Skip to content

Latest commit

 

History

History
768 lines (622 loc) · 48.4 KB

CHANGELOG.md

File metadata and controls

768 lines (622 loc) · 48.4 KB

Changelog

All notable changes to this project will be documented in this file.

[0.31.0] - 2026-03-15

Added

  • Cross-platform contact identity mapping — map agent identities across Discord, Telegram, Slack, and AlgoChat with unified contact resolution (#1113)
  • User response feedback tied to reputation scoring — users can rate agent responses, feeding into the reputation system for trust-aware routing (#1110)
  • AlgoChat worktree isolation and smart branch cleanup — AlgoChat sessions now use isolated git worktrees with automatic stale branch cleanup (#1115)
  • Flock Directory automated testing framework — structured test harness for validating Flock Directory agent discovery and heartbeat flows (#1108)
  • Session metrics tracking and analytics endpoints — track token usage, tool calls, and duration per session with new analytics API (#1107)
  • CLI per-command --help output — every CLI command now supports --help with usage, options, and examples (#1116)
  • Context exhausted error handling — emit context_exhausted error type on context reset with differentiated Discord error messages (#1124)
  • Discord archive thread button — replace "New Session" with "Archive Thread" button; rename "Resume" to "Continue" (#1124)

Fixed

  • Discord duplicate error message dedup — prevent duplicate error messages with sentErrorMessage flag (#1124)
  • Session metrics persistence on error — metrics are now saved even when sessions terminate with errors or aborts (#1109)
  • Migration retry on failure — reset cached initDb promise when migrations fail, allowing retry without restart (#1106)

Changed

  • Missing export specs documented for infra and response-feedback modules (#1112)
  • Expanded test coverage for feedback routes, reputation scorer (#1114), and validation edge cases (#1117)
  • Fixed stale README badges and references (#1111)

[0.30.0] - 2026-03-15

Added

  • Multi-tool chain continuation — limited-tier models (Haiku/Sonnet) can now chain multiple tool calls across continuation rounds, enabling complex multi-step workflows (#1097, #1018)
  • Session stats in Discord — completion embeds now show token usage, tool call count, and duration for finished sessions (#1101)

Fixed

  • Session view race condition — subscribe to WebSocket before HTTP fetch in Angular session view to prevent missed updates (#1099)
  • Disabled agent filtering — listAgents and getAlgochatEnabledAgents now filter out disabled agents by default (#1100)
  • Agent display columns — hotfix migration 088 to add missing display customization columns (#1104)

[0.29.0] - 2026-03-15

Added

  • MCP delegation tools — delegate subtasks to specialist models with corvid_delegate_task and corvid_dispatch_model; automatic tier routing (Opus/Sonnet/Haiku) based on task complexity (#1082)
  • Flock Directory search & sorting — sort agents by reputation, name, or last-seen; aggregate reputation scores across the directory (#1077)
  • Welcome wizard templates — agent template selection during onboarding (Code Reviewer, DevOps, Researcher, etc.) for faster first-agent setup (#1076)

Fixed

  • Stale session reaping — reap orphaned sessions on startup to prevent ghost processes after restart
  • Duplicate Discord messages — unsubscribe event callbacks on session teardown to prevent duplicate message delivery (#1092)
  • Discord channel affinity — enforce channel affinity so Discord-originated sessions reply on the same channel (#1079)
  • Messaging safety — prevent agents from generating scripts that send messages outside MCP tools (#1086)
  • CLI login/logout — wire up login and logout commands in CLI dispatcher (#1083)
  • CLI chat polling — replace polling loop with direct callback for lower latency (#1084)
  • Agent-to-agent messaging — stop orphaned sessions and handle session_stopped events in A2A messaging (#1081)
  • Spec scaffold TOCTOU — eliminate race condition in spec scaffold generation (#1072)
  • CLI empty response — handle empty response body gracefully in CLI client (#1066)

Security

  • CI checkout actions — align all checkout actions to v6.0.2 and enforce dependency audit in CI (#1062)

Tests

  • +150 tests — close coverage gaps in 6 under-tested modules (#1091); total now 6,982 across 293 files

Chores

  • Worktree cleanup — remove 10 stale git worktrees from .claude/worktrees/ (#1065)
  • Discord worktree isolation — add worktree isolation to Discord /session command (#1096)
  • README stats — fix stale stats and add missing API entries (#1063)

Stats

  • 6,982 unit tests across 293 files (18,918 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 138 module specs with automated validation (100% file coverage: 369/369)
  • 43 MCP tools, ~300 API endpoints, 44 route modules, 90 tables
  • 16 commits on main

[0.28.0] - 2026-03-14

Added

  • Spec coverage detection — scan server/ for .ts files not referenced in any spec's files: frontmatter, with per-module grouping (#1058)
  • Spec scaffold generation — --generate flag creates draft .spec.md files for uncovered modules from template (#1058)
  • Coverage reporting — --coverage flag shows full unspecced file report; summary always shows file coverage percentage (#1058)
  • 100% spec coverage — all 368 server files covered by 137 module specs with CI enforcement via --require-coverage 100
  • Spec coverage badge — shields.io badge in README showing 100% spec coverage
  • Convenience scripts — bun run spec:coverage, bun run spec:generate, and bun run spec:coverage:require shortcuts (#1058)
  • Tiered Claude model dispatch — Opus/Sonnet/Haiku routing based on task complexity with fallback chains (#1052)
  • Ollama feature flag — gate Ollama behind ENABLE_OLLAMA flag per council decision; provider abstraction preserved (#1052)
  • Discord slash commands — /tasks, /schedule, /config commands for server interaction (#1025)
  • Dashboard UI polish — duration display, empty states, skeleton loading, mobile responsiveness (#1023, #1024, #1026, #1027)
  • Council list enhancements — search, sort, pagination + schedule execution stats (#1024)
  • Test data purge utility — admin endpoint for purging test data (#1017)
  • In-memory test DB — use in-memory SQLite for test runs to prevent production pollution (#1016)

Changed

  • CLI utilities deduplicated — extracted shared helpers (truncate, formatUptime, resolveProjectFromCwd, handleError) into cli/utils.ts (#1055)
  • Cross-platform path resolution — use path.sep for Windows compatibility in resolveProjectFromCwd

Fixed

  • CLI streaming — fix streaming display and WebSocket keepalive (#1033)
  • Discord gateway intents — correct privileged intent configuration (#1033)
  • Sidebar scrolling — fix sidebar not scrollable on desktop (#1015)
  • Unique project names — sync unique name index across schema layers (#1008)
  • Insecure temp file — fix code scanning alert #307 (#1010)

Documentation

  • Contributor welcome — make project welcoming to new contributors (#1040)

Chores

  • Repo hygiene — unique project names, doc updates, CLI fixes (#1028)
  • Dead code removal — remove dead execMarketplaceTrialExpiry handler (#1009)

[0.27.0] - 2026-03-13

Added

  • Cloud model families & tier boosting — support cloud model family grouping and tier-based model selection (#1005)
  • Auto-clone projects — automatically clone projects to temp/worktree directories on demand (#1004)
  • Ollama exam persistence — persist model exam results to SQLite for tracking cloud model capabilities (#999)

Fixed

  • Discord typing timeout — reduce false typing timeout warnings (#1002)

Documentation

  • API reference expansion — add detailed API reference for 13 undocumented modules (#1000)

Tests

  • +3 test suites — coverage for DbPool, OwnerQuestionManager, and expanded ReputationScorer tests (#1003)

[0.26.0] - 2026-03-13

Added

  • Agent security hardening — tier-based agent permissions (untrusted/standard/trusted/admin), per-agent session limits, and input sanitization (#986)
  • RBAC role templates — pre-built role templates for agent permission provisioning (#979)
  • Typed WebSocket messages — enforce typed ServerMessage emission in WS broadcasting (#957, #972)
  • Git worktree session isolation — isolate chat sessions with dedicated git worktrees, including Discord /session command (#983, #1096)
  • Flock Directory heartbeat — periodic heartbeat and stale sweep for on-chain agent directory (#903, #961)
  • RC checklist expansion — 9 additional gating criteria checks for v1.0.0-rc (#310, #977)

Security

  • Ephemeral HMAC key — replace hardcoded HMAC fallback with ephemeral random key (#982)
  • Plaintext wallet key removal — eliminate plaintext wallet key escape hatch entirely (#924, #973)
  • Branch protection — enable branch protection on main branch (#966)
  • CORS production warning — repo-blocklist tenant scoping and CORS hardening (#963)
  • Permissions guard — admin role guard for /api/permissions routes (#962)
  • SECURITY.md formatting — fix formatting issues in security documentation (#976)

Fixed

  • Discord typing indicator — fix typing indicator liveness checks (#995)
  • Discord permission spam — stop spamming permission denials in monitored channels (#980)
  • Discord role mentions — respond to role mentions, not just direct bot mentions (#964)
  • Duplicate work tasks — prevent duplicate work tasks and PRs for the same issue (#974, #978)
  • Injection false positives — skip prompt-injection false positives in markdown code spans (#960)
  • SQLite transactions — convert remaining DEFERRED transactions to BEGIN IMMEDIATE (#959)
  • Key rotation mock — fix mock readKeystore in key rotation test (#970)
  • Silent catches — add debug logging to silent fire-and-forget catch handlers (#975)

Tests

  • +209 new tests — expanded coverage for 5 untested modules (+67), worktree isolation, role templates, work task dedup, tenant route tests (#971, #978, #979, #983, #994)

Documentation

  • API module docs — add 8 undocumented API modules and fix stale refs (#993)
  • README + API sync — sync README and API reference with current codebase (#968)
  • TaskQueueService spec — document exports for TaskQueueService module (#969)
  • Module specs — add specs for bash-security, code-scanner, fetch-detector, github-searcher (#997)

Stats

  • 6,655 unit tests across 278 files (18,335 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 127 module specs with automated validation
  • 41 MCP tools, ~300 API endpoints, 44 route modules, 90 tables
  • 27 commits on main

[0.25.4] - 2026-03-11

Added

  • TaskQueueService — parallel work task dispatch with configurable concurrency (#951)
  • Flock Directory specs — module specs for on-chain client and service (#952)
  • Discord autocomplete — agent and project choice autocomplete for Discord commands (#954)

Fixed

  • Silent catches — add debug logging to silent catch blocks in Discord thread-manager (#955)

Security

  • Wallet validation — validate Algorand address format on all wallet routes (#953)

[0.25.3] - 2026-03-11

Fixed

  • Install script — read user input from /dev/tty so prompts work when piped via curl (#944, #949)

[0.25.2] - 2026-03-11

Fixed

  • Discord typing indicator — add safety timeout to prevent typing indicator interval leaks (#947)
  • GitHub mention allowlist — bypass allowlist for assignment-type GitHub mentions (#946)
  • Process exit errors — pass error details through process exit to session messages (#945)
  • Discord mentions — resolve Discord mentions to @username in message text (#942)

Chore

  • Test coverage — coverage expansion for response, builtin middleware, key rotation (#943)

[0.25.1] - 2026-03-11

Fixed

  • Discord typing indicator — keep typing indicator alive during AI warm-up with continuous 8-second interval refresh (#940)

[0.25.0] - 2026-03-11

Security

  • KMS migration enforcement — encrypted in-memory key cache, startup enforcement requiring KMS migration, key access audit logging with 18 new tests (#931)
  • Hono CVE override — update hono override to >=4.12.7 for CVE GHSA-v8w9-8mx6-g223 (#934)
  • Admin role guard — add admin role guard to repo-blocklist routes (#930)

Fixed

  • Ollama cloud model serialization — cloud models get maxWeight in the slot system to force serialization and prevent proxy timeouts (#937)
  • Hide admin Discord commands from non-admin users (#921)

Changed

  • Discord bridge decomposition — decompose discord/bridge.ts from 2,688 to 367 lines (#933)

Added

  • RC verification script and mainnet config template (#917)
  • Spec documentation for polling modules: auto-merge, auto-update, ci-retry (#936)

Chore

  • Align @opentelemetry/exporter-prometheus to 0.213.0 (#922)
  • Sync MCP tools, migration count, and directory structure in docs (#935)

[0.24.2] - 2026-03-10

Fixed

  • Hide admin-only Discord commands (/admin, /council, /mute, /unmute) from non-admin users via default_member_permissions (#921)

[0.24.1] - 2026-03-10

Fixed

  • Remove unused execSync and existsSync imports from bin/corvid-agent.mjs (#919)
  • Add CodeQL config to exclude auto-generated *.generated.ts files from static analysis (#919)

[0.24.0] - 2026-03-10

Added

  • Discord /admin commands — manage bot configuration directly from Discord using native mentions. Subcommands: channels add/remove/list (#channel mentions), users add/remove/list (@user mentions), roles set/remove/list (@role mentions with permission level dropdown), mode (chat/work_intake toggle), public (role-based access toggle), show (full config summary). All mutations audit-logged and persisted to discord_config table with 30s hot-reload
  • Discord /work command — fire-and-forget work task creation from Discord with rich embed confirmations, @mention notifications on completion/failure, and PR link delivery
  • AlgoChat /work improvements — --project flag for project targeting, clear status indicators, PR URL in completion messages
  • Project discovery MCP tools — corvid_list_projects and corvid_current_project tools for agent project awareness
  • RC verification script — scripts/verify-rc.sh for release candidate validation and mainnet config template

Changed

  • Discord bridge spec v8 — /admin command fully documented with subcommand groups, recursive DiscordInteractionOption type
  • AlgoChat commands spec v2 — --project flag, behavioral examples for project resolution

Tests

  • 20 new tests for /admin command handlers (channels, users, roles, mode, public, show)
  • Discord /work and AlgoChat /work spec coverage

Stats

  • 6,347 unit tests across 262 files (17,499 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 115 module specs with automated validation (0 warnings)
  • 41 MCP tools, ~205 API endpoints, 44 route modules, 90 tables
  • 4 commits, 2 PRs merged this release

[0.23.2] - 2026-03-10

Fixed

  • Persist Discord interacted users to DB across restarts (#914)

[0.23.1] - 2026-03-10

Fixed

  • Remove eyes reaction on Discord message receipt to reduce noise (#912)

[0.23.0] - 2026-03-10

Added

  • Discord onboarding — guided setup flow for new users with welcome embeds, server configuration wizard, and role assignment (#890, #910)
  • Discord dynamic configuration — DB-backed config for Discord settings (channel modes, auto-archive, rate limits) with hot-reload via settings API (#909)
  • Hybrid FlockDirectoryService — on-chain sync for Flock Directory with local cache fallback (#902, #907)

Fixed

  • Grant contents:write permission to docker-publish workflow for SBOM release asset attachment (#908)

Tests

  • Coverage for scheduler orchestration and marketplace subscriptions (#906)

Stats

  • 6,327 unit tests across 261 files (17,461 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 116 module specs with automated validation
  • 39 MCP tools, ~205 API endpoints, 44 route modules, 90 tables
  • 5 commits, 5 PRs merged this release

[0.22.0] - 2026-03-10

Added

  • Marketplace ecosystem — tiered pricing plans (#842), per-use credit billing (#800), verification badges and quality gates (#851), free trial periods (#873), usage metering and analytics (#854)
  • Flock Directory — on-chain agent registry with MCP tool and API (#806), ARC56 contract client (#901)
  • Governance v2 frontend — vote panel and governance service UI (#802), real-time WebSocket vote events (#846)
  • MCP expansion — standalone corvid-agent-mcp server package (#815), agent-agnostic MCP support for Cursor, Copilot, and OpenCode (#843), VibeKit smart contract integration (#839), skills-as-markdown for AI assistant discovery (#838)
  • Work task priority queue — preemption support for higher-priority tasks (#816)
  • Branch protection — enforce branch protection on main branch (#808)
  • Enhanced init — corvid-agent init with --mcp, --yes, and auto-clone (#837)
  • WebSocket shared type layer — shared types between server and client (#870)
  • Discord public channel mode — role-based access control, multi-channel support, smart message splitting, typing indicators, and stale thread auto-archiving (#899)

Security

  • Injection hardening — unicode bypass detection, API route scanning, and prompt leakage prevention (#875)
  • Rate-limit device auth — rate-limit device auth flow endpoints (#868)

Refactored

  • Extract MessageTransport interface for bridge swappability (#871)
  • Module boundary decomposition step 1: split shared/types.ts (#814)
  • Spec strict CI gate enforcement (#805)
  • Squash migrations into single baseline for faster fresh installs (#872)

Fixed

  • Keep direct-process sessions alive for multi-turn conversations (#900)
  • Refresh Discord slash commands on agent CRUD (#879)
  • Document undocumented exports in marketplace billing (#801) and connection spec (#867)
  • Log errors in broadcast listener callbacks instead of silently swallowing (#850)
  • CI: auto-regenerate bun.lock on Dependabot PRs (#847, #848, #849, #853, #856)

Documentation

  • System requirements and RAM benchmarks (#840)
  • Improved landing page messaging and onboarding (#836)
  • "Most tested AI agent platform" positioning (#844)
  • Blog page on GitHub Pages site (#876)
  • Expanded API reference with agent, session, schedule, and work-task details (#813)
  • Spec coverage for algochat/init and repo-blocklist handler (#818)
  • Sync stale stats across README and deep-dive (#885, #897)

Tests

  • Route integration tests for flock-directory endpoints (#888)
  • Route integration tests for governance proposals (#812)
  • Route tests for permissions, mention-polling, performance, and health (#820)
  • Protected-paths and shutdown-coordinator module specs (#811)

Stats

  • 6,215 unit tests across 255 files (17,166 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 116 module specs with automated validation
  • 39 MCP tools, ~205 API endpoints, 44 route modules, 88 tables
  • 57 commits, ~42 PRs merged this release

[0.21.0] - 2026-03-08

Added

  • Work pipeline v2 — parallel execution, task dependencies, and retry policies for work tasks (#793, #632)
  • Governance v2 — proposals with weighted voting, quorum rules, and proposal lifecycle (#789, #633)
  • Encryption at rest — encrypt env_vars using AES-256-GCM (#791)
  • Container sandboxing — integrate SandboxManager into ProcessManager with resource limits and network isolation (#786, #382)
  • KeyProvider abstraction — wallet encryption via pluggable KeyProvider interface (#772, #383)
  • Bridge delivery receipts — message delivery receipt tracking for bridge hardening (#780, #631)
  • Security Overview dashboard — new dashboard page showing live security posture (#779)
  • Stats automation — automated stats collection and drift detection (#792, #537)
  • GH_TOKEN scope validation — OAuth scope validation on server startup (#787)
  • Discord passive channel mode — threads via /session only, no unsolicited replies (#761)
  • Council Discord notifications — post council synthesis to Discord on completion (#766)
  • Real-time session status — live thinking/tool_use status with silent death prevention (#775)
  • Telegram bridge hardening — poll backoff and message deduplication (#762)

Security

  • Error message sanitization — sanitize error messages in WS and AlgoChat handlers to prevent info leakage (#763)
  • Auto-merge hardened — security scan now comments instead of closing PRs, preventing accidental PR destruction (#784)

Fixed

  • Document undocumented exports in security, infra, and marketplace specs (#790, #801)
  • Prevent duplicate Discord messages and normalize coding-tool paths (#788)
  • Relax protected files and stop auto-closing PRs (#784)
  • Remove duplicate properties in playwright config (#770)
  • Discord slash commands failing due to interaction token too short (#765)
  • Eliminate any types in subscription manager tests (#744)

Refactored

  • Extract Discord gateway into dedicated module (#769)
  • Extract scheduler service into handler modules (#759)

Tests

  • Wait-sessions, tenant middleware, and plugin permissions coverage (#773)

Documentation

  • API reference for workflows, councils, marketplace, reputation, and billing routes (#798)
  • TaskQueue design doc — prerequisite for work pipeline (#771)
  • Document all 17 remaining undocumented exports (#768)
  • Sync stale stats across README and doc files (#767)
  • Discord passive channel mode spec (#760)

Cross-Repo

  • corvid-agent-chat: toast, chat-messages, search coverage (+69 tests, #58); split chat.ts view (#57); device-name and wallet lifecycle coverage (+51, #56); wallet idle lock (#55); icon sizing fixes (#49, #52, #54)
  • specl: WelcomeComponent tests (#74); GitHubOAuthService and FrontmatterEditor coverage (+26, #73); wildcard route redirect (#72); missing spec.md files (#68)
  • corvid-reputation: standardize git author identity in generate workflow (#1)

Stats

  • 5,838 unit tests across 237 files (16,222 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 113 module specs with automated validation
  • 38 MCP tools, ~200 API endpoints, 42 route modules, 82 tables
  • 29 commits, 13 PRs merged this release

[0.20.0] - 2026-03-07

Added

  • Discord slash commands — /ask, /status, /help commands with threaded conversations, session resume, and GitHub mention acknowledgment (#736)
  • Telegram bridge work-intake mode — submit work tasks directly from Telegram conversations (#698)
  • One-line install script — simplified onboarding for newcomers with a single curl command (#717)
  • Work task retry button — retry failed work tasks from the dashboard UI (#735)
  • Governance tier in council edit — configure governance tier directly from the council edit UI (#731)
  • Graceful work task drain — server shutdown waits for in-progress work tasks to complete before exiting (#721, #723)
  • Wallet idle lock — auto-lock wallet on tab focus when idle timeout exceeded (corvid-agent-chat#55)
  • WCAG AA accessibility — full accessibility compliance across the chat client (corvid-agent-chat#41)
  • Client-side rate limiting — message rate limiting in the chat client to prevent spam (corvid-agent-chat#35)

Security

  • CodeQL alert remediation — resolved 4 log-injection alerts (#701, #703)
  • CVE fix — override express-rate-limit >=8.2.2 for rate-limit bypass vulnerability (#696)

Fixed

  • Polling injection false positive and infinite retry loop (#737)
  • E2E test isolation from production database (#730, #732)
  • PR dedup check to prevent duplicate work tasks (#725)
  • Layer 0 governance violation in waitForSessions extraction (#722)
  • Heartbeat polling for council waitForSessions (#716)
  • DB-filter test lifecycle and marketplace spec enhancements (#711)
  • Test global state leaks from PR #699 (#702)
  • Deduplicate exports and fix inflated warning count in spec-check (#692)
  • Use bun x instead of bunx for work task validation (#684)
  • Prevent duplicate user messages in chat client (corvid-agent-chat#40)
  • Icon sizing consistency across 6 chat UI elements (corvid-agent-chat#45-54)
  • Wildcard route redirect in specl (specl#72)
  • Accessibility issues across 5 specl components (specl#66)
  • Standardize git author identity in corvid-reputation workflow (corvid-reputation#1)

Refactored

  • Extract repo map and symbol extraction into repo-map.ts (#693)
  • Extract WS event bus into councils/events.ts (#694)
  • Extract validation pipeline from work service (#695)
  • Remove unused printPrompt and clearAgentCardCache exports (#691)
  • Split chat.ts view into smaller components (corvid-agent-chat#57)

Tests

  • 6 untested modules covered with 66 new tests (#699)
  • Broadcasting, workflow service, and algochat init coverage (#697)
  • Toast, chat-messages, and chat-search coverage (+69 tests, corvid-agent-chat#58)
  • Device-name and wallet lifecycle coverage (+51 tests, corvid-agent-chat#56)
  • WelcomeComponent, GitHubOAuthService, FrontmatterEditor coverage (+26 tests, specl#73, #74)
  • SectionEditor, GithubConnect, TableEditor, SectionNav, SpecPreview coverage (specl#65, #67)

Documentation

  • Use case gallery and how-it-works guide (#712)
  • Document all undocumented exports — 23 to 0 warnings (#690)
  • Update stale test and spec counts across docs (#689)
  • Fix stale session endpoint, update route count (#687)
  • Add missing spec.md files for 4 specl modules (specl#68)

Dependencies

  • Angular 21.2.0 and minor dependency bumps (specl#60)

Stats

  • 5,704 unit tests across 228 files (15,958 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 111 module specs with automated validation
  • 37 MCP tools, ~200 API endpoints, 70 migrations, 81 tables
  • 52 PRs merged across 4 repositories this week

[0.19.0] - 2026-03-06

Security

  • 232 security tests across 3 dedicated test suites — security-audit (104), jailbreak-prevention (81), rate-limit-bypass (47)
  • SECURITY.md threat model — expanded from 90 to 324 lines with asset inventory, threat actors, attack surfaces, injection detection, rate limiting, and incident response playbook
  • Jailbreak prevention tests — multi-turn attacks, encoding bypasses (base64, hex, ROT13), persona hijacking, instruction hierarchy, payload splitting, language-switching
  • Rate limit bypass tests — IP rotation, header manipulation (X-Forwarded-For, X-Real-IP), concurrent floods, sliding window, content length guard
  • Dependency audit — manual audit of all direct and transitive dependencies; 0 HIGH/CRITICAL CVEs in direct deps, 5 transitive overrides analyzed (docs/dependency-audit.md)
  • External review scope document — P0–P3 critical paths, test coverage map, access instructions for third-party auditors (docs/external-review-scope.md)

Added

  • ProcessManager decomposition — extracted TimerManager and ResilienceManager from ProcessManager for cleaner separation of concerns (#453)

Stats

  • 5,427 unit tests across 212 files (15,465 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 111 module specs with automated validation
  • 37 MCP tools, ~200 API endpoints, 70 migrations, 81 tables

[0.18.0] - 2026-03-06

Added

  • Governance tier architecture — council launches support vote types (standard, weighted, unanimous) and governance tiers; governance_votes and governance_member_votes tables for structured multi-agent voting (#627)
  • Empty state components — dashboard pages (agents, councils, work tasks, schedules, sessions) show helpful empty states with ASCII art icons, descriptions, and quick-action buttons (#623)
  • Skeleton loading states — animated skeleton placeholders replace "Loading..." text across all list views (#623)
  • Deduplication state persistence — dedup_state table for crash-resilient dedup across polling, messaging, and bridge modules (#613)
  • Tooltip directive — reusable appTooltip directive for truncated text throughout the dashboard (#618)
  • McpServiceContainer — extracted MCP tool dependencies into a typed service container for cleaner dependency injection (#615)
  • Global PR review polling — centralized PR review detection across all configured repos (#628)
  • Auto-merge dedup — prevents duplicate merge attempts on the same PR (#626)

Improved

  • Council list cards — show last launch synthesis summary, stage badges, member chips, and chairman highlighting (#618)
  • Schedule list — improved layout with status badges, next-run display, and test-data filtering (#616)
  • Work task list — inline create form, status/type filters, and rich task cards with validation indicators (#616)
  • Migration robustness — IDEMPOTENT_CREATE_INDEX regex handles CREATE UNIQUE INDEX IF NOT EXISTS; safeAlter pattern for idempotent column additions

Fixed

  • Escaped backtick rendering in Angular template literals (#623, #616, #618)
  • Migration 066/067 collision between dedup_state and governance_tiers (#627)
  • Baseline migration schema drift for governance tables (#627)
  • Dead template references (filteredSchedules, activeFilter) in schedule-list component (#616)
  • Octal escape sequences in CSS template literals (#616)
  • Duplicate empty-state blocks in work-task-list component (#616)

Stats

  • 5,192 unit tests across 206 files (14,598 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 108 module specs with automated validation
  • 37 MCP tools, ~200 API endpoints, 70 migrations, 81 tables

[0.17.0] - 2026-03-06

Added

  • Repo blocklist with auto-block on PR rejection — rejected PRs automatically add repos to the blocklist; org-wide wildcard support (e.g. vapor/*); enforced in polling and webhooks (#547, #554, #561, #565)
  • Branch protection — script to enable branch protection on unprotected public repos (#560)
  • Dashboard summary batch endpoint — GET /api/dashboard/summary aggregates agents, sessions, work tasks, and recent activity in a single call (#567)
  • Schedule management via MCP — update action added to corvid_manage_schedule tool (#568)
  • Daily review schedule action — automated end-of-day retrospective with PR outcome analysis (#570)
  • Selective tool gating — scheduler sessions can restrict which MCP tools are available (#578)
  • Permission broker — capability-based security layer for agent actions (#578, #579)
  • 69 module specs for full server coverage — 109 total specs with automated validation (#552)

Security

  • CSP and Permissions-Policy headers — Content-Security-Policy and Permissions-Policy middleware on all responses (#566)
  • WebSocket post-connect auth timeout — 5-second deadline to authenticate after connection (#564)
  • @hono/node-server authorization bypass patch — override for GHSA-wc8c-qw6v-h7f6 (#575)
  • Blocklist enforcement in polling and webhook handlers (#565)

Fixed

  • Localhost exempted from rate limiting for local dashboard access (#562, #563)
  • Logger special characters test handles JSON format in production mode (#574)
  • Cosign installer pinned to current v3 SHA; id-token permission for release workflow (#545, #546)

Changed

  • Service bootstrap extracted from server/index.ts into server/bootstrap.ts — cleaner startup, testable initialization (#579)
  • Architecture docs synced with all 200+ API endpoints and 70 tables (#569)
  • README updated with At a Glance stats section, accurate counts (#573, #576, #577)

Stats

  • 5,040 unit tests across 202 files (14,118 assertions)
  • 360 E2E tests across 31 Playwright specs
  • 109 module specs with automated validation
  • 37 MCP tools, ~200 API endpoints, 64 migrations, 82 tables

[0.16.0] - 2026-03-04

Added

  • RBAC enforcement across all routes — tenantRoleGuard on 75+ write endpoints across 17 route files; owner-only guards on settings and billing (#529, #530, #531)
  • Admin role enforcement — system-logs, performance, github-allowlist, wallet-summary endpoints restricted to admin (#505)
  • Tenant isolation on usage endpoints — usage and allowlist routes scoped per-tenant (#494)
  • EntityStore signal store pattern — extracted reusable EntityStore<T> for Angular services; migrated ProjectService, CouncilService, WorkflowService (#499, #501)
  • WebSocket heartbeat — server-sent timestamps for connection health monitoring (#461)
  • Cache-Control headers for static assets (#493)
  • Retention policies for append-only tables (#481)
  • SSRF private IP range blocking (#479)
  • SBOM generation and Docker image signing via Cosign (#495)
  • Multi-tenant route isolation across all 14 previously unscoped handlers (#439)
  • Test coverage expansion — 15 new test suites covering DB modules (agents, sessions, projects, councils, spending, allowlist, work-tasks, pr-outcomes, health-snapshots, notifications, webhooks, reputation, backup, algochat-messages, github-allowlist, mcp-servers, plugins), polling service, scheduler cron-parser, priority-rules, auto-merge, and ci-retry (#489, #502–#504, #512–#514, #516–#519, #532, #533)

Changed

  • WebSocket auth — query-string ?key= deprecated in favor of Authorization: Bearer header (#496)
  • Polling service decomposed into 3 focused services (#498)
  • shared/types.ts split into domain-specific modules (#497)
  • DB count queries consolidated into queryCount() helper (#511)
  • Bumped @anthropic-ai/claude-agent-sdk to 0.2.68 (#518)
  • Bumped actions/github-script from 7 to 8 (#434)
  • CI workspace setup extracted into composite action (#482)
  • GH Actions pinned to SHA with explicit permissions (#478)

Fixed

  • Condenser tries all providers before truncation fallback (#484)
  • Missing subscription_items table for UsageMeter (#483)
  • Auto-refill agent wallets before on-chain sends (#476)
  • Git author identity in LaunchAgent plist (#515)
  • schedule_skip added to AuditAction type (#470)
  • Baseline migration schema drift (#469)
  • Respect human issue assignments and fix log rotation (#440)
  • Script injection in GitHub Actions workflows (#492)
  • 5 npm audit vulnerabilities resolved via overrides (#488)

Docs

  • Clarified Claude auth — API key not required when Claude Code CLI is installed (#543)
  • Updated stale stats across README, docs site, and CLAUDE.md (#534)
  • Documented inline API endpoints (#490)
  • Security review of API key authentication (#460)

[0.15.0] - 2026-03-04

Added

  • RBAC enforcement across all routes — tenantRoleGuard added to all write endpoints across 17 route files (75 endpoints), with owner-only guards on settings and billing (#520, #526, #527, #528)
  • Test coverage: scheduler — 70 new tests for cron-parser (37 tests) and priority-rules (33 tests) (#524, #532)
  • Test coverage: polling — 28 new tests for auto-merge (14 tests) and ci-retry (14 tests) (#525, #533)

Fixed

  • TypeScript strict mode errors in plugins.test.ts (double cast for PluginCapabilityRecord)
  • Unused @ts-expect-error directives in polling test files

[0.14.0] - 2026-02-28

Added

  • Slack notification integration — schedule approval notifications, work task results, and agent questions routed to Slack channels
  • Health-gated scheduling — priority rules engine suppresses non-critical work when system health is degraded
  • Auto-merge polling — automatically merges agent PRs when all CI checks pass
  • CI retry service — detects failed CI on agent PRs and spawns fix sessions
  • Performance metrics — collection, trend detection, and regression alerts
  • Usage monitoring — schedule execution frequency tracking and anomaly detection
  • Feedback loop — PR outcome tracking for schedule effectiveness learning

Changed

  • Database schema version bumped to 62 (15 new migrations since v0.13.0)
  • Route modules expanded from 28 to 34
  • Module specs expanded from 33 to 38

[0.13.0] - 2026-02-25

Added

  • Centralized DedupService — unified deduplication with TTL expiry, LRU eviction, and SQLite persistence; replaces scattered per-feature dedup logic (#254)
  • Test coverage reporting — CI pipeline now generates and uploads coverage reports (#252)
  • Unit tests for critical untested services — expanded test coverage for previously untested code paths (#255)

Fixed

  • Cross-repo dedup collisions — mention polling now scopes dedup keys per-repository to prevent false-positive suppression (#232)
  • SQL injection prevention — replaced string interpolation with parameterized queries across database layer (#251)
  • Dockerfile bun version — pinned to bun:1.3.8 to match lockfile (#229)
  • CI/Docker workflow timeouts — added timeout-minutes to prevent runaway jobs (#228)
  • bun.lock regeneration — resolved lockfile drift from dependabot dependency merges (#230)

Changed

  • Tool handler decomposition — split monolithic tool-handlers.ts into domain-specific modules for maintainability (#253)
  • AlgoChatBridge decomposition — refactored into focused single-responsibility services (#256)

Dependencies

  • Bumped @opentelemetry/auto-instrumentations-node (#225)
  • Bumped @anthropic-ai/sdk from 0.74.0 to 0.78.0 (#227)

[0.11.0] - 2026-02-23

Added

  • Slack integration — bidirectional Slack bridge for channel-based agent interaction, notification delivery, and question routing (#143, #212)
  • ChannelAdapter interface — unified adapter pattern for messaging bridges; AlgoChatBridge refactored to conform (#142, #209)
  • Koa-style middleware pipeline for agent messaging — composable request processing (#151, #217)
  • OnChainTransactor extraction — separated on-chain transaction handling from AgentMessenger for cleaner separation of concerns (#152, #219)
  • Fire-and-forget async messaging — non-blocking message delivery mode for AgentMessenger (#153, #220)
  • Circuit breaker + per-agent rate limiting — protects against overwhelming individual agents and the system (#154, #221)
  • Parallel council responses — agents respond concurrently during council discussion rounds, improving throughput (#216)
  • AST symbol context in work tasks — work task sessions now receive richer code context through AST analysis (#141, #211)
  • UI audit (phases 1–9) — complete dashboard and component rework: agent list/detail tabs, session state display, council panels, settings sections, analytics charts, work tasks, schedules, system logs, feed, and personas — all with consistent styling
  • Client pages for personas, skill bundles, reputation, marketplace, and MCP servers — full Angular services and E2E test coverage
  • Reputation auto-compute — stale scores auto-recompute on read (5-minute threshold); SVG score rings, trust badges, color-coded component bars with weight percentages
  • Marketplace enhancements — detail panels with star ratings, trust badges from reputation, federated listings from remote instances
  • 100% testable API E2E coverage — 348 tests across 30 Playwright spec files covering 198/202 endpoints
  • Module specs for reputation scorer, marketplace service, and marketplace federation

Fixed

  • Persona manager UX — replaced vertical card list with horizontal chip picker; detail form now immediately visible without scrolling
  • Suppressed expected 404 toasts on persona endpoints (unconfigured personas are normal)
  • Reputation score ring SVG arcs — switched from CSS style bindings to SVG attribute bindings for cross-browser rendering
  • Mention polling no longer blocks on idle sessions (#214) or permanently skips deduped mentions (#213)
  • LaunchAgent PATH for claude CLI discoverability (#218)
  • Replaced Math.random() with crypto.randomBytes() in E2E fixtures
  • Removed unused variables flagged by code scanning
  • Copilot review feedback (two rounds): marketplace query param alignment, system-log server-side filtering, cronToHuman() NaN handling, federation SSRF mitigation with DNS rebinding/IP validation, Promise.allSettled for resilient persona checks

Changed

  • E2E fixtures consolidated with gotoWithRetry extraction and api.seedWorkflow() helper
  • Reputation scorer exposes computeAllIfStale() and computeAll() for bulk operations
  • Marketplace component injects ReputationService for cross-feature trust badge display

[0.10.0] - 2026-02-21

Added

  • Schedule approval notifications — when a schedule execution needs owner approval, notifications are sent via all configured channels (Telegram, Discord, AlgoChat, etc.) instead of only showing in the dashboard
  • Proactive schedule prompts — all custom/suggest schedules now use corvid_notify_owner, corvid_web_search, corvid_deep_research, and corvid_create_work_task where appropriate
  • Dynamic community engagement — Weekend Community schedule searches for trending repos instead of starring a hardcoded list
  • Rotating self-improvement focus — corvid-agent self-improvement rotates by day-of-month: test coverage (1st-7th), type safety (8th-14th), error handling (15th-21st), performance (22nd-31st)
  • Issue-first project improvement — CorvidLabs project self-improvement checks open issues first before looking for generic improvements

Changed

  • All schedules now run on CorvidAgent (Claude Opus) — removed Qwen Coder agent dependency
  • Self-improvement schedules bumped to 2x/week (Mon+Thu for projects, Tue+Fri for corvid-agent)
  • Removed PR Comment Response and Morning PR Review schedules (covered by mention polling + Stale PR Follow-Up)
  • Weekly Improvement Suggestions refocused on public API correctness for ts-algochat and swift-algochat

[0.9.0] - 2026-02-20

Added

  • Rich polling activity feed — activity endpoint parses session initialPrompt to return structured fields (repo, number, title, sender, url, isPR, triggerType); UI shows status dots, PR/Issue labels, @sender, trigger type badges, and summary bar
  • Stampede throttling — MAX_TRIGGERS_PER_CYCLE (5) caps sessions spawned per config per poll cycle, preventing runaway session creation when many mentions arrive at once
  • Model library refresh — added Claude Sonnet 4.6, GPT-4.1/Mini/Nano, o3, o4-mini, Qwen 3 32B (local), and 5 new Ollama cloud models (Qwen 3.5, DeepSeek V3.2, Qwen 3 Coder Next, Devstral Small 2, Nemotron 3 Nano)
  • Ollama cloud model support — :cloud suffix routing, local proxy for auth, merged local+remote model listings
  • Model exam system — 18 test cases across 6 categories (coding, context, tools, algochat, council, instruction) with per-category scoring
  • Expanded model family detection — qwen3, qwen3moe, deepseek2, command-r, nemotron, hermes, firefunction added to inferFromName
  • Cloud model test suite — 32 tests covering parseModelSizeB, isCloudModel, hostForModel routing, and size gating
  • Model capability name inference for 12 families (up from 5)
  • AST code navigation tools: corvid_code_symbols and corvid_find_references for cross-file symbol analysis (#183)
  • AST-powered work task repo maps for smarter context in agent sessions (#184)
  • WebSocket authentication improvements (#184)
  • Module specification system with bun run spec:check CI enforcement (#185)
  • Ollama reliability improvements — testable tool parser, retry logic, context-aware truncation, smarter nudges, fuzzy repeat detection (#186)
  • Claude Code subscription auth support (no API key required)

Fixed

  • Opus 4.6 pricing corrected to $5/$25 (was legacy $15/$75 from Opus 4/4.1 era)
  • Haiku 4.5 pricing corrected to $1/$5 (was $0.80/$4 from Haiku 3.5)
  • Anthropic provider model IDs updated to current versions (claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5-20251001)
  • save_memory tool returning confusing "on-chain send failed" messages that caused model retry loops (#181)
  • Model family detection ordering — specific families like qwen3moe matched before generic qwen
  • GitHub mention dedup race condition (#174)
  • Ollama slot weight leak on session kill (#173)
  • Failed on-chain sends marked as failed, not completed (#171)
  • Project bundle tool merging for explicitly scoped agents (#176)
  • Ollama multi-tool chain hallucination in text-based tool calling (#180)
  • Ollama tool call reliability — smarter context management and retry logic (#186)

Changed

  • Fallback chains updated — GPT-4.1 replaces GPT-4o in high-capability/balanced, GPT-4.1 Nano added to cost-optimized, cloud chain expanded
  • Default Anthropic model changed from claude-sonnet-4 to claude-sonnet-4-6

[0.8.0] - 2026-02-17

Major release with 1757 server tests, 47 database migrations, and full-stack agent orchestration across five development phases.

Phase 5 — Bridges, Personas, Skills, Voice

  • Bidirectional Telegram bridge — talk to agents from your phone via long-polling; voice note support with automatic STT transcription; per-user sessions; authorization via TELEGRAM_ALLOWED_USER_IDS
  • Bidirectional Discord bridge — talk to agents from Discord via raw WebSocket gateway (no discord.js dependency); auto-reconnect with exponential backoff; heartbeat and session resume; per-user sessions
  • Character/Persona system — give agents distinct personalities with archetype, traits, background, voice guidelines, and example messages; persona is composed into the system prompt for both SDK and direct processes
  • Skill bundles — composable tool + prompt packages assignable to agents; 5 built-in presets (Code Reviewer, DevOps, Researcher, Communicator, Analyst); custom bundle creation; tools and prompt additions merged at session start
  • Voice support (TTS/STT) — OpenAI TTS API with 6 voice presets and SQLite-backed audio caching; OpenAI Whisper STT for voice message transcription; per-agent voice configuration
  • DB migrations 44-47 — agent personas table, skill bundles + assignment tables with preset data, voice cache table, voice columns on agents
  • 61 new tests across 8 test files (personas, skill bundles, routes, bridges, voice)
  • New API endpoints — /api/agents/{id}/persona (GET/PUT/DELETE), /api/skill-bundles (CRUD), /api/agents/{id}/skills (assign/unassign)

Phase 1-4 (prior development)

  • CLI mode — npx corvid-agent chat "..." for terminal-first interaction with device authorization flow
  • Plugin SDK — dynamic tool registration allowing agents to extend their capabilities at runtime
  • OpenAPI documentation — auto-generated API docs served from /api/docs
  • Container sandboxing — isolated execution environments for agent-generated code with resource limits
  • Agent marketplace — publish, discover, and consume agent services with credit-based payments
  • Reputation & trust scoring — track agent reliability, quality, and trustworthiness over time
  • WhatsApp & Signal channels — reach agents from mobile messaging apps
  • Multi-model cost-aware routing — automatic model selection based on task complexity, latency, and budget
  • Multi-tenant isolation — team workspaces with tenant-scoped data access
  • Billing integration — usage metering and billing for hosted deployments
  • Kubernetes & Helm deployment — production-grade orchestration with Helm charts and K8s manifests
  • Security hardening — enhanced input validation, rate limiting improvements, and audit coverage

Migration Notes

  • Database: SCHEMA_VERSION 43 → 47 (4 new migrations run automatically on startup)
  • No breaking changes: All new DB columns have DEFAULT values; all features opt-in via env vars
  • No new npm dependencies: Discord gateway uses raw WebSocket; TTS/STT are direct API calls

[0.7.0] - 2025-12-15

  • Graph-based workflow orchestration with suspend/resume
  • OpenTelemetry tracing, Prometheus metrics, and immutable audit logging
  • A2A protocol support (Google Agent-to-Agent interoperability)
  • Structured memory with vector embeddings and FTS5 search
  • GitHub webhook automation with @mention triggers
  • Multi-channel notifications (Discord, Telegram, GitHub Issues, AlgoChat)
  • Agent-to-owner communication (corvid_ask_owner, corvid_notify_owner)
  • Tree-sitter AST parser for code understanding
  • Automation bootstrap scripts
  • 850 tests across 34 files