The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to facilitate autonomous command execution and file system modification through the mcp__claude-code__claude_code tool. It explicitly instructs the agent to 'Make all required file changes' and 'Run the test suite after implementation'.
[PROMPT_INJECTION]: The skill contains instructions (the 'PROVENANCE NOTE') that attempt to influence the downstream agent's (Claude's) reasoning by telling it to 'apply its own independent judgment' and 'not defer to any assumptions' from the source agent (OpenAI Codex). While intended for behavioral alignment, this pattern demonstrates how instructions are being passed through to influence autonomous behavior.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface:
Ingestion points: The skill interpolates untrusted data from the user or prior agent turns into the {clear description of what to implement}, {explicit requirements}, and {any constraints} variables within the tool prompt.
Boundary markers: The prompt lacks clear boundary markers (e.g., XML tags or specific delimiters) to separate the system instructions from the potentially untrusted task description.
Capability inventory: The downstream tool (claude_code) possesses high-risk capabilities including full workspace read/write access and the ability to execute code (tests).
Sanitization: There is no evidence of sanitization, validation, or filtering of the input variables before they are passed to the high-privilege tool.

claude-implement