The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill checks and utilizes the DATABASE_URL environment variable to test database connectivity. Passing this variable directly to the psql command line exposes potential credentials to system process logs and shell history.
[PRIVILEGE_ESCALATION]: The skill includes a --fix flag that executes chmod +x on shell scripts within the .claude/hooks/ directory to modify execution permissions.
[DYNAMIC_EXECUTION]: The skill performs runtime execution of build scripts (npm run build) and Python code via uv run python during the audit and fix phases.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted configuration and metadata from project files to generate a health report.
Ingestion points: Reads contents from .claude/skills/SKILL.md, .claude/agents/*.md, and .claude/settings.json (SKILL.md, SKILL.md).
Boundary markers: Absent; the skill does not use delimiters or instructions to prevent the agent from obeying instructions embedded within the audited files.
Capability inventory: Includes shell command execution, file permission modification, and database querying across multiple phases.
Sanitization: Absent; extracted strings are processed using grep and sed without escaping before being interpolated into the final report.

mot