The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub PR reviews and comments to drive automated code modifications and PR updates.
Ingestion points: Extracts review thread status and comment content via gh api graphql and monitors external review activity through a shared polling mechanism.
Boundary markers: The orchestration logic lacks explicit instructions to ignore adversarial directives embedded in the external data; it relies on the safety protocols of the sub-agents it dispatches (pr-reviewer and implement-suggestion).
Capability inventory: The skill possesses significant capabilities, including automated code implementation, execution of git push operations, modification of PR metadata (gh pr edit), and the ability to dispatch additional sub-agents (Task or Agent tools).
Sanitization: While the skill utilizes shell heredocs for updating PR descriptions, it does not implement content-level sanitization to filter out potential prompt injection attacks contained within the PR threads it processes.

review-loop