The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from conversation transcripts and existing skill files to generate new instructions.
Ingestion points: The skill mines history from workspace transcripts and skill files located in .agents/skills/ and ~/.agents/skills/.
Boundary markers: No specific boundary markers or instructions to ignore embedded malicious content within the mined transcripts are present in the instructions.
Capability inventory: The skill can create and modify files in the user's skill directories and perform git operations (worktrees, commits, PRs).
Sanitization: The skill mitigates risks by incorporating human-in-the-loop review through the AskQuestion tool and a final manual review of the generated PR before landing changes.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with the environment and manage the skill creation process.
Evidence: Uses git log to check history and performs git workflow operations (creating worktrees, committing, and opening PRs) in Step 6.

automate-me