The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted backend code and diffs, creating a surface for indirect prompt injection attacks where malicious instructions could be embedded in source code or comments to influence the agent's behavior.
Ingestion points: In SKILL.md, the workflow requires the agent to "establish the requested review scope and inspect the relevant diff or files" specifically under the api/ directory.
Boundary markers: The skill defines clear review rules in references/ and a specific output format (P0-P3 severities), which helps constrain agent behavior, but it lacks explicit instructions to ignore or sanitize natural language prompts that might be hidden within the code comments or diff content being reviewed.
Capability inventory: The skill is configured for analysis and reporting; the provided files do not request additional tool permissions like network access or shell execution, though it operates within the context of an agent that may possess such tools.
Sanitization: No sanitization, escaping, or filtering mechanisms are described for the content being reviewed before it is processed by the agent.

backend-code-review