| name | prime |
| description | Orient agents to a repository by creating concise scoped guidance from verified source structure, commands, conventions, and code-navigation paths. |
Repository Priming
Prime this repository for downstream subagents. Use this before large multi-agent work, onboarding a new repo, or after major folder/architecture changes.
Workflow
- Inspect manifests and source roots, excluding generated, vendor, ignored, private, and
credential paths.
- Re-derive repository facts from source; treat existing context prose as an
unverified hint.
- Create or update concise
AGENTS.md files only at meaningful ownership boundaries.
Include scoped commands, code-navigation guidance, conventions, guardrails, and a
staleness note.
- Keep
CLAUDE.md as a one-line @AGENTS.md pointer when both formats are needed.
- Create or update a root
.claudeignore for high-volume generated/vendor paths.
- Re-read changes and verify that no product source, secrets, or private inputs were
written.
For very large monorepos, delegate at most three non-overlapping source roots in
parallel. Fall back to a sequential main-thread scan when delegation is unavailable.
Expect: a source map, list of context files created/updated (each carrying ## Scoped Commands + ## Code Navigation), a root .claudeignore, ignored folders, and verification that no source code or secrets were changed.
Evidence Receipt
Return spk.evidence/v1 with scanned roots, context artifacts, corrections to stale
claims, exclusions, verification, risks, and next action.
Guardrails
- Do not change product source or scan private/generated/vendor paths.
- Preserve human-authored intent while re-verifying factual claims from source.
- Keep guidance scoped, concise, and free of secrets or volatile duplicated facts.
