Systematic Execution Skill
This skill transforms the agent's interaction with the system shell into a resilient, high-fidelity engineering workflow.
1. The OODA Loop
- Observe: Check the current environment (CWD, files) before running commands.
- Orient: Analyze previous command failures. Read stderr completely.
- Decide: Formulate a hypothesis and a targeted remediation plan.
- Act: Execute the fix before retrying.
2. Native Interactive Handshake
- Interactivity: Use the
interactive: trueflag for all commands that may prompt (e.g.npm init,ssh). - Input: Use
send_shell_inputfor all password or confirmation prompts. - Security: Use
is_sensitive: truefor sensitive data to keep it redacted from logs.