Update TODO.md with prioritized improvement roadmap

TODO.md

@@ -1,20 +1,40 @@
 # PRDforge TODO
 
-## High Priority
+## High Impact, Moderate Effort – Do Next
 
-## Medium Priority
+- [ ] Demo GIF/video in README — record 30-45s terminal session (import markdown PRD, read a section with token count, leave a comment, have Claude resolve it). Convert with `vhs` or `asciinema`, place above the fold in README
+- [ ] Token savings dashboard in Web UI — chart showing cumulative tokens saved over time, savings per section read, comparison bar ("full doc: 15,000 tokens vs loaded: 1,320 tokens"). Proof-of-value screen users will screenshot and share
+- [ ] `prd_import_url` tool — fetch a Google Doc, Notion page, or raw URL and import as sections. Support public Google Docs and raw GitHub markdown files to reduce onboarding friction
+- [ ] Section templates — offer project templates beyond blank/SnapHabit: "SaaS MVP", "API Design", "Mobile App", "Infrastructure Migration". Each is a seed SQL or importable markdown showcasing different dependency patterns
 
-(none)
+## Medium Impact, Lower Effort – Quick Wins
 
-## Low Priority
+- [ ] Health check in docker-compose — add `healthcheck` directives for MCP server and UI containers (UI already has `/health`). Fixes startup ordering, makes `docker compose up` more reliable
+- [ ] CONTRIBUTING.md — how to run tests locally, code style expectations, PR process
+- [ ] GitHub repo metadata — description, topics/tags (`mcp`, `claude`, `prd`, `product-requirements`, `ai-tools`, `developer-tools`, `mcp-server`), website URL
+- [ ] Multi-arch Docker builds — add `linux/amd64,linux/arm64` to GH Actions build for Apple Silicon support
+- [ ] `prd_diff_sections` tool — unified diff between two revisions of a section, avoids loading both and diffing manually
 
+## Medium Impact, Higher Effort – Plan For These
+
+- [ ] Webhooks / notifications — POST to a configured URL on section update for Slack/Discord/pipeline integration
+- [ ] Export to Google Docs / Notion — push assembled PRD back to where stakeholders read it (Notion API is more reasonable than Google Docs)
+- [ ] Section-level permissions / locking — advisory lock with timestamp and owner field so two Claude sessions don't clobber each other
+- [ ] Embeddings-based dependency suggestions — store embeddings in pgvector, compute cosine similarity to catch semantic relationships that FTS keyword overlap misses
+- [ ] Version tagging / snapshots — `prd_tag_version` tool that snapshots all current section revision numbers under a named tag for milestone tracking
+
+## Lower Priority – Worth Tracking
+
+- [ ] Conflict detection — optimistic concurrency with revision check on write (warn/fail if section was edited since last read)
+- [ ] Section ordering visualization — Gantt-style view based on `blocks` dependency type for timeline sections
+- [ ] Prometheus `/metrics` endpoint — request counts, latency histograms, DB connection pool stats
+- [ ] claude.ai MCP marketplace listing — be ready when Anthropic opens MCP server discovery
 - [ ] Replace heuristic markdown import parser with `markdown-it-py` AST parser
 - [ ] Add WebSocket push to UI for real-time updates when sections change
 - [ ] Support `### ` (h3) splitting in import for nested section hierarchies
 - [ ] Add `prd_merge_sections` tool (combine two sections into one)
 - [ ] Add `prd_reorder_sections` tool (bulk sort_order update)
 - [ ] Export as PDF via headless browser
-- [ ] Add Prometheus `/metrics` endpoint to MCP server (request count, latency histograms)
 - [ ] UI: keyboard navigation (j/k to move through sections, Enter to select)
 
 ## Done