The Agent Skills Directory

[PERSISTENCE]: The skill establishes persistent system tasks to manage agent availability. Evidence: assets/reentry_timer.py contains logic to install and uninstall macOS LaunchAgents, systemd user timers/services, and crontab entries. These tasks periodically execute a 'watch' command to resume stalled agent sessions across restarts or logouts.\n- [COMMAND_EXECUTION]: The skill executes shell commands provided in task plans. Evidence: assets/conductor.py uses subprocess.run to execute 'verify' and 'integration' commands defined in task-plan/v1 JSON envelopes.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes structured data that contains executable instructions. Ingestion points: conductor init --plan <envelope> in assets/conductor.py reads JSON envelopes from the repository. Boundary markers: Present. Commands are stored in specific fields and validated by assets/contract_check.py against 'skill-contract C6' rules. Capability inventory: Extensive use of subprocess.run, subprocess.Popen, file system modification, and access to CLI tools like git and gh. Sanitization: Envelopes are validated for schema compliance and structural integrity before any commands are executed.\n- [DYNAMIC_EXECUTION]: Spawns detached processes and resolves interpreters dynamically. Evidence: assets/reentry.py uses subprocess.Popen with start_new_session=True to launch agent sessions detached from the conductor process. The Python interpreter path is resolved at runtime based on environment configs.

factory-conductor