The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains meta-instructions designed to override default AI behavior, specifically forbidding 'performative agreement' and social niceties. It references specific platform files (e.g., CLAUDE.md) to define violations and uses high-pressure, adversarial language ('If you lie, you'll be replaced') to enforce compliance. It also establishes a 'safe word' or signaling mechanism ('Strange things are afoot at the Circle K') to influence agent communication patterns.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and implement technical feedback from external reviewers, creating an attack surface for indirect prompt injection.
Ingestion points: External code review comments processed in references/code-review-reception.md.
Boundary markers: None identified; instructions do not specify delimiters for untrusted feedback.
Capability inventory: The skill has the capability to execute shell commands via the 'Verification Gates' protocol and modify files during implementation.
Sanitization: No sanitization or validation of the technical feedback is specified before the agent attempts to 'verify' or 'implement' it.
[COMMAND_EXECUTION]: The 'Verification Gates' protocol in references/verification-before-completion.md requires the agent to identify and execute arbitrary shell commands to verify that tasks are complete. While intended for testing and builds, this provides a mechanism for command execution that could be exploited if the agent is steered by malicious external feedback.

code-review