Finding Prior Art

---

name: finding-prior-art description: Find what the org already knows before building or deciding. Use when starting non-trivial implementation or design work, when asked "have we done this before", "why is X like this", or "has anyone hit this error", when a decision smells like it was already made once, or before proposing an approach someone may have already rejected.

Finding prior art

The org's hard-won knowledge — decisions, gotchas, incident lessons, owners — is indexed alongside its code. Code alone won't show it, and rebuilding a rejected approach is the most expensive way to rediscover a decision.

Workflow: search, then fetch

1. bonez:search with the question as intent. Don't restrict entities to code when the question is a "why" or a "have we":

{"query": "retry strategy for webhook delivery", "entities": ["knowledge", "memory", "ticket", "pr", "doc"], "limit": 10}

1. bonez:fetch the top 2–3 handles before quoting anything:

{"ref": "~a1b2c3d4"}

Snippets are bait, not evidence. The fetched record carries provenance and temporal status (superseded? stale? what evidence backs it?) — prefer it over the snippet whenever they disagree.

Scope

Omit repos to search the whole org — that is the default and the point: prior art usually lives in a repo you are not looking at. Pass the working repo only when the question is genuinely repo-local.

Traps

• [INCOMPLETE] markers and unavailable-arm notices mean coverage gaps, not absence. Search is fused from several arms; when one arm reports unavailable, results are partial. Say so explicitly instead of presenting a partial answer as complete.
• Absence is not proof. "No results" never proves the org hasn't done it — it proves nothing was indexed under your phrasing. Vary the phrasing once (the domain word, the error text, the ticket vocabulary), then state coverage honestly: "nothing indexed under these terms", not "we've never done this".
• Prior art expires. A fetched decision may be superseded — check its temporal status before presenting it as current policy.

When to stop

Two searches with varied phrasing plus fetched top hits is proportionate diligence for most tasks. Escalate to bonez:query traversal only when you have a concrete seed entity and need its neighborhood — see querying-the-graph.

Hand-offs

• Relationship and impact questions ("what depends on this") → impact-analysis.
• "Who knows about this" → who-owns-what.
• Presenting what you found → citing-bonez-sources.