The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from the project's source code, configuration files, and product documentation to generate architectural summaries. This ingestion point represents a surface for indirect prompt injection. The skill mitigates this risk by requiring every claim in the documentation to be backed by verifiable code references (file:line) and by implementing two mandatory human-in-the-loop checkpoints before proceeding to subsequent phases.\n- [COMMAND_EXECUTION]: The skill uses local shell commands to execute its internal utility scripts (scripts/scan_repo.py and scripts/validate_kb.py) for repository auditing and quality verification. It also interfaces with the team-wiki CLI to perform deep code analysis. These executions are scoped to the development environment and are standard for codebase cognitive engineering tools.\n- [DATA_EXFILTRATION]: During the scanning phase, the tool identifies and reads configuration files, including environment templates and service definitions. While this involves reading potentially sensitive files, the scope is limited to architectural extraction for local documentation, with no evidence of remote transmission.

team-wiki-codebase