Idea: — Health command output should be more readable — cleaner formatting and visual hierarchy
Idea: — Add command to log bugs to task backlog — bug tasks can be grouped together into a single sprint task
Convert MCP server to prepare/apply pattern (API-to-local)
Consolidate MCP server source into papi-v1 repo
Idea: — Audit and update prompts.ts for API-to-local transition
Idea: — idea tool fails on feature branches — can't checkout main
Idea: — board_view output too large (95.9KB) — needs pagination or filtering
Idea: — Research: v2 retrospective learnings applicable to v1 — upstream learning, living documents, validation gates
Idea: — Research: Streams vs Sprints — evaluate v2 stream-based work model for v1
Idea: — Research: API-to-local (prepare/apply) — evaluate removing Anthropic API calls from MCP server
Idea: — Map out external user journey — what does a friend need to do step-by-step to get PAPI running on their project?
Idea: — Cross-project contract enforcement — shared CLAUDE.md context for collaborating Claudes. Two developers using Claude Code on separate repos have zero shared context. Each Claude has different memory about field formats, column constraints, parser expectations, and implementation conventions. This causes data mismatches (e.g. complexity S/M/L vs Small/Medium/Large, AD regex attribute order, displayId vs id). Current AD conflict detection catches decision-level contradictions but misses implementation-level incompatibilities. PAPI should solve this — either via a shared context layer that both Claudes read (cross-repo CLAUDE.md section, shared contract file), or via schema validation at the write boundary (reject non-conforming data before it enters Supabase). This is the same institutional memory gap identified in the memory guide — the product should be the bridge, not the humans.
MCP adapter-pg architecture docs and test plan for Kieran
Repair .papi/ data for Supabase migration
Add alpha testing docs and setup script
Commit brain.json API handler changes
Add PAPI Brain analytics dashboard
AD: Write path architecture — Go API vs MCP-only vs hybrid
Recommendation acceptance tracking — feedback loop for strategy recommendations