The Agent Skills Directory

[SAFE]: The skill performs read-only analysis of project files such as package manifests, configuration files, and directory trees. These operations are restricted to reconnaissance and documentation generation, with no evidence of network exfiltration, unauthorized file modifications, or credential harvesting.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes content from untrusted codebases.
Ingestion points: Manifest files (e.g., package.json, go.mod), configuration files, and source code content.
Boundary markers: Instructions limit the agent to structural reconnaissance (using glob and grep) rather than exhaustive reading of all files, which mitigates exposure to malicious instructions embedded in deep source code.
Capability inventory: File system read access for analysis and file write access for creating guide artifacts (CLAUDE.md). No network or shell execution tools are used.
Sanitization: The skill focuses on pattern matching and technical metadata, though it lacks explicit sanitization of the values extracted from the codebase before they are included in generated reports.

codebase-onboarding