The Agent Skills Directory

[SAFE]: The skill's operational logic involves read-only exploration of the project's directory structure and configuration files (e.g., package.json, tsconfig.json). It does not target sensitive user secrets, private keys, or environment-specific credentials beyond standard reconnaissance for documentation purposes.- [SAFE]: There are no indicators of prompt injection, obfuscation, or persistence mechanisms within the skill's instructions. The skill utilizes standard development tools like git and filesystem globs for its analysis.- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection (Category 8) as it ingests untrusted data from the local codebase being analyzed. Evidence: 1. Ingestion points: The agent reads various source files and manifests across all reconnaissance phases (SKILL.md). 2. Boundary markers: No explicit markers or 'ignore embedded instructions' warnings are provided for the agent when processing these files. 3. Capability inventory: The skill is limited to reading files and writing documentation artifacts (CLAUDE.md), minimizing the risk of automated compromise. 4. Sanitization: The skill does not implement content validation or sanitization for the ingested data. Despite this surface, the overall risk is low due to the lack of dangerous executable capabilities.

codebase-onboarding