The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/scan.mjs executes system commands including git, lsof, and du using Node's execFileSync. These calls are used to manage git worktrees, identify processes holding file locks, and calculate disk usage. The use of execFileSync with argument arrays instead of shell strings mitigates command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests a JSON session list via the --sessions argument. This represents an attack surface where external data influences the skill's logic (deciding which worktrees to keep or delete). However, the risk is minimal as the data is parsed as structured JSON and used solely for logical comparisons rather than being interpolated into natural language prompts for the agent.
[DATA_EXPOSURE]: The skill reads local file system metadata, git worktree lists, and process information. To prevent data loss, it writes patches of uncommitted changes to a local cleanup/ directory within the project's .git folder. No sensitive information or harvested data is exfiltrated over the network.

cleanup