Skip to content

PRODUCT.md

Latest commit

 

History

History
57 lines (42 loc) · 2.67 KB

PRODUCT.md

File metadata and controls

57 lines (42 loc) · 2.67 KB
Raw

HFQ Code — Product Map

One-liner

HFQ Code is a Windows desktop coding agent: open a workspace, run agent sessions, review diffs, manage skills/MCP/models, with permissions and audit built in.

Phase-1 pages

Page Primary jobs
Home Recent workspaces, continue / rename / delete session, new task
Chat / Session Streaming chat, tool trace, interrupt, rename, token usage totals
Changes File diff list, accept/reject all or selected hunks, open file
Terminal Session-linked terminal, re-run, approval for commands
Tasks Multi-step task list, status, retry
Skills Installed list + 商店 tab (preview / tags / conflict); local folder install; open folder
MCP Server list (persisted in config.json), connect status, tool inventory; live stdio + HTTP tools inject as mcp__*
Models Providers, API keys, default/fallback models
Permissions Tool risk matrix; ask/allow/deny; session overrides
Audit Timeline of tools + approvals; filter by category; copy/export JSONL; restore on session open
Settings Data dirs, session delete, theme/proxy/memory prefs, runtime summary

Beta additions (beyond Phase-1 map)

  • Local memory notes (memory_search / memory_save, %APPDATA%/HFQ-Code/memory)
  • Context compaction for long multi-turn sessions
  • HTTP MCP JSON-RPC client (tools/list + tools/call)

Phase-2 pages (shipped in 1.0)

  • Memory — user/project scope + retrieval UI (BM25; embedding optional later)
  • Import — OpenClaw / Claude / Cursor rules wizard
  • Sub-agents via Chat + Tasks (spawn_subagent, explore/edit/shell)
  • Usage — token / cost dashboard
  • Eval lab: pnpm eval (headless)

1.0.0 notes

完整里程碑见 ROADMAP.md（1.0.5 后主计划）· UI-REDESIGN.md · PHASE2.md · PHASE3.md · 审计 AUDIT.md。

Core user loops

  1. Implement feature — Chat → tools edit → Diff review → Terminal test
  2. Fix bug — paste error → agent investigates → patch → verify
  3. Extend agent — install skill / enable MCP → use in session
  4. Govern — tighten permissions → audit risky shell

Success metrics (engineering)

  • Session crash recovery rate
  • Time-to-first-successful-edit
  • Permission false-block rate
  • Skill load success rate
  • Median tool-round latency