The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute standard shell commands for repository analysis, including git status, git diff, git rev-parse, grep, and cat (optionally using sed or head for chunking).
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes potentially untrusted data from the codebase (e.g., code diffs, finding descriptions, and source code content) and interpolates it into prompts for an external verification model (Codex).
Ingestion points: Untrusted data enters the agent context through the relevant_diff, original_finding_text, and code content retrieved via cat or grep commands.
Boundary markers: The skill uses markdown code blocks (e.g., ```diff and ```) to delimit external content within the verification prompt.
Capability inventory: The skill possesses capabilities for file system access, shell command execution, and interaction with an external analysis model.
Sanitization: The skill instructions explicitly require the redaction of secrets/tokens from logs and finding texts, but there is no specific logic to sanitize prompt injection markers or control sequences embedded within the source code being analyzed.

seek-verdict