Spec Retrieval (Prior Work)

When a spec is created, Monday Morning searches finished specs for similar work and writes the matches to prior-work.md in the spec folder. Agents read it before planning, so they build on what exists instead of re-deriving it.

Note

This is plumbing for agents, not a screen. There’s no Prior Work panel in the app.

When it runs

At spec creation, on /mm:retrieve-prior-work (or mm_retrieve_prior_work), and whenever an agent calls mm_search.

How matching works

Semantic similarity over an embedding index, falling back to word overlap, with boosts for shared tags and recency. Matches below the similarity threshold are dropped, and the file says so rather than listing weak results.

The indexes live in .mm/index/: specs.json for spec metadata and ledger/ for embeddings. Rebuild or check them with mm_index.