Skip to content

TODO.md

Latest commit

 

History

History
53 lines (39 loc) · 3.27 KB

TODO.md

File metadata and controls

53 lines (39 loc) · 3.27 KB

Relay — Task Board

See ROADMAP.md for full context and rationale behind each item.

Phase 1 — Visibility ✅

  • Structured failure reports — classify each failure mode (timeout, no commits, tests failed, git error, parse error) into a specific failure_reason stored in DB; surface it in Telegram messages and relay show
  • relay logs <id> command — store raw Claude NDJSON output per issue run; new CLI command to tail it; --follow for live tailing during active fix; --pretty to render as readable text
  • Streaming Telegram progress — wire tool_use events from onEvent into the live progress edit; show current file being edited, tool call count, and running cost estimate
  • Local web dashboard — Bun.serve() at localhost:7842 with WebSocket live updates; views: queue overview, issue detail (log + diff + cost), event timeline; no frameworks, no build step

Phase 1.5 — Web UI as Primary Interface ✅

  • Make Telegram optional — WorkspaceConfig.telegram is optional; all Telegram calls guarded behind ws.telegram?
  • Web UI action buttons — Start Fix / Skip / Accept / Discard / Retry; POST /api/issues/:id/action; shared executeAction() handler used by both Telegram and REST
  • Web UI streaming progress — stream_progress WS events; live tool name, call count, elapsed time, cost in issue detail
  • Web UI triage result view — verdict, confidence, plan, Start Fix / Skip buttons
  • Web UI approval view — diff viewer, cost/token stats, Accept / Discard buttons
  • Web UI failure view — failure reason badge, human-readable hint from FAILURE_HINTS, Retry button
  • Integrations / settings pages — workspaces, projects, notifications, setup wizard (5-step)
  • Preserve CLI debugging — relay logs, relay show, --follow/--pretty flags intact

Phase 3 — Open Source Readiness

  • Distribution — published as @adriandmitroca/relay on npm; bunx @adriandmitroca/relay works zero-install; data dir fixed to ~/.relay/
  • Releases & updates — semantic-release on merge to main; relay update self-update command; silent version check on relay start
  • relay doctor command — validates Bun, claude CLI, gh auth, data dir, database, and all project repo paths; actionable hints on failure
  • X.com launch announcement — write and publish announcement thread positioning Relay as autonomous dev agent; include demo GIF or screenshot of dashboard + Telegram in action

Icebox

Ideas worth revisiting later, not prioritised now.

  • Confidence gating (warn/block Start Fix based on triage confidence score)
  • Test-gate on Accept (run testCommand after fix, block Accept if failing)
  • PR review feedback loop (poll GitHub comments, re-queue with review context)
  • Multi-directory projects (api/ + frontend/ as separate dirs)
  • Plugin architecture (sources and channels as npm packages)
  • Slack notification channel
  • Webhooks for Sentry/Asana (faster than polling, needs tunnel)
  • Auto-accept for high-confidence small diffs (opt-in)
  • Issue batching (multiple errors → one fix)
  • Scheduled batch runs (overnight mode)
  • New source integrations (GitHub Issues)
  • Browser toast notifications via Notification API