autocomputer v0.10.0 — Deep-Slate Rebrand + Path Standardization

[v0.10.0] - 2026-08-01

🎨 Highlights

• Visual rebrand: Obsidian brass identity replaced with a deep-slate + run-green
  developer palette (#020617 base, #22C55E accent, #38BDF8 pop) — designed
  with ui-ux-pro-max design intelligence. Share Tech Mono (display) + Fira Code
  (body) fonts for a HUD feel. Zero-build preserved.

🔧 Changed

• Runtime data standardized to %APPDATA%/autocomputer/ (was ~/.qclaw):
  workspace/screenshots, unified_state.db, flows.json, autocomputer_audit.db.
  Existing data migrated automatically.
• Privacy: git history rewritten — all commit emails now yygg693@users.noreply.github.com,
  personal paths removed from history and docs.
• README fully rewritten (7-page dashboard, MCP server, real test counts 61).
• prompt permission semantics: fails closed with a distinct error so callers can
  distinguish "needs confirmation" from a hard deny.

🐛 Fixed

• CI python job missing typer/rich deps → 5 CLI tests failed on CI only.
• README GitHub rendering (centered header, accurate badges).

[v0.9.0] - 2026-08-01

autocomputer v0.9.0 — MCP Server + Obsidian Dashboard

[v0.9.0] - 2026-08-01

🚀 Highlights

• MCP Server (mcp-server/) — 8-tool Model Context Protocol integration for AI agents (see / find / act / verify_target / apps / audit / permissions / memory), with Windows built-in OCR (zero downloads, Chinese zh-Hans-CN). Fully integrated with the Reasonix skill (junction link + SKILL.md hardlink, single source of truth).
• Obsidian design-system Dashboard — near-black layered palette + brass #c9a962 + cyan #4fd6ff, glass cards, jelly press/spring physics and gold particle FX (pure CSS/JS, zero build). 7 pages: Dashboard / Desktop Console / Recorder Editor / Flows / Monitor / Security / Memory.
• Rust core online — compiled _core.pyd (GNU toolchain); real window listing, click/focus/move through the browser UI.
• One-click launch — scripts/start-dashboard.cmd starts the server and opens the browser automatically.

✨ New features

• Desktop Console page: live window list, focus, coordinate click/move (/api/execute).
• Memory page: learned actions + per-app permissions + MCP audit from unified_state.db.
• Flows persisted server-side (%APPDATA%/autocomputer/flows.json); POST/DELETE /api/flows.
• /api/logs ring buffer + /api/security (audit stats from SQLite).
• Recorder editor redesigned as a 3-column RPA-style editor: action library | step timeline | property panel with per-action param forms and a simulated-screen preview.
• Jelly physics (press squash + spring-back) and click-burst particles across the UI.

🐛 Fixed

• _bridge.py: _get_rust_attr now catches ImportError — pure-Python fallback actually works (tests went 15 failed → 32 passed).
• Audit DB path moved from CWD to %APPDATA%/autocomputer/; ALTER TABLE now conditional.
• CI now runs cargo test + a Python pytest job (previously no tests ran in CI).
• Server crashed on GBK consoles due to emoji in the banner — UTF-8 reconfigure + UnicodeDecodeError guard on JSON bodies.
• html/body missing width:100% collapsed the layout (~400px dead space on the right of every page).
• Duplicate deleteFlow shadowed the API version, breaking flow deletion.
• ScreenContext.capture now degrades gracefully when an OCR backend is installed but unusable.

🔧 Changed

• Version bumped to 0.9.0 (was 0.8.0); project description updated across pyproject.toml, __init__.py, Cargo.toml, README.
• UI fully localized to Chinese.
• Dashboard stat "tests passed" now reports the real dynamic count (61 = 29 Rust + 32 Python).

[v0.8.0] - 2026-07-22

v0.8.0 — Production Release 🦀🐍🌐

v0.8.0 — Production Release 🦀🐍🌐

autocomputer v0.8.0 is the result of a 4-phase overhaul based on deep code audit, fixing 30+ bugs and adding major new capabilities.

Phase 0: Foundation

• Fixed UB in window.rs (Box-based FFI + catch_unwind)
• Removed dead crates: ac-security, ac-browser
• Cleaned 4 unused dependencies, fixed CLI entry point
• input.rs: +8 unit tests (26 → 29 Rust tests)

Phase 1: Agent Pipeline

• Agent ↔ Perception: ScreenContext.capture(with_ocr=True) wired to PerceptionManager
• LLM planning: Agent.plan(goal) via litellm with structured JSON output
• Self-correction: max_retries=3, re-captures context on failure
• Record: speed_factor fix, from_dict() classmethod, ocr_text/window_title in serialization

Phase 2: Security

• Audit log: ISO-8601 timestamps (chrono), error logging, config-driven retention cleanup
• AgentExecutor.enable_security(): one-call guard, click/press auto-checked

Phase 3: Ecosystem

• Plugin system: AutoComputerPlugin ABC + PluginRegistry with lazy entry_points discovery
• CLI: 10 commands (version, see, monitors, serve, run, record, replay, plan, plugins-list)
• Utils: logging, config paths, TOML loader, security config auto-find

Phase 4: Intelligence

• ContextMemory: cross-step state persistence with LLM snapshot formatting
• FlowGeneralizer: recorded flows → parametrized templates ( variables)
• ProfilingAgent: per-step timing breakdown (capture/ocr/execute/verify ms)

P0+P1+P2 Audit Fixes

• AutoComputerError: proper PyO3 create_exception! macro
• Error types: wired capture_err/input_err into actual modules
• Smooth mouse: true coordinate interpolation at ~60fps
• Digit/punctuation key support: parse_key now accepts 0-9, ., -, etc
• Server: dynamic module detection, real flow scanning
• GUI: dynamic module count from API, correct test badge

Quality

• Rust: 29 tests / Python: 27 tests = 56 total, all passing
• Clippy: zero warnings
• Import chain: 19/19 modules verified
• Privacy audit: 10/10 clean