The Agent Skills Directory

[SAFE]: The skill functions as documentation for a permission-handling feature. It provides clear guidance and warnings regarding the use of manual, auto, and bypass modes.
[COMMAND_EXECUTION]: The documentation includes example bash scripts (permission-denied.sh, pause-for-review.sh) meant to be used as local hooks. These scripts demonstrate safe practices, such as using jq for structured JSON parsing and printf for logging tool metadata.
[INDIRECT_PROMPT_INJECTION]: The skill describes an attack surface where hooks ingest tool metadata (tool name and failure reason).
Ingestion points: permission-denied.sh and pause-for-review.sh read tool data via stdin.
Boundary markers: Not applicable as these are internal tool metadata hooks.
Capability inventory: The skill uses Bash, Read, Write, and Edit tools.
Sanitization: The example scripts use jq -r to safely extract values from JSON input before processing, which prevents common shell injection patterns.

cc-auto-mode