The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands (grep, bash) and a specialized CLI tool (gbrain) to perform data maintenance tasks.
**[Step 1
Search]:** Uses gbrain search and grep to locate factual errors within the local file system (brain repository).
**[Step 2/3
Memory/Identity Check]:** Uses grep to inspect local workspace files such as MEMORY.md, SOUL.md, and USER.md for misleading information.
**[Step 6
Fix]:** Employs gbrain bootstrap and gbrain remember/forget commands to update internal state and render configuration files.
[REMOTE_CODE_EXECUTION]: No remote code execution was detected. All commands interact with local tools and data stores.
[DATA_EXFILTRATION]: No network operations or data exfiltration patterns were detected. Data access is restricted to local files and the gbrain database.
[PROMPT_INJECTION]: The skill includes instructions to ignore certain conversational tendencies (e.g., "No exceptions. No 'I'll note that.'"), but these are functional constraints to ensure the agent follows the correction protocol rather than attempts to bypass safety or system-level instructions.

correction-pipeline