Skip to content

CHANGELOG.md

Latest commit

 

History

History
132 lines (110 loc) · 12.5 KB

CHANGELOG.md

File metadata and controls

132 lines (110 loc) · 12.5 KB
Raw

Changelog

All notable changes to this project are documented below.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.0] — 2026-06-06

Added

  • Marketing landing page — new website/ directory (Vite + React 19 + Framer Motion 12 + Tailwind 3) with a redesigned hero, features grid, architecture diagram, terminal demo, metrics, tech stack, use cases, docs preview, changelog timeline, FAQ accordion, and CTA panel. Liquid-glass surfaces, animated orb background, custom dark palette (purple / pink / cyan / green / yellow), and prefers-reduced-motion support.
  • Multi-platform adapter gateway — src/adapters/ now ships with Discord, Slack, SMS (Twilio), Voice (Twilio), WhatsApp, Email (Nodemailer), Webhook, and Bot-Commands adapters behind a single gateway.ts interface. Includes 4 adapter test files (gateway, sms, whatsapp, hmac).
  • HMAC-signed REST API — src/api/hmac.ts with timing-safe comparison, replay-protection window, and a hmac.test.ts covering sign / verify / replay / tampering cases.
  • New AI providers — Mistral, Azure OpenAI, and Together AI are now first-class in src/ai/providers.ts, with a providers.test.ts smoke test.
  • Test framework migration — agent / audit / bench / chat / cron / harness / mcp / memory / mesh / sandbox / tools / vault suites migrated from test-*.ts to *.test.ts naming so they run under bun test --coverage and Codecov. New tests added: planner, ratchet, supervisor, runtime, docker / filesystem / process (sandbox), runner, reporter, embedding, computer, rbac.
  • GitHub Actions coverage reporting — ci.yml runs bun test --coverage and uploads to Codecov with a dashboard flag for vitest output.
  • docs/superpowers/specs/2026-06-05-cost-attribution-design.md — design spec for Sprint 17 (cost attribution & agent benchmarking), tracking real implementation that replaces the stubbed src/billing/tracker.ts and src/telemetry/cost.ts.

Changed

  • tsconfig.json now excludes website/ from the root tsc --noEmit (the website has its own tsconfig.app.json); the pre-commit hook runs both typechecks in sequence.
  • .husky/pre-commit runs bunx tsc --noEmit (root) and bunx tsc -b (website) so both projects stay clean.
  • Dashboard files were refactored from dashboard/src/site/*.tsx to dashboard/src/site/components/ to keep imports relative now that they mirror the website's component layout.
  • README and ROADMAP.md updated to mention the website, the adapter gateway, and the upcoming v0.6.0+ vision.

Removed

  • Runtime SQLite databases (data/audit/*.db*, data/telemetry/*.db*, data/experience/*.db*) and ephemeral state directories (.aegis/, .superpowers/, .commandcode/, .worktrees/) are now gitignored — they are runtime state, not source.
  • Test stub utilities (src/utils/doSomething.ts, greet.ts, hello.ts), the one-time copy-site.js script, and the top-level task.md / implementation_plan.md planning artifacts are no longer tracked.

[Unreleased]

Added

  • Phase 6 — Self-improvement skill pipeline: src/improve/eval-validator.ts validates skill candidates against the GraderSuite before publishing; src/improve/skill-monitor.ts tracks skill performance over time with trend analysis, degradation detection, and improvement suggestions. CLI: aegis improve validate (run/list/stats), aegis improve monitor (status/list/degrading/top/record).

  • Phase 7 — Multi-agent orchestration harness: src/harness/multi-agent.ts defines 6 coordination patterns (sequential, parallel, debate, hierarchical, voting, refine) with typed MultiAgentTest definitions and a scenario factory; src/harness/multi-agent-collector.ts collects handoff accuracy, context loss, consensus quality, parallel speedup, Gini coefficient, and per-agent metrics. CLI: aegis eval multi-agent (list/run/metrics).

  • Phase 8 — Golden dataset manager: src/harness/golden-dataset.ts implements the Silver→Gold→Audit→Archived pipeline for human-verified eval tasks with file persistence; src/harness/golden-validator.ts cross-validates tasks across multiple LLM models. Tasks auto-discovered via evals/golden/ with aegis eval run --golden. CLI: aegis eval golden (create/list/promote/audit/archive/stats).

  • Unit test suites (66 tests total): src/improve/eval-validator.test.ts (11 tests), src/improve/skill-monitor.test.ts (17 tests), src/harness/golden-dataset.test.ts (24 tests), src/harness/multi-agent-collector.test.ts (14 tests).

  • Dashboard eval panels: Golden Dataset (🥇), Skill Pipeline (🔧), and Multi-Agent (🤖) panels in dashboard/src/routes/Eval.tsx with stat summaries and CLI quick-refs.

  • Calibration comparison script improvements: scripts/calibration-compare.ts now supports LIMIT=N env var to run only N examples (default: all 40), and Groq + Ollama models in DEFAULT_MODELS with proper API key handling for local providers.

  • JSON test loader: src/harness/discover.ts now includes a jsonLoader that parses GoldenTask JSON files (single or array) into TestCase objects.

  • Auto-generated docs source — scripts/extract-commands.ts parses src/cli/commands/*.ts via the TypeScript Compiler API and emits shared/commands.json (128 commands as of this write). The dashboard and website docs sections now consume this file as a single source of truth instead of hand-maintained command arrays. New scripts: bun run docs:generate / bun run docs:check. pretest runs docs:check to fail CI on drift.

Fixed

  • SkillMonitor recursion bug: generateSuggestions() no longer calls this.getPerformance() internally, which caused infinite recursion and Maximum call stack size exceeded.

  • Generator test suite — scripts/__tests__/extract-commands.test.ts with 3 fixture files (simple.ts, with-options.ts, with-subcommands.ts) covering simple chains, alias/description/options/defaults, and parent+subcommand extraction. All 3 tests pass; the test is registered in scripts/run-tests.ts.

  • Dashboard docs layer — dashboard/src/data/commandGroups.ts derives 11 hand-curated groups (system, setup, agents, orchestration, memory, knowledge, schedule, serve, adapters, sessions, runtime) from shared/commands.json and adds icon/tag metadata. dashboard/src/routes/Docs.tsx now imports this layer; the hand-maintained commandGroups array is gone.

  • Website docs layer — website/src/data/docTopics.ts exposes navGroups, docTopics, and defaultTopic (15 curated topics) sourced from shared/commands.json via a row(name) helper that resolves command descriptions from the JSON. website/src/sections/DocsSection.tsx now imports this layer; the hand-maintained navGroups / docContent / defaultContent constants are gone.

  • Design spec — docs/superpowers/specs/2026-06-06-docs-section-update-design.md captures the contract: schema, file layout, CI wiring, risks, migration plan.

  • Multi-stage Dockerfile with oven/bun:1-slim production image, HEALTHCHECK, non-root user

  • docker-compose.yml with named volume, env passthrough, and dashboard-dev profile (Vite HMR)

  • .dockerignore with comprehensive exclusion rules

  • AES-256-GCM vault encryption: src/vault/crypto.ts with key management (AEGIS_VAULT_KEY env var or auto-generated ~/.aegis/.vault-key), auto-migration from legacy vault.json, no plaintext fallback

  • Structured logger (src/cli/logger.ts): levels, JSON output in non-TTY, pretty-print in TTY, module-scoped instances, writes to stderr

  • Error boundaries (src/cli/guard.ts): registerErrorBoundaries() for unhandledRejection + uncaughtException

  • Graceful shutdown (index.ts): SIGINT/SIGTERM handlers with agent cleanup via agentManager.destroy()

  • Zod config validation (src/config.ts): validateConfig() with field-level salvage on invalid config

  • API hardening (src/api/server.ts): CORS with configurable origins, rate limiting (100/min), input validation, security headers

  • Unit test suites:

    • src/memory/test-memory-system.ts — 33 tests: init, user profile CRUD, daily logs, auto memories, fact extraction, search, context building
    • src/memory/test-vector.ts — 21 tests: add, search, category, stats, remove, edge cases
    • src/memory/test-session-store.ts — 13 tests: CRUD, rename, export, agent traces
    • src/agent/test-engine.ts — 85 assertions: IPC handling, hooks, kill, routing, events, listing

  • Architecture documentation: agent system, memory system, sandbox system deep-dives

  • Developer guides: creating modes, tools, and agent types step-by-step

  • REST API reference: all endpoints with examples, validation rules, error codes

  • CHANGELOG.md and SECURITY.md files

Changed

  • dashboard/src/routes/Docs.tsx — no longer hardcodes 13 stale command groups; renders directly from the generated JSON via dashboard/src/data/commandGroups.ts.
  • website/src/sections/DocsSection.tsx — no longer hand-maintains 16 marketing copy blocks; renders from website/src/data/docTopics.ts. The "Reflection loop" topic no longer references the non-existent aegis agent-run --ratchet flag; "Provenance" topic is gone (no provenance command exists).
  • dashboard/tsconfig.json and website/tsconfig.app.json — added resolveJsonModule: true so both can import shared/commands.json.
  • scripts/run-tests.ts — registers the new docs generator test before the integration tests.
  • src/vault/manager.ts — vault serialized as encrypted blob (vault.enc), auto-migrates from legacy vault.json, removes stale plaintext
  • src/cli/commands/config.ts — shows vault.enc path and "AES-256-GCM encrypted" in status
  • README.md — added Docker usage section with build/run/compose commands and security notes
  • docker-compose.yml — removed deprecated version field

[2026-05-31]

Added

  • Sandbox system: FilesystemSandbox, ProcessSandbox, DockerSandbox implementations with common interface
  • Computer use tool: screen interaction via src/tools/computer.ts
  • Agent evaluation harness: src/harness/ with reporter, runner, and test types
  • Dashboard routes for MCP, Memory, Serve, Setup, Skills, Status with corresponding UI components
  • AgentMemory sidecar integration: REST connector, smart-search fusion, CLI mode, mode registration
  • Mode launcher: keyboard-navigable mode selection, MCP CLI, memory CLI, vector memory, type fixes
  • Web tools: web_fetch and web_search tools
  • MCP integration: stdio and HTTP server modes
  • Shell mode: inline command execution in chat
  • Checkpoint/rewind system in chat store
  • Model picker UI: provider and model selection in chat
  • Vector memory: 128-dim hash-based embeddings with cosine similarity search
  • 11-mode agent type system: build, plan, read, write, test, validate, review, debug, document, refactor, deploy, monitor, explore
  • Session management actions: delete, rename, export with pendingAction confirmation flow
  • CI/CD pipeline: GitHub Actions with setup-bun, dependency caching, typecheck before tests

Fixed

  • Corrected TypeScript type issues in ChatState.config — removed any casts in renderer and store
  • Test exit codes, session persistence, slash command handling
  • README formatting and clarity improvements
  • Security: removed API key history from git

Infrastructure

  • GitHub Actions CI with matrix strategy (typecheck + test)
  • PR templates for bug reports, feature requests, and changes
  • PR description templates for security, session management, TUI tests, and type tightening

[2026-05-31] — Initial Release

Added

  • First commit with core project structure
  • Agent system: AgentManager, AgentEngine, HookRegistry, AgentRuntime
  • CLI framework with command routing
  • AI provider system with OpenAI, Anthropic, DeepSeek, Ollama support
  • Tool registry with 10 built-in tools (read, write, edit, bash, grep, glob, etc.)
  • GPT-4 based agent worker
  • Mode system with chat, config, status, and dashboard modes
  • TUI dashboard with agent list, activity log, status bar, command bar
  • Memory system with user profile, long-term memory, daily logs, auto memories
  • Session store with save/load/list/delete/rename/export
  • Fact extraction with regex-based pattern matching
  • Vault system for credential storage
  • Cron engine for scheduled tasks
  • MCP server for agent tool exposure
  • Skills system for reusable workflows
  • Web search and web fetch tools
  • Configuration system with environment variable support
  • Original 11-mode system