale working-agent recovery, official direct-API preset onboarding, Kimi Code provider support, and development-default API assignment across hydrated office packs.

Highlights

1. Docker Deployment Is Now a First-Class Path

• Added a production-oriented Dockerfile, .dockerignore, and docker-compose.yml.
• Container runtime now uses a non-root user and standardized writable data locations.
• Runtime path resolution was updated so database/log paths behave correctly inside Docker deployments.
• README deployment guidance was added across English, Korean, Japanese, and Chinese.

2. Stale working Agents Now Recover Automatically

• Startup and lifecycle sweeps now detect agents that are still marked working even though their current task is missing or no longer in_progress.
• Those agents are reset back to idle with current_task_id = NULL.
• Result: server restarts and interrupted runs no longer leave misleading "still working" agents behind in the UI.

3. Official Direct-API Presets Reduce Manual Setup Friction

• Settings > API now includes official presets for:
  • OpenCode Go (OpenAI)
  • OpenCode Go (Anthropic)
  • Bailian Coding Plan (OpenAI)
  • Bailian Coding Plan (Anthropic)
• Presets now:
  • lock and manage the expected Base URL
  • seed fallback model lists immediately
  • keep refresh/retry explicit instead of silently retry-looping after failures
  • show preset identity directly on provider cards
• Existing provider-card model search remains available, so large cached model lists are still easy to filter before assignment.

4. Kimi Code Is Now Treated as a First-Class Provider

• Kimi Code is now supported through the CLI execution path alongside the existing local coding providers.
• The integration was completed across:
  • provider selection and display labels
  • task execution/provider validation
  • prompt-skill rendering
  • skill learn/unlearn provider routing
  • video-preprod remotion bootstrap
  • provider checks in existing database schema/runtime migration paths
• Result: Kimi is no longer just a partial label-level addition; it is wired through the execution and skills flow end to end.

5. API Model Assignment Now Starts With Development and Includes Initialized Packs

• The assign flow now always loads the development office as the default baseline.
• If a user has already completed initial setup for other office packs and those packs have been hydrated into runtime state, their agents and departments also appear in the assignment modal.
• Pack sections are now grouped safely, so packs that reuse the same department IDs no longer collapse into each other inside the modal.
• Result: default development assignment remains predictable, while already-initialized office packs can also receive direct API model mappings from the same settings flow.

6. Local E2E Validation Is Safer and Leaves Less Residue

• pnpm run test:e2e now resets the isolated .tmp/e2e-runtime database both before and after the Playwright run.
• Playwright no longer reuses an already-running local 8810 server by default. Reuse is now opt-in via PW_REUSE_EXISTING_SERVER=1.
• CRUD-heavy E2E scenarios were hardened so local validation does not keep leaving stale departments, agents, and similar CI residue behind during normal development verification.
• Result: local E2E runs are more predictable, are less likely to touch the wrong runtime, and clean up their temporary state more aggressively.

PR Attribution

• Includes PR #56: Docker deployment support (Dockerfile, compose, runtime path handling, multilingual docs)
• Includes PR #60: stale/orphan working-agent recovery during lifecycle startup and interval sweeps
• Includes intent-preserving integration of current open PR work:
  • PR #62: development-pack-scoped API assignment departments
  • PR #63: Kimi Code provider support and migration/runtime completion
  • PR #64: official API presets, preset refresh hardening, and provider-card UX updates
  • post-integration follow-up: API assign modal now includes hydrated office packs while keeping development as the default baseline

Files Updated in This Release

• .dockerignore
• Dockerfile
• docker-compose.yml
• README.md
• README_ko.md
• README_jp.md
• README_zh.md
• docs/releases/README.md
• docs/releases/v2.0.4.md
• server/config/runtime.ts
• server/db/runtime.ts
• server/modules/bootstrap/schema/api-providers-schema.test.ts
• server/modules/bootstrap/schema/base-schema.ts
• server/modules/bootstrap/schema/oauth-runtime.test.ts
• server/modules/bootstrap/schema/oauth-runtime.ts
• server/modules/bootstrap/schema/task-schema-migrations.ts
• server/modules/lifecycle.ts
• server/modules/routes/ops/api-providers.test.ts
• server/modules/routes/ops/api-providers.ts
• server/modules/routes/ops/skills/learn-constants.ts
• server/modules/routes/ops/skills/learn-core.ts
• server/modules/routes/ops/skills/types.ts
• server/modules/workflow/agents/providers/usage-cli-tools.ts
• server/modules/workflow/core/cli-tools.ts
• server/modules/workflow/core/prompt-skills.test.ts
• server/modules/workflow/core/prompt-skills.ts
• server/modules/workflow/core/video-skill-bootstrap.test.ts
• server/modules/workflow/core/video-skill-bootstrap.ts
• server/modules/workflow/packs/video-artifact.ts
• src/api-provider-presets.test.ts
• src/api/providers-reports-github.ts
• src/api/workflow-skills-subtasks.ts
• src/components/AgentDetail.tsx
• src/components/AgentStatusPanel.tsx
• src/components/agent-detail/constants.ts
• src/components/agent-manager/constants.ts
• src/components/settings/ApiSettingsTab.test.tsx
• src/components/settings/ApiSettingsTab.tsx
• src/components/settings/Logos.tsx
• src/components/settings/constants.tsx
• src/components/settings/types.ts
• src/components/settings/useApiProvidersState.test.tsx
• src/components/settings/useApiProvidersState.ts
• src/components/skill-history/utils.ts
• src/components/skills-library/model.tsx
• src/types/index.ts

Verification

• pnpm run test:ci
• pnpm run build
• pnpm run test:e2e

Compatibility Notes

• A schema/runtime migration is included for existing provider checks so legacy databases can accept kimi in the relevant provider columns.
• Docker deployment is additive and optional; existing local non-Docker workflows remain supported.
• Official API presets are backward compatible with manual API provider configuration.

Claw-Empire v2.0.3 Release Notes

• Release date: 2026年03月08日
• Scope: selective PR #54 salvage, task worktree verification visibility, report reliability fixes, and optional self-host operations references.

Highlights

1. Final Branch Verification Is Now Visible Before Merge

• Added GET /api/tasks/:id/verify-commit so the app can inspect a task worktree before merge.
• The endpoint evaluates commit count, changed files, and uncommitted files, then returns a verdict such as:
  • ok
  • no_worktree
  • no_commit
  • dirty_without_commit
  • commit_but_no_code
• Internal helper artifacts such as .claude/skills are excluded from the code-change verdict so non-code support files do not create false positives.

2. Diff Modal Now Shows Final Branch Verification

• The task diff modal now loads worktree verification alongside the git diff.
• You can see:
  • final verification verdict
  • compare/base ref
  • commit count
  • changed files
  • uncommitted file count
• Result: merge and discard decisions are now made with explicit worktree evidence instead of raw diff text alone.

3. Completion Reports Now Preserve Final Verification Evidence

• Manual merge flow now appends a system log entry such as Final branch verification: passed (...) before merge.
• The completion report popup surfaces those verification logs in the planning summary view.
• Result: completed task reports now retain proof that the final branch was checked before merge.

4. Report Avatars No Longer Fall Back to Emoji for Missing Active Agents

• Report history and task-completion popup avatars no longer depend only on the currently active office-pack agent list.
• When a report references an assignee who is not present in the current agents array, the UI now builds a report-scoped fallback agent from the report payload and still renders a sprite avatar.
• Result: completed report rows and popups continue to show the agent face instead of degrading to emoji when the active agent roster has changed.

5. Cleanup Utility for Staff State Auditing

• Added scripts/cleanup-staff.mjs.
• The utility can:
  • reset break staff back to idle
  • report role distribution
  • detect duplicate names
• It prefers DB_PATH when provided, making it usable across different local environments.

6. Optional Self-Host Deployment References

• Added optional deployment reference files under deploy/:
  • deploy/README.md
  • deploy/.env.production.template
  • deploy/claw-empire@.service
  • deploy/nginx/claw-empire.conf
• These are reference templates for self-hosted setups, not a required runtime migration.

PR Attribution

• This release includes selected salvage from PR #54, limited to independent operational/documentation ideas that fit the current product structure.
• It does not include the broader PR #54 task model, admin console, local-server, or work-phase proposals.

Files Updated in This Release

• deploy/.env.production.template
• deploy/README.md
• deploy/claw-empire@.service
• deploy/nginx/claw-empire.conf
• docs/releases/README.md
• docs/releases/v2.0.3.md
• README.md
• README_ko.md
• README_jp.md
• README_zh.md
• scripts/cleanup-staff.mjs
• server/modules/routes/ops/worktrees-and-usage.test.ts
• server/modules/routes/ops/worktrees-and-usage.ts
• server/scripts/cleanup-staff.test.ts
• src/api.test.ts
• src/api/workflow-skills-subtasks.ts
• src/components/ReportHistory.test.tsx
• src/components/ReportHistory.tsx
• src/components/TaskReportPopup.test.tsx
• src/components/TaskReportPopup.tsx
• src/components/task-report-agent.ts
• src/components/taskboard/DiffModal.test.tsx
• src/components/taskboard/DiffModal.tsx

Compatibility Notes

• No schema migration is required for v2.0.3.
• Deployment templates under deploy/ are optional references only.
• The release is focused on safer merge visibility, more trustworthy completion reports, and small operational salvage from PR #54.

Claw-Empire v2.0.2 Release Notes

• Release date: 2026年03月07日
• Scope: CI/E2E coverage expansion, contributor-facing API documentation parity refresh, and reproducible OpenAPI validation for the main branch.

Highlights

1. CI Now Blocks OpenAPI Contract Drift

• test:ci now runs openapi:check before E2E execution.
• This makes undocumented or stale OpenAPI paths fail in CI instead of being discovered later in manual review.
• Result: API docs, generated OpenAPI, and shipped routes are now checked as one release gate.

2. New CI E2E Coverage for Operational API Paths

• Added CI-focused E2E coverage for execution-control and orchestration-critical paths:
  • POST /api/tasks/:id/run
  • POST /api/tasks/:id/stop
  • POST /api/tasks/:id/inject
  • POST /api/tasks/:id/resume
  • GET /api/tasks/:id/terminal
  • GET /api/tasks/:id/meeting-minutes
• Added directive-webhook coverage for POST /api/inbox so task-routing and announcement entry flow are exercised in CI.
• Added project-path and diagnostics coverage for:
  • GET /api/projects/path-check
  • GET /api/projects/path-suggestions
  • GET /api/projects/path-browse
  • GET /api/cli-status
  • GET /api/cli-usage
  • API provider preset/CRUD routes

3. Docs and Public API Surface Are Now Exercised in CI

• Added CI checks for live docs routes:
  • GET /api/docs
  • GET /api/docs/swagger-bootstrap.js
  • GET /api/openapi.json
• Added public-surface verification for contributor-facing utility and ops routes so documentation drift is caught alongside route regressions.
• Coverage is now split across dedicated CI specs for:
  • operational API flows
  • docs and ops route availability
  • public API surface expectations

4. API Documentation Was Brought Back to Route Parity

• Refreshed docs/api.md and docs/openapi.json to cover contributor-facing routes that had fallen behind implementation.
• Newly documented or clarified coverage includes categories such as:
  • task reports and task execution control
  • project detail/branch/path helper routes
  • subtasks and agent spawn flows
  • announcements and directives
  • GitHub status/repo/clone helpers
  • OAuth status/account lifecycle routes
  • skills learn/history/custom routes
  • sprite processing/registration
  • auto-update status/apply/config routes
• Result: main-branch docs now describe the practical API surface used by contributors and CI.

5. E2E Inbox Verification No Longer Depends on Local Secrets

• CI inbox-webhook coverage now uses a dedicated E2E-only secret instead of any developer-local .env secret.
• This keeps the test reproducible in CI and prevents accidental leakage of real local webhook secrets into tracked test fixtures.

Post-release Additions Still Included in v2.0.2

6. PR #49 Follow-up: Discord Channel Docs/Test Cleanup

• Removed the nonexistent 409 response from POST /api/messenger/discord/channels in generated OpenAPI output.
• Added a mockReset() guard for Discord channel listing tests so characterization coverage stays isolated across runs.
• Result: the shipped API contract and the actual Discord channel route behavior now stay aligned in both docs and tests.

7. PR #55: API Provider Model Search

• The API provider settings panel now includes an inline model search field when a provider's cached model list is expanded.
• Filtering happens locally per provider, so large model lists can be narrowed without reloading or collapsing the provider card.
• Added a dedicated UI regression test to ensure the filtered model list only shows matches and keeps non-matching models hidden.

8. PR #52 (Selective): Windows Local Dev Launcher Scripts

• Added Windows helper scripts for local development startup:
  • scripts/run-claw-empire-dev-local.cmd
  • scripts/kill-claw-empire-dev.ps1
• Adjusted the PowerShell cleanup logic so it matches processes by the current repository root instead of a hardcoded path.
• Result: Windows local launcher cleanup is portable across clones instead of only working for one machine-specific directory.

Files Updated in This Release

• docs/api.md
• docs/openapi.json
• docs/releases/v2.0.2.md
• scripts/kill-claw-empire-dev.ps1
• scripts/openapi-contract.mjs
• scripts/run-claw-empire-dev-local.cmd
• src/components/settings/ApiSettingsTab.test.tsx
• src/components/settings/ApiSettingsTab.tsx
• src/components/settings/GatewaySettingsTab.characterization.test.tsx
• tests/e2e/ci-api-ops-and-docs.spec.ts
• tests/e2e/ci-docs-and-ops.spec.ts
• tests/e2e/ci-public-api-surface.spec.ts
• package.json

Compatibility Notes

• No schema migration is required for v2.0.2.
• No user-facing workflow changes are required to adopt this release.
• This release is focused on release-gate accuracy, API discoverability, regression prevention on main, and small post-release UX/runtime follow-ups from PRs #49, #52, and #55.

Claw-Empire v2.0.1 Release Notes

• Release date: 2026年03月03日
• Scope: Discord direct-messenger bidirectional reliability update plus office-pack hydration-aware department precedence and first-install bootstrap safety.

Additional Update (2026年03月03日)

0. Office Pack Department Precedence Is Now Hydration-Aware

• For non-development office packs, department merge precedence now depends on hydration state:
  • pre-hydration: pack profile labels/icons are shown first
  • post-hydration: DB metadata becomes source of truth
• This keeps first-install users from losing pack-profile defaults before DB hydration is ready.

0.1 UI Wiring Updated for Deterministic Behavior

• AppMainLayout now computes isHydratedOfficePack before department display resolution.
• resolvePackDepartmentsForDisplay receives preferPackProfile: !isHydratedOfficePack to make the transition explicit and deterministic.

0.2 Regression Coverage Added for Merge Semantics

• Added test coverage for:
  • pre-hydration pack-profile priority
  • post-hydration DB priority
  • foreign-pack seed-agent filtering in non-development packs

Highlights

1. Discord Channel Auto-Discovery by Token

• Added POST /api/messenger/discord/channels to resolve Discord channels directly from a Bot token.
• Token normalization now accepts both raw token and Bot <token> input formats.
• Discovery flow reads:
  • /users/@me/guilds (guild list)
  • /guilds/:id/channels (channel list per guild)
• Channel candidates are filtered to text-capable Discord types (0, 5, 10, 11, 12), deduplicated, and sorted by guild/name for consistent UI rendering.
• Error mapping is now explicit:
  • discord_token_required (400)
  • discord_auth_failed (401/403)
  • discord_rate_limited (429)
  • discord_channel_lookup_failed (502)

2. Settings UX Upgrade for Discord Session Setup

• In Messenger Settings > Add/Edit Chat, Discord token input now auto-triggers channel lookup with debounce.
• Added auto-detected Discord channel selector:
  • selectable option format: Guild / #channel (channelId)
  • selected option auto-fills targetId and default display name when empty.
• Added multilingual guidance for lookup states (KO/EN/JA/ZH):
  • loading indicator
  • auto-loaded channel count
  • no-discoverable-channel hint (permission/server membership guidance)
  • detailed auth/rate-limit/failure error messages

3. Discord Receiver Added (Inbound -> Inbox Forward)

• Added Discord polling receiver runtime (server/messenger/discord-receiver.ts) and wired it into lifecycle start/stop.
• Receiver resolves enabled Discord session routes from settings and polls channel messages with per-route cursors.
• Human text messages are forwarded into /api/inbox with route context:
  • source: discord or token-hinted source when multi-token
  • chat: channel:<channelId>
  • author: Discord display/user identity normalization
• Bot-authored and empty-content messages are skipped to avoid self-echo loops/noise.
• Cursor state persists under discordReceiverCursor, preventing duplicate forwards across restarts.

4. Runtime Receiver Observability

• Added GET /api/messenger/receiver/discord for receiver health/status inspection.
• Settings runtime panel now displays Discord receiver state alongside Telegram:
  • active/inactive flag
  • polled route count
  • last error surface for fast diagnosis
• Result: one-way outbound test success can now be differentiated from inbound receive readiness directly in UI.

5. Multi-Token Routing Safety for Discord Inbound Flow

• Receiver route keys now combine token-hash key + channel ID (<tokenKey>:<channelId>) for cursor isolation.
• When multiple Discord tokens share the same target channel ID, inbound forwarding uses token-hinted source values to preserve session ownership.
• This mirrors existing messenger token-hint route disambiguation strategy and prevents cross-token mixing.

6. Test Coverage Expansion

• Added gateway tests for Discord channel lookup:
  • successful guild/channel aggregation path
  • auth failure behavior and error surfacing
• Added Discord receiver tests for:
  • inbox forwarding + cursor persistence
  • bot-message skip behavior
  • multi-token token-hint source separation
• Added Settings characterization test:
  • token input triggers Discord auto-discovery
  • loaded channel options are rendered and selectable

API and Behavior Notes

• New API endpoints:
  • POST /api/messenger/discord/channels
  • GET /api/messenger/receiver/discord
• No schema migration is required for this release.
• Existing outbound messenger send flow remains compatible.
• If Discord token/session or INBOX_WEBHOOK_SECRET is missing, receiver remains non-forwarding and exposes explicit status error text.

Compatibility Notes

• Backward compatible with existing messengerChannels structure.
• Existing Telegram receiver behavior is unchanged.
• Existing Discord outbound message sending continues to work; v2.0.1 adds inbound routing and setup ergonomics on top.

Claw-Empire v2.0.0 Release Notes

• Release date: 2026年02月28日
• Scope: Workflow Pack / Office Pack operational rollout and direct messenger routing stability update, covering pack-based office profiles, agent/department sync, multi-token messenger isolation, and inbox/session handling hardening.

Highlights

1. Workflow Pack Platform Rollout

• Added first-class workflow pack definitions and runtime APIs for pack-aware orchestration.
• Built-in pack keys now include:
  • development
  • report
  • web_research_report
  • novel
  • video_preprod
  • roleplay
• Added /api/workflow-packs route and schema/runtime support for pack metadata and enabled-state management.
• Messenger session settings now support per-session workflowPackKey, enabling channel-session based task style routing.

2. Office Pack UX and Real Data Isolation

• Moved Office Pack selector to the top header for faster context switching while keeping office canvas focus intact.
• Added pack-specific office presets (room themes, department naming, staff naming/persona seeds) with multilingual labels.
• Implemented pack profile isolation for non-development packs so each pack keeps independent agent/department datasets.
• Agent/Department management now applies CRUD/reorder changes against the selected pack profile in isolated packs, while development pack keeps existing DB-backed behavior.
• Fixed office avatar rendering consistency so non-development pack characters use proper sprite avatars instead of emoji fallbacks.

Office Pack Profiles Added in v2.0.0

• development (DEV): default engineering baseline profile (backward-compatible DB behavior).
• report (RPT): structured report production profile with Editorial Planning, Research Engine, Doc Design, and Review Desk.
• web_research_report (WEB): evidence-first web investigation profile with Research Strategy, Crawler Team, and Fact Check workflow.
• novel (NOV): narrative production profile tuned for worldbuilding, character consistency, and tone QA.
• video_preprod (VID): video pre-production profile focused on concept/script/shot-list/editing-note handoff quality.
• roleplay (RPG): character roleplay profile optimized for in-character dialogue immersion and character QA/safety consistency.

3. Messenger Multi-Token Route Isolation

• Hardened route resolution for same-provider multi-session setups where targetId may overlap across different bot tokens.
• Added token-hint route keying (<channel>#<tokenKey>) for internal source disambiguation without exposing raw tokens.
• Session route matching now validates both targetId and token-key context when present.
• Telegram receiver now supports per-token route processing and per-token offset persistence to avoid cross-token poll collisions.
• Result: when multiple Telegram bots point to the same chat/channel ID, replies are routed back to the correct originating bot/session.

4. Inbox Session Reset and Project Flow Reliability

• Added /new messenger command handling to reset direct-chat session binding state and start a fresh conversation thread.
• Reset ACK messages are localized (KO/EN/JA/ZH) and delivered on the same route.
• Improved existing/new project guidance flow with stronger multilingual intent fallback and recent-project selection stability.
• Existing-project candidate output now supports concise numbered selection patterns with clearer latest-project emphasis.

5. Decision Notice Readability v2

• Decision request formatting was further compacted for messenger readability:
  • short planning-lead summary block
  • concise option previews
  • clearer numeric single/multi-reply guidance
• Added recommendation extraction/output refinements so planning-lead recommendation is surfaced as a cleaner numeric suggestion line.
• Preserved multilingual parsing and numeric reply compatibility (1, 1,3, 1 3) with route-aware ACK delivery.

6. Stability and Test Coverage Additions

• Added/expanded regression coverage for:
  • messenger client session routing behavior
  • messenger token-aware route resolution
  • Telegram receiver multi-token handling
  • office pack normalization and sync planning logic
• Runtime typing and route dependency wiring were tightened around collab/inbox/session handling paths.

7. New Menu Updates (Workflow/Office Pack)

• video_preprod pack received production-level render orchestration updates:
  • Remotion-based real MP4 output guidance is now built into pack execution rules.
  • [VIDEO_FINAL_RENDER] subtask is seeded at planning stage and delegated only after non-render subtasks complete.
  • When review meeting routing is blocked, final render delegation is queued and resumed automatically.
  • Review flow now prevents duplicate final-render subtask creation and reconciles stale delegated render links.
• Delegation runtime for pack flows was hardened:
  • origin-team-first sequential delegation order is enforced.
  • workflow-pack scope is preserved when creating delegated tasks.
  • in-flight drain/re-entry race windows and render-delegation stale-state paths are guarded.
• Office Pack operational lifecycle was expanded for real usage:
  • first-entry pack hydration bootstraps pack-specific department/agent seeds into DB.
  • hydrated packs switch to DB-backed mode and skip re-hydration to preserve user customizations.
  • provider edits from Office tab now persist correctly for hydrated pack agents.
• Settings UX update for new pack operations:
  • toggle-based controls (auto-assign, YOLO mode, etc.) were reorganized for clearer two-column scanning.
• Report/PPT operation update inside pack-driven flow:
  • report output policy now targets both HTML and PPTX artifacts together.
  • python-pptx fallback is constrained to PPT Team unavailable/hard-fail cases with explicit fallback logging.
• Prompting/context quality update for new workflow menu:
  • video generation guidance was upgraded for higher output quality.
  • auto-generated worktree CLAUDE.md context scaffolding was enriched.

8. Fixes on Existing Flows (Non-New-Menu)

• Worktree lifecycle now recovers from branch name collisions by choosing safe suffixed branch/worktree candidates.
• Claude CLI spawn path now preserves --no-tools argv token semantics across shell execution.
• Decision YOLO/autopilot retry handling was hardened to avoid busy-loop style stalls on retryable failures.
• WebSocket unauthorized-close recovery was stabilized to reduce force-bootstrap retry loops.
• Messenger decision delivery/retry and progress-target inference paths were tightened for better continuity.

API and Behavior Notes

• New/expanded workflow-pack surface:
  • GET /api/workflow-packs
• Messenger session payloads include optional:
  • workflowPackKey
  • per-session token override
• Internal source route may include token hint suffix:
  • <channel>#<tokenKey> (for route disambiguation only)

Compatibility Notes

• Existing development office behavior remains backward compatible.
• Non-development packs use isolated profile data by design and do not overwrite the development office baseline.

Claw-Empire v1.2.3 Release Notes

• Release date: 2026年02月27日
• Scope: Direct messenger integration and decision-routing stabilization update, covering multi-channel session UX, channel-isolated relay behavior, in-messenger decision replies, and project-binding safety hardening.

Highlights

1. Unified Built-in Messenger Channels

• Added a shared messenger channel catalog used across runtime, routing, gateway, and UI layers.
• Built-in channels are now normalized to:
  • telegram
  • whatsapp
  • discord
  • googlechat
  • slack
  • signal
  • imessage
• Runtime channel configuration is persisted in SQLite (settings.messengerChannels) and consumed consistently by server gateway and UI.

2. Settings Tab UX Redesign (Chat Sessions)

• Replaced per-channel fragmented editing flow with a unified "Add Chat" / "Edit Chat" modal UX.
• Session create/edit/delete now supports channel, token, target, name, enabled flag, and mapped agent in one place.
• Confirm in modal immediately persists settings (no extra global save step for chat-session changes).
• Chat list now shows mapped agent avatar + name next to each session for faster operator context.
• Added explicit helper copy at the top of Channel Messaging tab to guide users on registering messenger/token/target ID/conversation agent.

3. Channel-Isolated Task Relay

• Task relay remains pinned to the origin route via [messenger-route] <channel>:<target>.
• Relay policy was expanded from completion-only reports to task broadcast messages:
  • report
  • chat
  • status_update
• Result: meeting kickoff/progress/completion messages are delivered only to the originating channel route, preventing cross-channel spread.
• Added in-memory route cache guardrails (TTL + max size) for route lookup stability.

4. Decision Inbox Messenger Bridge

• Decision requests are now delivered to the same messenger route tied to task/project context.
• Added persistent dedupe reservation/sent markers to prevent duplicate decision notices across retries/restarts.
• Users can answer decisions directly in messenger with numeric replies:
  • single: 1
  • multi-pick (review round): 1,3 or 1 3
• Decision reply parsing is multilingual and route-aware, then applied via the existing decision handlers.
• Reply ACK/ERR is sent back to the same messenger channel/target.
• Numeric multi-choice parsing was hardened so plain replies like 1,2,3,4 are treated as decision replies when a pending decision exists.
• If no pending decision exists, plain numeric chat remains regular agent conversation flow (no forced decision interception).

5. Decision Message Readability Improvements

• Decision request messages were reformatted for messenger readability:
  • compact summary text
  • concise option previews
  • plain numeric reply guidance
• Removed default heavy technical token noise (advanced token reply line) from normal operator view.
• Kept token-based explicit reply parsing supported internally for compatibility.

6. Direct Chat Workflow and Safety Hardening

• Direct-chat project flow enforces existing/new project selection before escalation.
• Improved multilingual intent/project-kind fallback to reduce looped prompts.
• New project path creation is restricted by allowed roots (PROJECT_PATH_ALLOWED_ROOTS) to harden path traversal risk.
• Added duplicate sentence normalization mitigation for cleaner outbound replies.
• Task-delegation/runtime typing tightened (db: any -> strict runtime DB types in deps).

7. Messenger Completion Report Readability Patch

• Long completion reports (table-heavy/large payload) are now transformed into compact messenger summaries:
  • title + top key results
  • progress summary line when available
  • pointer to full details in Claw-Empire chat
• The first report line is now generated with agent identity (avatar + name) instead of fixed generic completion wording.
• This keeps in-messenger readability high while preserving full report detail in the in-app chat thread.

API and Behavior Notes

• Direct messenger endpoints remain under /api/messenger/*.
• /api/inbox now includes decision-reply bridge handling before regular agent-chat reply scheduling when applicable.
• Messenger session payloads support per-session agentId mapping.
• .env messenger token/channel variables are no longer the primary config path; UI/DB configuration is the canonical runtime source.

Verification Summary

• Type check:
  • pnpm -s tsc -b --pretty false (pass)
• Targeted tests:
  • pnpm -s vitest --config server/vitest.config.ts run server/gateway/client.test.ts server/messenger/session-agent-routing.test.ts server/modules/routes/collab/direct-chat.normalize.test.ts (pass)

Compatibility Notes

• Slack typing indicator remains no-op due provider/API constraints.
• For stricter environments, set PROJECT_PATH_ALLOWED_ROOTS explicitly instead of relying on defaults.

Claw-Empire v1.2.2 Release Notes

• Release date: 2026年02月26日
• Scope: Post-v1.2.1 functional release introducing the interrupt-inject workflow (paused-session prompt injection), plus task-control security hardening and per-agent CLI model override controls from Office Agent Detail.

Highlights

Interrupt-Inject Feature Rollout

• New paused-session prompt injection endpoint: Added POST /api/tasks/:id/inject to queue sanitized interrupt prompts for paused runs.
• Session-proofed control path:
  • Added task interrupt control token validation (x-task-interrupt-token or request body interrupt_token).
  • Added CSRF validation gate for cookie-authenticated task mutations (stop, resume, inject).
• Queued injection persistence:
  • Added task_interrupt_injections flow helpers (sanitize/hash/queue/consume) and prompt hash audit logs.
  • Terminal API now returns interrupt proof payload (session_id, control_token, requires_csrf) to support safe UI resume/inject flows.
• TerminalPanel operational UX:
  • Added Interrupt/Inject panel with pause-only, inject+resume, and resume-only actions.
  • Added short retry strategy for pending proof retrieval and clear multilingual operator feedback messages.
• Verification coverage:
  • Added route and workflow unit tests for interrupt token handling, CSRF constraints, and injection queue behavior.
  • Added HTTP smoke helper script: scripts/qa/interrupt-inject-http-smoke.mjs.

Agent-Level CLI Model Overrides (Office)

• Schema/API extension for agents:
  • Added agents.cli_model and agents.cli_reasoning_level.
  • Extended /api/agents/:id patch validation and provider-aware cleanup rules.
• Agent Detail editor upgrade:
  • CLI providers (claude, codex, gemini, opencode) can now set per-agent main model overrides directly in Office.
  • Codex adds per-agent reasoning effort override (low/medium/high/xhigh) with default fallback support.
  • When set to default, runtime uses global Settings model config as before.
  • Sub-agent model selection remains driven by global Settings (providerModelConfig.subModel), not per-agent override.
• Runtime override propagation:
  • Applied to all major execution entry points:
    • task run
    • resume/start orchestration path
    • one-shot runner
    • direct spawn
    • delegated subtask batch
    • cross-department cooperation

Planning-Lead No-Tools Fast Path

• Team-lead meetings keep no-tools policy:
  • Planned approval and review consensus one-shot rounds execute with noTools: true for faster turn completion.
• Decision inbox planning consolidation now no-tools:
  • Project-level review planning consolidation (project_review_ready) runs with noTools: true.
  • Review-round planning consolidation (review_round_pick) runs with noTools: true.
• Final report consolidation now no-tools:
  • Planning leader's final consolidated archive generation runs with noTools: true to reduce tool-call overhead.

UI/Theme Readability Polish

• Interrupt action contrast fix:
  • Added theme-level danger tokens and updated interrupt button styling for better readability in light mode.

API/Docs Updates

• Updated API docs (docs/api.md) to include:
  • CSRF bootstrap/header expectations
  • Interrupt session proof requirements
  • New POST /api/tasks/:id/inject endpoint

Notable Files

• Interrupt control and injection:
  • server/modules/routes/core/tasks/execution-control.ts
  • server/modules/workflow/core/interrupt-injection-tools.ts
  • src/components/TerminalPanel.tsx
  • src/api/organization-projects.ts
• Agent-level CLI model overrides:
  • src/components/AgentDetail.tsx
  • server/modules/routes/core/agents/crud.ts
  • server/modules/routes/core/tasks/execution-run.ts
  • server/modules/workflow/orchestration/execution-start-task.ts
  • server/modules/workflow/core/one-shot-runner.ts
• Planning-lead no-tools policy coverage:
  • server/modules/workflow/orchestration/planned-approval.ts
  • server/modules/workflow/orchestration/meetings/review-consensus.ts
  • server/modules/routes/ops/messages/decision-inbox/project-review-planning.ts
  • server/modules/routes/ops/messages/decision-inbox/review-round-planning.ts
  • server/modules/workflow/orchestration/planning-archive-tools.ts
• Schema/type updates:
  • server/modules/bootstrap/schema/base-schema.ts
  • server/modules/bootstrap/schema/oauth-runtime.ts
  • src/types/index.ts
  • server/modules/routes/shared/types.ts

Verification Summary

• pnpm lint (passes; existing repository warnings unchanged)
• pnpm build (passes)
• pnpm test:api (passes)

Claw-Empire v1.2.1 Release Notes

• Release date: 2026年02月26日
• Scope: Post-v1.2.0 hardening and maintainability release focused on CI reliability, runtime safety, type debt reduction, and documentation/API discoverability.

Highlights

Core Hotfixes

• Task route split recovery: Fixed production build/runtime breakage by restoring missing core task route registrations after modular split (crud, execution, subtasks) in the main route wiring.
• Workflow Unicode guard: Added CI protection against hidden/bidi Unicode characters in .github/workflows/*.yml to mitigate Trojan Source-style risks.

CI and Toolchain Hardening

• pnpm version alignment: Removed conflicting pnpm version declarations between workflow and packageManager, resolving ERR_PNPM_BAD_PM_VERSION class failures.
• Trigger coverage update: CI now runs on push to main, pull_request, and manual dispatch.
• Least privilege: Explicit permissions: { contents: read } applied to CI workflow.
• Explicit build gates:
  • pnpm exec tsc -p tsconfig.json --noEmit
  • pnpm run build
• Formatting enforcement reliability: Repository-wide Prettier normalization to satisfy format:check deterministically in CI.

Lint Baseline and Gradual Tightening

• ESLint Flat Config rollout with TypeScript and React Hooks coverage in CI.
• Safe warning cleanup pass across runtime/orchestration hotspots (empty blocks, useless escapes where safely removable, and related non-behavioral lint debt).
• Incremental strictness path introduced via lint-staged for staged-file quality gates (targeted tightening without forcing full-repo churn).

RuntimeContext and Type-Debt Refactor

• Dependency-surface reduction across route modules to reduce giant runtime context coupling and improve reviewability.
• Decision Inbox type unification:
  • Removed duplicated DecisionInboxRouteItem declaration.
  • Consolidated shared type usage through decision-inbox/types.ts.
• App orchestration extraction:
  • Added src/app/useAppBootstrapData.ts for initial bootstrap fetch/sync logic.
  • Added src/app/useLiveSyncScheduler.ts for live sync scheduling and dedup logic.
  • Reduced logic density in src/App.tsx without changing user-facing behavior.

Meeting Timeout Reliability and Clarity

• Explicit timeout unit default:
  • Standardized REVIEW_MEETING_ONESHOT_TIMEOUT_MS default to 65000 (milliseconds) in docs/migration defaults to avoid unit ambiguity.
• Backward compatibility:
  • Existing values like 65 are still accepted and interpreted as seconds (<= 600 -> seconds), preserving prior user setups.
• Timeout resilience:
  • Added one-shot meeting timeout retry path with compacted prompts for review/approval phases to reduce interruption risk during longer planning discussions.

Test Runtime Safety

• Test DB/runtime isolation hardening:
  • Reinforced separation of test runtime data from production runtime paths.
  • Added setup-level safety posture for future integration tests to prevent accidental production DB usage.

API and Security Documentation Surface

• Swagger/OpenAPI route exposure: Added Swagger documentation endpoint with linked OpenAPI spec (docs/openapi.json) for contract visibility.
• OpenAPI contract automation:
  • Added openapi:sync and openapi:check workflows to keep spec drift visible and enforceable.
  • CI now validates OpenAPI structure and fails when docs/openapi.json is out of sync.
  • Added baseline request/response examples and standardized error response coverage (401/403/409/429/500) across documented operations.
• Documentation refresh:
  • Updated multilingual READMEs (README.md, README_ko.md, README_jp.md, README_zh.md).
  • Added CI status badges and current CI validation flow.
  • Added clear links to API docs and SECURITY.md.

Notable Files and Areas

• CI / workflow:
  • .github/workflows/ci.yml
• Quality/tooling:
  • eslint.config.mjs
  • package.json
  • pnpm-lock.yaml
  • scripts/openapi-contract.mjs
• Runtime/routes refactor and type cleanup:
  • server/modules/routes/core.ts
  • server/modules/routes/ops/messages/decision-inbox-routes.ts
  • server/modules/routes/ops/messages/decision-inbox/types.ts
  • server/modules/routes/** (dependency-surface narrowing series)
• App orchestration split:
  • src/App.tsx
  • src/app/useAppBootstrapData.ts
  • src/app/useLiveSyncScheduler.ts
• Docs/API/security:
  • docs/api.md
  • docs/openapi.json
  • SECURITY.md
  • README*.md
  • .env.example
  • scripts/auto-apply-v1.0.5.mjs
• Meeting orchestration:
  • server/modules/workflow/orchestration.ts
  • server/modules/workflow/orchestration/planned-approval.ts
  • server/modules/workflow/orchestration/meetings/review-consensus.ts

Verification Summary

• Local quality/build checks executed during stabilization:
  • pnpm run lint (passes with warnings policy)
  • pnpm run build (passes)
• CI workflow now includes explicit format, lint, typecheck, build, and test gates with pnpm-version consistency and minimum permissions.

Claw-Empire v1.2.0 Release Notes

• Release date: 2026年02月25日
• Scope: Agent management (hire/edit/delete), department CRUD, department management tab, new pixel character sprite (#13), project-level manual agent assignment, meeting participant filtering for manual projects, mobile-responsive Project Manager, custom skill upload system, bug fixes, plus post-release hardening and engineering-quality updates merged into v1.2.0.

Highlights

Agent Management (CRUD)

• Hire New Agent - Full agent creation UI in Agent Manager with name (multilingual: en/ko/ja/zh), department, role (team_leader/senior/junior/intern), CLI provider, sprite number, avatar emoji, and personality fields.
• Edit Agent - Inline edit for all agent fields including provider/model reassignment.
• Delete Agent - Safe deletion with working-status guard; cascading NULL-ification on tasks, subtasks, meeting minutes, report archives, and review decision states.
• Backend: POST /api/agents, PATCH /api/agents/:id, DELETE /api/agents/:id endpoints with WebSocket broadcast (agent_created, agent_deleted).

Department CRUD

• Create Department - Full department creation with ID validation (a-z0-9 pattern), multilingual names, icon, color picker, description, and prompt.
• Edit Department - Inline editing of all department properties.
• Delete Department - Removal with associated agent/task guards.
• Backend: POST /api/departments, PATCH /api/departments/:id, DELETE /api/departments/:id endpoints with departments_changed broadcast.

Department Management Tab

• Added subtab toggle (Agents | Departments) inside Agent Manager.
• Department list view with sort order, icon, name, color, agent count, description, and ID.
• Sort order editing via arrow buttons and drag-and-drop, with batch save.
• Backend: PATCH /api/departments/reorder endpoint using temporary high-value sort_order to avoid UNIQUE constraint conflicts during reorder.
• Routing guard added so /api/departments/reorder is not captured by /api/departments/:id.

New Character Sprite (Sprite #13)

• Added a new pixel character with full directional sprite set (D-1/D-2/D-3, L-1, R-1).
• Added a sprite generation pipeline script.
• AgentAvatar and OfficeView updated for sprite #13 support.

Project Manual Agent Assignment

• Projects now support assignment_mode: auto (default) or manual.
• In manual mode, specific agents can be assigned via multi-select checkbox UI with sprite avatar icons and department/role labels.
• New DB schema: assignment_mode column on projects table + project_agents junction table.
• Backend CRUD: POST /api/projects and PATCH /api/projects/:id accept assignment_mode and agent_ids.
• GET /api/projects and GET /api/projects/:id return assigned_agent_ids per project.

Meeting Participant Filtering (Manual Mode)

• When a project uses assignment_mode: manual, kickoff and review meetings only include:
  • Planning team leader (always included)
  • Team leaders from departments of the assigned agents
• fallbackAll expansion (invite all active team leaders) is skipped for manual-mode projects.
• Applies to both kickoff (startPlannedApprovalMeeting) and review (startReviewConsensusMeeting) meetings.

Task Delegation with Manual Mode

• findBestSubordinate in collab.ts accepts optional candidateAgentIds.
• Manual mode queries project_agents and restricts candidates to the assigned agent pool.
• Candidate filtering is additionally constrained to the active team leader's department.
• If no assigned candidate exists for that department in manual mode, fallback to full-department auto selection is blocked.
• Auto mode projects retain existing delegation logic unchanged.

ChatPanel Manual Mode Indicator

• Project confirm step displays assigned agent count for manual-mode projects (for example: "Manual mode - 2 assigned agents will work on this").

Mobile-Responsive Project Manager

• List-detail toggle pattern for mobile viewports (below md / 768px):
  • Project list takes full width when no project is selected.
  • Selecting a project or creating new hides the list and shows detail view.
  • Mobile-only back button returns to project list.
• Desktop layout remains side-by-side.

Bug Fixes

• Fixed project save 500 caused by unresolved runInTransaction binding in core.ts.
• Improved light-mode error readability in ProjectManagerModal (text-rose-800/text-cyan-800 variants).
• Fixed /api/departments/reorder save failure caused by route collision with /api/departments/:id.
• Added internal scroll behavior for Agent Manager emoji picker (max-h + overflow-y-auto).
• Added internal scroll behavior for Agent Manager hire/edit modal (max-h + overflow-y-auto).

Custom Skill Upload System

• Custom skill CRUD — Users can upload .md skill files directly through the Skills Library UI, name the skill, select CLI representatives to train, and delete existing custom skills.
• Classroom training animation — Uploading a custom skill triggers a full-screen chalkboard classroom animation showing the claw-empire mascot as teacher and CLI representative agents as students.
• Backend endpoints: POST /api/skills/custom (upload), GET /api/skills/custom (list), DELETE /api/skills/custom/:skillName (delete) with file-based storage under custom-skills/ directory and skill_learning_history DB records.
• Frontend API client: uploadCustomSkill(), getCustomSkills(), deleteCustomSkill() added to src/api.ts.
• Light mode support: Full light-mode CSS overrides for custom skill buttons, modal, list cards, and classroom animation overlay.

Post-release Hardening (still v1.2.0)

• Manual assignment pre-save safeguard modal: Project Manager now warns when manual assignment is saved with no selected agents or team-leader-only selections, showing selection summary and requiring explicit confirmation.
• Manual selection visibility: Manual assignment UI now shows live counts (total, leaders, subordinates) to prevent accidental subordinate omission.
• Project API agent_ids validation: POST /api/projects and PATCH /api/projects/:id now validate agent_ids type and existence; invalid IDs are rejected with explicit error payloads.
• Delegation fallback auditability: In manual mode, when no eligible subordinate exists in assigned candidates, system logs explicit fallback context and sends a CEO-facing safeguard notice before team-leader direct execution.
• Sprite registration conflict guard: POST /api/sprites/register now returns 409 sprite_number_exists if target sprite files already exist, preventing accidental overwrite.
• Portable sprite generation script: Sprite generation now resolves output path relative to repository and auto-creates public/sprites, removing machine-specific absolute path dependency.
• Department sort_order UNIQUE index safe migration: server-main.ts now drops and re-creates the idx_departments_sort_order UNIQUE index around sort_order seed updates to avoid constraint violations during migration.

Engineering Quality Pack (still v1.2.0)

• Codebase modularization (frontend + backend):
  • Backend route/workflow decomposition moved large route domains into feature folders under server/modules/routes/* and server/modules/workflow/* (for example: core/agents, core/projects, core/tasks, ops/messages, ops/oauth, ops/skills, ops/task-reports, ops/terminal, collab/coordination).
  • Frontend decomposition moved oversized UI surfaces into feature directories under src/components/* (for example: office-view, taskboard, settings, chat, project-manager, agent-manager) with main entry modules plus focused submodules.
  • Goal: keep behavior equivalent while reducing single-file complexity for reviewability, targeted testing, and safer incremental changes.
• Type-debt cleanup (@ts-nocheck removal):
  • Removed server-side @ts-nocheck usage in runtime/security-related modules and replaced broad escapes with explicit shared/runtime type boundaries.
  • Refactoring was done alongside helper extraction to preserve pre-existing API behavior and response contracts.
• Repository-wide formatting standardization:
  • Added Prettier baseline (.prettierrc.json, .prettierignore) and package scripts (format, format:check).
  • Wired CI to run pnpm run format:check so formatting regressions fail fast before merge.
  • Result: consistent diffs and easier multi-file review after large modular refactors.
• Test code expansion (E2E + unit):
  • New CI E2E gap suite: tests/e2e/ci-coverage-gap.spec.ts.
  • Expanded coverage areas include full task lifecycle, Department/Agent/Project CRUD smoke flows, chat/directive path, settings + stats load, decision inbox path, report/terminal flows, and WebSocket event reception.
  • Frontend unit tests expanded with src/api.test.ts, src/hooks/useWebSocket.test.tsx, src/hooks/usePolling.test.tsx, and src/i18n.test.ts.
  • Backend unit tests expanded with server/security/auth.test.ts, server/ws/hub.test.ts, server/db/runtime.test.ts, and server/gateway/client.test.ts.
• CI E2E stabilization:
  • Added transient 502/503 retry handling in polling paths used by E2E coverage flows.
  • Set deterministic Playwright CI defaults (workers: 1, reuseExistingServer: false) to reduce flaky outcomes.

Changed Files

Frontend

• src/types/index.ts - Added AssignmentMode and extended Project.
• src/api.ts - Added reorderDepartments(), extended project CRUD fields, custom skill API functions (uploadCustomSkill, getCustomSkills, deleteCustomSkill).
• src/api.test.ts - Request format, retry/auth, idempotency-post, and error-shape coverage.
• src/App.tsx - Agent/department live-sync updates.
• `src/components/AgentManag...

Claw-Empire v1.1.9 Release Notes

• Release date: 2026年02月23日
• Scope: Post-v1.1.8 reliability fixes for sub-agent status sync, delegated review-gate pause/resume behavior, and Decision Inbox review-round skip routing.

Highlights

• Sub-agent Status Sync Hardening

  • Unknown or out-of-order agent_status payloads now trigger canonical live sync instead of appending unknown agents directly.
  • Added bounded Codex thread-binding guardrails (TTL + max-size pruning) to prevent stale thread mapping growth.

• Codex Thread Binding Cleanup on Completion

  • When a sub-agent is marked done, related Codex thread bindings are removed immediately.
  • Prevents delayed stream fragments from incorrectly finalizing unrelated/stale sub-agent entries.

• Delegated Pause/Resume Review-Gate Hotfix

  • Paused delegated runs no longer finalize linked subtasks as blocked on graceful interrupt exits.
  • Linked delegated subtasks are now reconciled after run completion (done on success, blocked on real failure).

• Review Completion Auto-heal for Stale Delegations

  • During review completion, stale blocked delegated subtasks are auto-recovered when their delegated task already reached review/done.
  • Parent review completion is retried automatically once all remaining linked subtasks are resolved.

• Decision Inbox Round-Skip Routing Fix

  • Fixed runtime wiring for scheduleNextReviewRound used by review_round_pick -> skip_to_next_round.
  • Prevents false reply errors and unintended fallback into project-level decision/team-lead meeting flow.
  • Added rollback guard: if next-round scheduling fails, the meeting state is reverted to revision_requested instead of staying completed.

Changed Files

Frontend

• src/App.tsx

Backend

• server/modules/routes/collab.ts
• server/modules/routes/ops/messages.ts
• server/modules/runtime-helper-keys.ts
• server/modules/workflow.ts
• server/modules/workflow/orchestration.ts
• server/types/runtime-context.ts

Docs

• README.md
• README_ko.md
• README_jp.md
• README_zh.md
• docs/releases/README.md
• docs/releases/v1.1.9.md

Verification

• pnpm run test:api
  • vitest --config server/vitest.config.ts passed (29 tests)