The Agent Skills Directory

[PERSISTENCE]: The skill modifies global configuration files for various agent harnesses (e.g., ~/.config/devin/AGENTS.md, ~/.cursor/rules/global.mdc) and installs a utility script to ~/.local/bin. These changes are persistent across sessions but are necessary for the skill's stated goal of cross-harness rule synchronization.\n- [COMMAND_EXECUTION]: The run-watch.mjs script executes shell commands provided by the agent via spawn('sh', ['-c', ...]). It implements a quoting mechanism (q()) that escapes single quotes to prevent command injection when interpolating task arguments into the shell script.\n- [INDIRECT_PROMPT_INJECTION]: The devin-hook-exec-guard.mjs script acts as a filter for commands executed by the agent. 1. Ingestion points: Consumes PreToolUse event JSON from process.stdin. 2. Boundary markers: None; expects structured JSON input. 3. Capability inventory: Blocks execution by exiting with code 2; no file-write or network operations. 4. Sanitization: Parses input using JSON.parse() and utilizes regex to match and intercept sleep commands based on duration.\n- [SAFE]: No network requests or attempts to read sensitive system credentials (e.g., .ssh or .aws) were identified in any of the scripts. The skill interactions are limited to managing agent guidelines on the local filesystem.

setup-env