The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon code review feedback from 'External Reviewers', which is an untrusted data source that could contain malicious instructions or misleading code suggestions.
Ingestion points: External feedback received during the code review cycle as described in Part 2 of SKILL.md.
Boundary markers: The skill explicitly defines a '6-Step Response Pattern' which acts as a manual boundary, requiring the agent to restate, verify, and evaluate feedback before implementation.
Capability inventory: The skill utilizes git for diffing and gh (GitHub CLI) for API interactions, along with the ability to dispatch subagents.
Sanitization: Implementation is gated by technical verification ('Check against codebase reality') and skepticism protocols for external feedback, reducing the likelihood of accidental execution of injected instructions.
[COMMAND_EXECUTION]: The skill uses local shell commands to facilitate the review process.
Evidence includes git rev-parse, git diff, and gh api calls in SKILL.md and code-reviewer.md.
These commands are used for retrieving commit SHAs and repository metadata, which are standard development operations.

code-review