The Agent Skills Directory

[DATA_EXFILTRATION]: Access to sensitive file paths. The skill is designed to read .env files to check for the presence of API keys and verify 'Connections' coverage. It also accesses project memory files (e.g., ~/.claude/projects/<id>/memory/MEMORY.md) to score 'Context'.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests and processes untrusted data from multiple project files to generate scores and recommendations.
Ingestion points: The skill reads content from CLAUDE.md, MEMORY.md, and various markdown files within the .claude/ directory.
Boundary markers: Absent. The skill does not use specific delimiters or instructions to prevent the agent from obeying instructions that might be embedded within the files being audited.
Capability inventory: The skill utilizes file read capabilities across the project directory and has the ability to write a final report to the audits/ folder.
Sanitization: Absent. The skill performs word counts and pattern matching directly on the ingested text without escaping or filtering.
[COMMAND_EXECUTION]: Dynamic Context Injection. The skill uses the !command`` syntax (!date +%Y-%m-%d) to execute a shell command at load time to populate the date field in the audit report. While this specific command is benign, it demonstrates a shell execution point.

audit