The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from past conversation logs to generate instructions for new skills.
Ingestion points: Transcript files located in the workspace-scoped agent-transcripts/ directory.
Boundary markers: Absent; the instructions do not specify delimiters to isolate potentially malicious commands within the mined transcripts.
Capability inventory: The skill utilizes create-skill to write to the local filesystem and executes git commands.
Sanitization: The skill employs a high-confidence filtering strategy, requiring patterns to be detected across multiple independent history slices before codifying them, which mitigates single-instance injection attempts.
[COMMAND_EXECUTION]: The skill uses shell commands to manage its workflow and track changes.
Evidence: Uses git log -1 --format=%cI <path> to determine the last modification time of existing skills for incremental mining.
Evidence: Instructs the agent to work in a worktree and open pull requests for changes, involving multiple git operations.

automate-me