The Agent Skills Directory

[SAFE]: The skill incorporates a 'Re-verification Pass' as a core component of its workflow, which requires the agent to confirm findings against the actual source code at specific line numbers before reporting. This design significantly reduces the likelihood of hallucinations or false findings during the code review process.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted source code files, which constitutes a standard ingestion surface for indirect prompt injection.
Ingestion points: Source code files are identified via the Glob tool and read using the Read tool in SKILL.md.
Boundary markers: The instructions lack specific boundary markers or instructions to the sub-agent to ignore potential prompts embedded in the code comments or strings.
Capability inventory: The reviewing sub-agent has the capability to modify files via TodoWrite, and the skill uses Glob and Read for file discovery and access.
Sanitization: No sanitization or escaping mechanisms are described for the code content before it is analyzed by the agent.

code-review