The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely prescriptive and aggressive language to override the agent's default operational guidelines and persona, using terms like 'Iron Law,' 'Non-negotiable,' and 'Be honest, be brutal.' It explicitly forbids standard polite responses (e.g., 'Thanks,' 'Great point!') and establishes a specific code phrase ('Strange things are afoot at the Circle K') for the agent to communicate discomfort or internal conflict covertly, which could be used to signal bypasses or issues without alerting the user.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow for receiving external feedback creates a surface for indirect prompt injection where malicious instructions could be embedded in code review comments.
Ingestion points: External code review comments, particularly from 'External Reviewers' as described in references/code-review-reception.md.
Boundary markers: Absent. There are no instructions to use delimiters or strictly isolate the feedback text from the agent's instruction context.
Capability inventory: The agent is authorized to run shell commands (tests, builds, git, grep) and dispatch tasks to subagents. The 'Verification Gates' protocol (references/verification-before-completion.md) specifically instructs the agent to 'IDENTIFY' and 'RUN' commands to prove claims, which could be exploited if an external reviewer provides a malicious command disguised as a necessary 'verification step.'
Sanitization: None. While the skill encourages technical skepticism and verification against the codebase, it provides no mechanical sanitization, validation, or escaping for the content of the feedback received.

code-review