The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'maker' agent is instructed to process verdicts and findings returned by a spawned 'checker' agent. This data flow creates a vulnerability where instructions embedded in a finding could influence the maker agent's logic during the implementation or fix phase.
Ingestion points: Findings and verdicts are received from the adversarial-reviewer spawned via the Task/Agent tool as described in SKILL.md.
Boundary markers: No specific boundary markers or 'ignore' instructions for the checker's output are defined in the skill instructions.
Capability inventory: The skill context includes the ability to execute shell commands, modify code files, and spawn additional sub-agents.
Sanitization: No sanitization of the checker's findings is specified before the maker agent interprets them.
[COMMAND_EXECUTION]: The skill instructions require the agent to execute shell commands for targeted testing, QA, and git version tracking (git rev-parse HEAD). These capabilities are necessary for the skill's primary purpose of behavioral change verification but represent an broad execution surface.

review-loop