The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code, which serves as a surface for indirect prompt injection. Maliciously crafted files under review could contain hidden instructions or adversarial text designed to influence the agent's findings or behavior.
Ingestion points: The skill accepts a [target-file-or-directory] argument and uses the Read, Grep, and Glob tools to load external file content into the agent's context.
Boundary markers: The skill includes a dedicated 'Prompt Injection Safety Notice' that explicitly instructs the agent to treat reviewed code as inert text and to disregard any instructions found within code comments, strings, or variable names.
Capability inventory: The skill's environment is restricted to the Read, Grep, and Glob tools. It lacks network access, file-writing capabilities, or command execution tools, which severely limits the potential impact of an injection attack.
Sanitization: The skill uses static analysis patterns and does not execute, evaluate, or interpret the code it reviews.

secure-code-review