The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Bash to run established security tools such as semgrep, gitleaks, and trivy. Execution of active scanning tools (like nmap or sqlmap) is strictly controlled by the appsec-active-scan-guard.js hook, which requires a valid Rules of Engagement (ROE) document.
[DATA_EXFILTRATION]: The skill is configured with forbidden-tools: WebFetch to prevent direct outbound network calls. Furthermore, it employs multiple defensive layers, including the appsec-secret-access-guard.js hook to block reading production secrets (e.g., .env, .pem files) and the appsec-secret-redaction.js hook to scrub credentials from all agent outputs.
[DYNAMIC_EXECUTION]: The orchestrator dispatches local workflows and sub-agents. This runtime execution is secured by the "Execution Preview Contract" (§16.13), which mandates a human-reviewed plan and cryptographic verification of the execution spec before the Workflow tool is permitted to launch.
[INDIRECT_PROMPT_INJECTION]: The skill handles potentially untrusted data from security scan results. It mitigates injection risks by using structured YAML schemas for findings, employing a specialized triager agent for normalization, and enforcing strict schema validation through the appsec-finding-schema-prewrite.js hook.

appsec-security-orchestrator