docs: mark v0.4 roadmap items complete

Author: igmagollo

docs/ROADMAP.md

@@ -0,0 +1,96 @@
+# suggestion-box Roadmap
+
+## v0.2 — Foundation & Trust
+> Goal: fix adoption blockers, make the project safe to recommend to strangers
+
+- [x] Fix issue #57 — `init` modifying global `~/` (adoption blocker)
+- [x] Add `--dry-run` flag to `init` that shows what would be written without touching anything
+- [x] Add config schema validation with clear error messages on startup
+- [x] Add per-session rate limiting to prevent runaway agent spam
+- [x] Add minimum quality filter rejecting vague/too-short submissions before hitting the queue
+- [x] Store session ID on every feedback entry now, even before transcript capture exists
+- [x] Capture dependency/package versions on every feedback entry for regression analysis
+- [x] Tie feedback entries to git SHA at submission time; auto-flag when related files change
+- [x] ~~Add `CHANGELOG.md`~~ (covered by release-drafter)
+- [x] Add `CONTRIBUTING.md`
+
+---
+
+## v0.3 — Developer Experience
+> Goal: make it feel polished and serious to first-time visitors
+
+- [x] Add a demo GIF/screen recording to the README showing the full submission → triage → publish loop
+- [x] Add a "Built with suggestion-box" section to the README linking to actual agent-filed issues
+- [x] Give cross-agent compatibility (Claude Code + Codex + OpenCode) its own README section
+- [x] Build a `suggestion-box doctor` command to verify environment health (`gh` CLI, embedding model, write permissions)
+- [x] Ship pre-authorized tool list in Claude Code config for safe operations (`submit`, `upvote`, `list`, `status`, `triage`) — keep `publish_to_github` and Linear outside the allowlist
+- [x] Make feedback categories configurable per project
+- [x] Create a README badge for adopters to add to their own repos — passive distribution
+- [x] Publish a Homebrew formula
+
+---
+
+## v0.4 — Review Workflow
+> Goal: make human triage fast enough to actually happen
+
+- [x] Fix CLI/MCP server DB lock contention — CLI commands fail when server is running (#149)
+- [x] Make `init` create the database so `doctor` passes immediately (#150)
+- [x] Fix GitHub dedup for custom categories — `searchExistingIssues` only handles hardcoded title prefixes (#148)
+- [x] Auto-triage by vote count to surface high-signal items without manual review
+- [x] Build interactive `suggestion-box review` TUI with keyboard shortcuts (`p`=publish, `e`=edit, `d`=dismiss, `s`=skip)
+- [x] Add a `/review` slash command that kicks off the pre-triage agent flow conversationally inside Claude Code
+- [x] Add a pre-triage MCP tool that groups entries, deduplicates against target repo/Linear, enriches with impact summary, and moves results to a `pending_review` queue
+- [x] Inject a digest of top-voted unresolved friction into SessionStart so new agents arrive pre-warned
+- [x] Add a webhook option that pings Slack/Discord when high-vote items arrive
+
+---
+
+## v0.5 — Transcript & Context
+> Goal: give the triage agent enough context to reason well
+
+- [ ] Leverage Claude Code's native session transcripts at `~/.claude/projects/<project-hash>/<session-id>.jsonl`
+- [ ] Add a PostToolUse hook that appends tool calls/responses to `.suggestion-box/transcripts/<session-id>.jsonl` for non-Claude Code agents
+- [ ] Document the transcript capture integration path clearly
+- [ ] Add per-session agent quality scoring based on feedback vote patterns — use it to measure SessionStart prompt quality
+
+---
+
+## v0.6 — Publisher Ecosystem
+> Goal: make suggestion-box work wherever teams already live
+
+- [ ] Linear publishing support (team + project required)
+- [ ] Design generic publisher backend interface to support future targets
+- [ ] Auto-detect issue templates from target GitHub repo, use local fallback when none exist
+- [ ] Build reverse sync — poll GitHub/Linear and auto-resolve local feedback when linked issues close
+- [ ] Track resolver attribution and build resolution rate metric per target repo/project
+- [ ] Design publisher plugin interface so the community can build Jira, Notion, Azure DevOps targets
+
+---
+
+## v0.7 — Distribution & Discoverability
+> Goal: get in front of the community while the MCP ecosystem is still sparse
+
+- [ ] Publish to `modelcontextprotocol/registry` at `io.github.igmagollo/suggestion-box`
+- [ ] Add `server.json` with proper tags (`feedback`, `ai-agents`, `claude-code`, `developer-tools`)
+- [ ] Add a `suggestion-box check` CI command that fails/warns based on unresolved high-vote items
+- [ ] Write post framed as "my agents filed their own GitHub issues" targeting Claude Code / Codex communities
+- [ ] Write a longer thought leadership piece on "agents as contributors" — stake out the category
+
+---
+
+## v1.0 — Ecosystem Signal
+> Goal: become the quality feedback layer for the MCP ecosystem
+
+- [ ] Design opt-in telemetry layer for anonymized MCP tool friction signals only (target identifier + category + timestamp — no content, no repo, no identity)
+- [ ] Add `suggestion-box telemetry on/off` command with clear status in `suggestion-box doctor`
+- [ ] Open-source the aggregation server
+- [ ] Build public dashboard of systemic friction patterns across opted-in users
+- [ ] Build public leaderboard of most-reported MCP tool friction across the ecosystem
+
+---
+
+## Post-v1.0 — Sustainability
+> Goal: build a model that funds continued development
+
+- [ ] Design B2B analytics tier (MCP tool author monitoring) — avoid architecture decisions that foreclose this path
+- [ ] Introduce MCP tool author dashboard: "here is how agents experience your tool across the ecosystem"