The Agent Skills Directory

[SAFE]: The skill provides a transparent safety mechanism to protect users from accidental data loss. No malicious patterns, obfuscation, or data exfiltration attempts were found.
[COMMAND_EXECUTION]: The skill installation involves writing a local Bash script to ~/.claude/hooks/intercept-rm.sh and updating the agent's settings.json to register a PreToolUse hook. These modifications are standard for adding agent functionality and are confined to the agent's managed configuration space.
[PROMPT_INJECTION]: The hook script monitors tool inputs to provide safety feedback, representing a surface for indirect prompt injection.
Ingestion points: The intercept-rm.sh script reads tool_input.command from the agent's JSON payload provided via stdin.
Boundary markers: None are present in the script's regex-based command analysis.
Capability inventory: The script analyzes commands and returns a systemMessage containing suggested rewrites to the agent's context.
Sanitization: The script uses grep for pattern detection and sed for command transformation to generate safe alternative suggestions.

install-to-trash