The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is designed to autonomously generate code changes and execute them (e.g., via npm test or the codex CLI) in response to test failures. The full-auto and yolo modes allow for iterative code modification and execution with minimal human intervention.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (external source code and test error logs) and uses high-privilege tools (Bash, Write, Edit) to modify the environment based on that data.
Ingestion points: The agent reads failing test logs and project source code (SKILL.md, Phase 1).
Boundary markers: Explicit delimiters or 'ignore instructions' warnings for external data are absent.
Capability inventory: The skill uses Read, Write, Edit, and Bash tools.
Sanitization: No sanitization or filtering of external content is described.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands and scripts (e.g., ./scripts/multi-model/codex-yolo.sh) to perform autonomous tasks. The yolo and full-auto modes specifically bypass standard human-in-the-loop review steps for shell execution.
[DATA_EXFILTRATION]: The 'LIBRARY-FIRST PROTOCOL' in SKILL.md instructs the agent to search and extract code from D:\Projects*. This represents broad file system access that could lead to the exposure or unauthorized reuse of code from unrelated local projects on the user's machine.

codex-iterative-fix