The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/gates.py script executes shell commands found in settings.json and settings.local.json using subprocess.run(shell=True). This is designed to test if security hooks are working, but it means the skill executes arbitrary command strings defined in local configuration.
[DATA_EXPOSURE]: The skill accesses sensitive local data for auditing purposes. scripts/collect.sh searches for hardcoded secrets (API keys, private keys) using grep and gitleaks. scripts/usage.py reads Claude Code session transcripts from ~/.claude/projects/ to generate usage telemetry.
[COMMAND_EXECUTION]: The audit process relies on multiple shell scripts and Python scripts that execute system commands like find, git, grep, and ls to gather repository facts.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect injection as it ingests and processes untrusted data from the repository and previous transcripts.
Ingestion points: Reads all markdown and source files in the repository (scripts/collect.sh) and all session transcript files (scripts/usage.py).
Boundary markers: None; the skill parses file contents and transcripts directly for inventory and telemetry.
Capability inventory: Includes arbitrary shell command execution via gates.py and file writing for report generation in context/audits/.
Sanitization: None; it uses raw matches from grep and transcript logs to populate its reports.

system-audit