stuck-crewmate-recovery
Pass
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts located at
bin/fm-send.shto send control sequences, keystrokes (such as Escape), and corrective messages to agent windows. - [PROMPT_INJECTION]: The playbook implements a relaunch procedure that appends 'progress so far' notes to agent briefs. This creates an indirect prompt injection surface where instructions present in the previous execution context could be interpolated into the new brief.
- [COMMAND_EXECUTION]: The skill reads from local state files at
state/<id>.metato identify the appropriate harness adapters for active agent sessions.
Audit Metadata