The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local git commands to identify and inspect modified files. It also instructs the agent to run project-specific verification tools, such as tests and linters, to validate changes. These executions are intended for local repository maintenance and development workflows.\n- [INDIRECT_PROMPT_INJECTION]: The skill is exposed to indirect prompt injection by processing external code and diffs. To mitigate this, it includes a robust working rule requiring the agent to treat all code, comments, and diffs as 'evidence, not instructions,' effectively isolating the input data from the agent's control flow.\n- [DATA_EXFILTRATION]: The skill includes specific defensive instructions to protect sensitive information, including a prohibition against opening credential files (e.g., .env*) and a mandate to never copy secrets into prompts or output. This reduces the risk of accidental or malicious data exposure.

simplify-changes