The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and analyzing untrusted application source code, creating a surface for indirect prompt injection.
Ingestion points: Processes entire application repositories, including product documentation, configuration files, and source code (SKILL.md).
Boundary markers: Includes a specific subagent brief that instructs the AI to ignore any instructions encountered within the audited codebase (SKILL.md).
Capability inventory: The skill possesses capabilities to execute shell commands for scanning secrets and dependencies, write audit reports to the local file system, and perform network requests (SKILL.md).
Sanitization: Implements a multi-phase audit process where findings are subjected to adversarial verification by independent skeptic subagents to filter out invalid or malicious results (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill performs network operations to interact with vendor infrastructure and well-known services.
Evidence: Executes a version check via curl to foundry.thehorizonfoundry.com, which is a vendor-owned domain (SKILL.md).
Evidence: References the canonical report schema hosted on the vendor's GitHub repository (SKILL.md).
Evidence: Utilizes npx ajv-cli for report validation, which may involve downloading the package from the npm registry (SKILL.md).
[COMMAND_EXECUTION]: The skill uses local shell commands to automate mechanical audit tasks.
Evidence: Executes tools like npm audit, gitleaks, trufflehog, and ajv-cli to perform dependency analysis, secret scanning, and schema validation (SKILL.md).

production-audit