The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as code for review and architecture descriptions, which presents a surface for indirect prompt injection.
Ingestion points: Processes user-provided code, Jira stories, and architecture descriptions in SKILL.md and references/code-review-checklists.md for security evaluation.
Boundary markers: The skill explicitly instructs the agent to treat external text as untrusted data and use fence markers to prevent instruction following in its 'safety calibration' section.
Capability inventory: Includes file writing (approvals/secops-security.md), execution of security scanners (Semgrep, Trivy, Gitleaks, OSV-Scanner), and network research capabilities via MCP and WebSearch.
Sanitization: The skill includes instructions for input sanitization and output filtering in the context of AI/LLM security defense.
[EXTERNAL_DOWNLOADS]: Fetches security tools and intelligence from trusted organizations and well-known services.
Evidence: references/supply-chain-and-scanning.md contains instructions to download tools like Syft from Anchore's official GitHub repository, Trivy from Aquasecurity's repository, and SLSA verifier from the SLSA framework repository.
Evidence: SKILL.md references documentation and vulnerability data from established sources including NIST (NVD), CISA, MITRE (ATT&CK/CWE), and OWASP.
[COMMAND_EXECUTION]: Provides instructions for executing various security scanning, build, and identity tools.
Evidence: references/supply-chain-and-scanning.md contains command-line instructions for running syft, grype, semgrep, trivy, osv-scanner, slsa-verifier, and cosign.
Evidence: references/infrastructure-security.md and references/owasp.md include configuration for Docker, Kubernetes network policies, and Java-based security controls.

secops-engineer