The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted codebases to identify security vulnerabilities.
Ingestion points: Source code files from target repositories (SKILL.md).
Boundary markers: Explicit instructions to generalize patterns into structured rules and use a verification pass (Step 2, references/verification.md).
Capability inventory: Access to Read, Write, Grep, and specific Bash commands for Semgrep.
Sanitization: Implementation of an independent verification pass and utility scripts for validating finding records (references/validate-findings.mjs).
[COMMAND_EXECUTION]: The skill uses specific, restricted Bash commands to run the Semgrep analysis engine, ensuring that only pre-approved auditing tasks are performed (SKILL.md).

variant-analysis