The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow, workflows/triage.md, involves the agent reading and analyzing snippets from external code repositories to identify and rank vulnerabilities. This ingestion of untrusted code creates a surface for indirect prompt injection, where an attacker could embed instructions in a repository to influence the agent's triage results. However, the skill explicitly acknowledges this risk in references/bounty_patterns_2024_2026.md, providing detection rules (Pattern P8) to identify and mitigate such injections during the scanning process.
[COMMAND_EXECUTION]: The orchestration script scripts/sast_scan.sh and the multi-tool workflow workflows/multi_tool_scan.md execute several shell commands to invoke SAST utilities like Semgrep, Bandit, and CodeQL. This is the intended behavior for an orchestration skill and is restricted to well-known, established security tools.
[EXTERNAL_DOWNLOADS]: The documentation and scripts reference official sources for installing dependencies, including GitHub, PyPI, NPM, and the Go module registry. These references target official tool repositories (e.g., github.com/github/codeql) and represent standard, safe practices for tool setup.

sast-orchestration