The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository files, build manifests, and code metadata, creating a surface for indirect prompt injection.\n- Ingestion points: Scanning logic in references/checklist.md and references/java-build-and-dependency-map.md reads content from the workspace.\n- Boundary markers: references/graph-operation-gates.md enforces path resolution within the repository root to prevent scope escape.\n- Capability inventory: The instructions limit the tool to read-only mapping and rendering, with no subprocess or network capabilities described in the skill body.\n- Sanitization: references/project-grounding.md and references/java-build-and-dependency-map.md explicitly forbid the capture or output of secrets, tokens, and sensitive configuration data.

repo-map