The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses common shell utilities including find, grep, and wc to map the codebase and search for specific code patterns. These commands are used for discovery and metrics within the project directory.
[DATA_EXFILTRATION]: No network operations or attempts to access sensitive system files (such as SSH keys or environment variables) were detected. The skill writes its final report to a local directory (.audits/).
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface because it ingests and processes untrusted data from the user's source code.
Ingestion points: All source files discovered during Phase 1.
Boundary markers: Absent; the skill does not use specific delimiters to separate code content from agent instructions.
Capability inventory: Shell execution (find, grep) and file-write access to the .audits/ directory.
Sanitization: Absent; content is analyzed directly as text. Despite the lack of boundaries, the risk is minimal as the instructions guide the agent toward structural auditing rather than executing logic from the files.
[REMOTE_CODE_EXECUTION]: There are no instructions to download external scripts or execute code from remote repositories.

architectural-analysis