The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform project discovery, executing find commands to inventory project files and git rev-parse to determine repository roots.
[COMMAND_EXECUTION]: In the final phase, the agent generates a shell script (move-script.sh) containing filesystem move (mv) and directory creation (mkdir) commands. It then executes this script via Bash to finalize reorganization. The script is saved to a .tmp folder, providing a persistent audit trail of the actions performed.
[PROMPT_INJECTION]: The skill ingests untrusted content from CLAUDE.md and various project documents, which creates an indirect prompt injection surface. 1. Ingestion points: Reads policy from CLAUDE.md and classifies file content during auditing. 2. Boundary markers: No explicit delimiter markers are used for external data. 3. Capability inventory: Accesses Edit, Write, and Bash (for moves) tools. 4. Sanitization: No data sanitization is implemented. Mitigation: The potential for autonomous abuse is mitigated by required human approval gates between the audit, planning, and execution stages.

sweep-project