The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple terminal-based commands to orchestrate the approval process:
Uses tmux list-sessions and tmux list-panes to identify validator lanes.
Uses tmux capture-pane to extract transcripts from validator sessions.
Uses tmux send-keys and atm send to dispatch requests to other agents.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) because it processes untrusted plan data and forwards it to another agent without sanitization or strict boundary markers.
Ingestion points: The skill reads PLAN, RESEARCH, and SynthesisPacket files from the local repository as specified in Phase 1 and 3 of the workflow.
Boundary markers: The prompt template in Phase 3 does not utilize delimiters (e.g., XML tags or triple backticks) or "ignore embedded instructions" warnings when interpolating the content of the plans into the request sent to the validator.
Capability inventory: The skill possesses the ability to send content to terminal panes (tmux send-keys) and record results into the filesystem (.agents/council/).
Sanitization: No sanitization or validation of the input plan content is performed before it is sent to the Fable/Claude-family reviewer.

codex-approval