skills/karlorz/agent-skills/autopilot/Gen Agent Trust Hub

autopilot

Pass

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: Instruction Override. The autopilot-keep-running.sh stop hook returns instructions to the agent such as "Continue from where you left off" and "Do not ask whether to continue," which serve to override the agent's standard autonomous decision-making regarding task completion.
  • [COMMAND_EXECUTION]: Bash Tool Usage. The /autopilot_reset command defined in templates/commands/autopilot_reset.md provides Bash script templates for the agent to execute, which perform file operations in the /tmp directory to manage autopilot session state.
  • [PROMPT_INJECTION]: Indirect Injection Surface. The skill aggregates session identifiers from hook inputs without performing validation, creating a potential surface for indirect manipulation.
  • Ingestion points: The templates/hooks/session-start.sh script extracts the session_id from hook input JSON and persists it to a local file.
  • Boundary markers: None are implemented to delimit or validate the session identifier.
  • Capability inventory: The slash command /autopilot_reset uses the stored identifier in Bash commands (rm, touch) to manage state files.
  • Sanitization: There is no validation or escaping of the session identifier before it is used in shell commands, which could allow for path traversal within the temporary directory if a malicious identifier is provided by the environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 12, 2026, 11:22 AM