メニュー
Performs semantic and filtered searches across DNA truth artifacts. Use when looking for relevant decisions, constraints, or contracts before implementing features, resolving ambiguity, or understanding why something works the way it does.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | searching-artifacts |
| description | Performs semantic and filtered searches across DNA truth artifacts. Use when looking for relevant decisions, constraints, or contracts before implementing features, resolving ambiguity, or understanding why something works the way it does. |
DNA uses vector embeddings for semantic search. Queries are natural language, not keywords.
When exploring an unfamiliar area:
dna search "how does authentication work in this system"
dna search "what are the rules for billing"
Natural language queries work well because the embedding model understands meaning, not just word overlap.
When you know what type of truth you need:
dna search "payment processing" --kind contract
dna search "password requirements" --kind constraint
dna search "order total calculation" --kind algorithm
Kind filters narrow the search space. Use them when you want a specific category of truth.
When you want to browse rather than search:
dna list --kind intent
dna list --kind constraint --label domain=auth
Listing does not use semantic similarity -- it returns all matching artifacts.
When you know the ID (from a previous search or reference):
dna get <artifact-id>
Search results include a similarity score (0.0 to 1.0, cosine similarity).
| Score | Interpretation |
|---|---|
| > 0.8 | Strong match. Almost certainly relevant. |
| 0.6 - 0.8 | Good match. Read and assess applicability. |
| 0.4 - 0.6 | Weak match. May be tangentially related. |
| < 0.4 | Noise. Probably not relevant. |
These thresholds depend on the embedding model. The local model (bge-small) tends to produce lower scores than OpenAI models for equivalent relevance.
Cast a wide net:
dna search "<the task or feature>" --limit 10
Read everything with score > 0.5. You are building a picture of what truth exists in this area.
Targeted search:
dna search "<the specific thing you are changing>" --kind contract --limit 5
dna search "<the specific thing you are changing>" --kind constraint --limit 5
Focus on contracts and constraints -- these are the artifacts most likely to conflict with a change.
Search for glossary terms when the meaning of a concept is unclear:
dna search "<the ambiguous term>" --kind glossary
If no glossary artifact exists, the term is undefined. Consider capturing a definition.
No results means either:
Try alternative phrasings before concluding no truth exists:
dna search "user authentication"
dna search "login flow"
dna search "session management"
If multiple phrasings return nothing, the area is genuinely uncovered.
Interactive conversation for deliberate truth changes. Acts as a truth guardian that searches DNA, surfaces conflicts, debates proposals, and only applies changes with explicit human reasoning. Use via text or voice mode in Claude Code.
Assembles DNA truth artifacts into a self-contained specification that an autonomous agent can execute against without further human guidance. Use when handing off a well-defined deliverable to an agent for extended autonomous execution.
Consults DNA truth artifacts during task planning to understand system identity before designing an implementation approach. Use at the start of any non-trivial task before choosing an architecture or decomposing work into steps. Searches for intents, contracts, and constraints that shape the solution space.
Creates escalation artifacts that define when an agent should stop autonomous work and involve a human. Use when encoding decision boundaries for agent autonomy -- the guard rails on what agents can decide alone.
Creates tradeoff artifacts that encode which concern wins when competing concerns clash. Use when two valid goals, constraints, or intents conflict and agents need to know which to prioritize.
Stores truth artifacts in DNA with correct kind, format, metadata, and semantic context. Use when codifying decisions, contracts, intents, or constraints that should persist across code changes, team transitions, and framework evolutions.