The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (conversation history and task logs) to generate new agent instructions, which could theoretically contain malicious prompts intended to influence the output skill.
Ingestion points: Recent conversation surveys and Codex task APIs (identified in Step 1 of SKILL.md).
Boundary markers: No specific delimiters or "ignore instructions" markers are mentioned to isolate historical data during the mining phase.
Capability inventory: Writing files to the project directory or home folder (Step 4), and performing git operations including worktree creation and PR submission (Step 6).
Sanitization: The instructions do not specify any filtering or sanitization of the mined history before it is used to draft the new skill.
[COMMAND_EXECUTION]: The skill uses git log to check for existing skill updates and manages changes via git worktrees, commits, and pull requests.
Evidence: Step 0 uses git log -1 --format=%cI <path> to check last edit times, and Step 6 performs standard git workflow operations.

automate-me