The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes plan files and code diffs as part of its core review function, which creates a surface for indirect prompt injection.
Ingestion points: The agent reads plan files in ai-workspace/plans/ and code changes identified via git diff.
Boundary markers: There are no specific instructions to delimit this untrusted content or warn the agent about embedded instructions.
Capability inventory: The skill can perform git commit, create GitHub issues via gh issue create, and write to local files such as .branch-context.md.
Sanitization: No sanitization or content validation is performed on the ingested data before it is processed by the agent.
[COMMAND_EXECUTION]: The skill utilizes shell commands to interact with the repository and project environment.
Evidence: It invokes git diff for mode detection and git commit to automate the application of review fixes.
[DATA_EXFILTRATION]: The skill interacts with GitHub to record review summaries.
Evidence: Findings are recorded using gh issue create. As GitHub is a well-known service and the data transmitted consists of development-related review summaries, this is considered safe and standard functionality.

review