The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script scripts/run_audit.py executes shell commands and binary executables defined in agents.json and site.json. It uses asyncio.create_subprocess_exec for agent processes and asyncio.create_subprocess_shell for resetting the test environment.
[COMMAND_EXECUTION]: The inventory script scripts/build_inventory.py executes git commands via subprocess.run to extract version information from the audited repository.
[DYNAMIC_EXECUTION]: The skill launches independent agent processes with broad system tool access (Grep, Bash, Curl) configured through dynamic templates and configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill audits untrusted application code, presenting a vulnerability surface where malicious code could attempt to influence auditing agents. Mandatory evidence chain: 1. Ingestion points: scripts/build_inventory.py (via ast.parse) and agents reading the target directory. 2. Boundary markers: Explicit instructions in SKILL.md (Global Rule) warning agents to ignore self-safety claims. 3. Capability inventory: Subprocess and shell execution capabilities in run_audit.py and build_inventory.py. 4. Sanitization: Minimal sanitization of audited content before presentation to agents.
[EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions to download semgrep rules from the official vendor repository (https://github.com/frappe/semgrep-rules) and install the semgrep utility.

deep-app-audit