The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a node -e inline script in SKILL.md to update its run-history.json file. This involves dynamic script generation and local file system operations to maintain tool state.
[INDIRECT_PROMPT_INJECTION]: As a security scanning tool, the skill ingests untrusted data from the project source for analysis.
Ingestion points: Reads code from src/, config/, and .env files as specified in SKILL.md and references/compliance-agent-commands.md.
Boundary markers: No explicit instructions to delimit or ignore embedded instructions in scanned data were identified in the prompt logic.
Capability inventory: Execution of external security tools (npm audit, trufflehog, docker), local file system writes to run-history.json, and coordination of agent tasks.
Sanitization: No specific sanitization or escaping of the ingested code is described before processing.
[CREDENTIALS_UNSAFE]: The evaluation suite in evals/security-testing.yaml contains hardcoded secrets and connection strings within its test fixtures.
Evidence: Includes ADMIN_PASSWORD = 'admin123', API_KEY = 'sk-1234567890abcdef', and a DATABASE_URL with embedded credentials. These are used as test data for scanning and are not intended for production use.
[EXTERNAL_DOWNLOADS]: The skill references and utilizes several external security tools and repositories.
Findings: Downloads the TruffleHog scanner from its official source and uses the OWASP ZAP Docker image. It also references well-known security services such as SonarQube, Semgrep, and Snyk. These references are documented neutrally and correspond to established security tooling.

security-testing