synthesis-autopilot

Warn

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: MEDIUMPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [PERSISTENCE]: The skill instructs the agent to establish mechanisms for unattended execution that survive session termination, session death, and system reboots.
  • Evidence in SKILL.md: The "Continuation" section explicitly guides the agent to use cron or "Scheduled task" re-entry to ensure the agent is re-invoked automatically as a "dead-man's switch" for long-horizon tasks.
  • The scripts/autopilot_gate.py script functions as a registry and "stop-gate" that prevents the agent session from exiting unless a continuation mechanism, a blocker, or a completion status is recorded.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where the agent continuously reads from external state files that could potentially be influenced by untrusted data sources.
  • Ingestion points: The agent is instructed to read plan files (stored at resources/artifacts/*.md) and a shared coordination board at the start of every "wake" cycle to restore its internal state.
  • Boundary markers: While the handoff queue uses hash-pinning for shared work between agents, the primary plan file lacks explicit delimiters or instructions to ignore embedded instructions during re-entry.
  • Capability inventory: The agent possesses the capability to register tasks, manage project files, and is guided to configure system-level schedulers.
  • Sanitization: No specific sanitization or validation logic is defined for the content of the plan files or coordination board records processed during re-invocation.
  • [COMMAND_EXECUTION]: The test suite for the autopilot gate utilizes shell execution to verify script behavior.
  • Evidence in scripts/test_autopilot_gate.py: The run_cli function uses subprocess.run to execute the autopilot_gate.py script within a managed environment to verify registry logic.
  • [DYNAMIC_EXECUTION]: The skill's testing framework dynamically loads the gate script for behavioral verification.
  • Evidence in scripts/test_autopilot_gate.py: The script uses importlib.util to load and execute autopilot_gate.py as a module at runtime using exec_module.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 31, 2026, 02:10 AM
Security Audit — agent-trust-hub — synthesis-autopilot