docs-source-harvest

Files (4)

• SKILL.md
• smoke.yaml · evals/
• REFERENCE.md · references/
• collect.sh · scripts/

SKILL.md (first lines)

# Docs source harvest

## Inputs expected

- Product/library name and version (if known).
- Question focus: setup, API usage, config, auth, migration, limits.

## Retrieval strategy

1. Prioritize official docs and reference pages.
2. Prefer first-party vendor-provided skills/guides when available for that tool/platform.
3. Prefer version-matched docs for the caller's stack.
4. Use secondary sources only when official docs are silent.
5. Capture exact URLs for every material claim.

## Extraction checklist

- Minimal "how-to" path for the request.
- Required flags/fields and defaults.
- Version constraints and incompatibilities.
- Security/auth caveats.

## Output requirements

- `summary`: direct answer in 1-2 sentences.
- `findings`: concise bullets with one citation each.
- `artifacts`: optional snippet blocks (commands/config examples).
- `confidence`: based on source quality and version match.

## Guardrails

- No uncited recommendations for critical configuration.
- Explicitly call out gaps when authoritative docs are missing.
- If product/version/scope is incomplete, return `insufficient_evidence` and callback to parent for missing identifiers before final recommendations.

From bryanbarton525/prism, MIT.

More from bryanbarton525/prism