The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/lane_snapshot.py executes tmux commands (such as list-panes, capture-pane, and display) using the subprocess module. These calls are used to gather operational state from the local terminal environment and do not utilize a shell, reducing the risk of command injection.
[DATA_EXPOSURE]: The skill accesses local agent logs in ~/.codex/sessions and captures terminal pane history. This access is central to its purpose of reconstructing worker state. The implementation includes a redact() function in scripts/lane_snapshot.py that uses regex patterns to mask potential secrets—including GitHub tokens, OpenAI keys, AWS credentials, and private IP addresses—before the data is presented to the agent.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content originating from terminal panes and external log files.
Ingestion points: Terminal pane content and .jsonl log files are ingested via the capture_pane and extract_log_hints functions in scripts/lane_snapshot.py.
Boundary markers: The script output does not use explicit boundary markers to separate system instructions from ingested data.
Capability inventory: The script's capabilities are limited to reading local files and executing local tmux queries via subprocess.
Sanitization: The script performs sanitization by redacting common secret formats and private IP addresses from the ingested text prior to output.

tmux-agent-lane-orchestrator