The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external code and PR metadata, creating a surface for indirect prompt injection. ● Ingestion points: data enters via git diff, git log, and file read operations in the Preflight and Audit steps. ● Boundary markers: present; the skill explicitly instructs the agent to treat rationales as claims and to actively refute correctness. ● Capability inventory: the skill has access to terminal commands (git) and file system reads. ● Sanitization: absent; the skill relies on the agent's logical analysis rather than automated filtering.
[COMMAND_EXECUTION]: The skill utilizes several Git commands to gather information for the review. ● Evidence: SKILL.md preflight steps include git merge-base, git status, git log, git diff, and git ls-files. These are used strictly for read-only analysis as defined by the skill's IRON LAW.

code-review