The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data such as source code, diffs, and pull requests which can contain malicious instructions.
Ingestion points: The skill processes diffs, pull requests, code modules, and subsystems as specified in the description and process steps.
Boundary markers: There are no explicit technical delimiters or boundary markers mentioned for separating code from instructions, but the process includes a mandatory verification step.
Capability inventory: The skill utilizes integrated CLI tools (doctrine review, backlog new) to perform functions like raising findings and concluding reviews.
Sanitization: The instructions explicitly require the agent to 'Verify each candidate finding' by re-reading cited lines and checking behavioral claims against actual tests or execution rather than recollection, which serves as a manual sanitization step.
[SAFE]: The instructions include a specific security best practice directing the agent to 'Never include any secrets, credentials or API keys directly in the ledger,' effectively mitigating potential data exposure during the documentation phase of a review.

code-review