The Agent Skills Directory

Command Execution: The skill executes local Node.js scripts (validate-findings.cjs and validate-coverage-ledger.cjs) during the workflow's structured output phase. These scripts perform file system operations to validate the integrity and structure of audit results.
Evidence: Phase 4 in SKILL.md and VALIDATION-AND-REPORTING.md specifies running node <skill-dir>/validate-findings.cjs <output-dir>/findings.json.
Dynamic Execution: The audit process involves running Node.js scripts that use the fs and path modules to process and validate data generated during the audit waves.
Evidence: validate-findings.cjs uses fs.openSync, fs.readSync, and fs.fstatSync to enforce safety limits on input files.
Indirect Prompt Injection Surface: The skill is designed to ingest and analyze untrusted third-party source code, which could contain instructions meant to influence the agent's auditing logic.
Evidence Chain:
Ingestion points: The skill maps and reads the absolute repository root under review (Agent 1a in RECONNAISSANCE.md).
Boundary markers: The skill requires an "OS-enforced sandbox" for running target-controlled code and uses dummy data to isolate the audit from real environments (SKILL.md).
Capability inventory: The agent can read files and execute local validation scripts via node (Phase 4 in SKILL.md).
Sanitization: The validation scripts check for structural schema compliance, text visibility, and specific path safety markers (validate-findings.cjs).

security-audit