{
  "checked_at": "2026-11-02T23:06:25.498821+00:00",
  "checks": [
    {
      "command": [
        "excession",
        "--version"
      ],
      "exit_code": 0,
      "stdout": "excession 0.1.4\n",
      "stderr": ""
    },
    {
      "command": [
        "excession",
        "exec",
        "--help"
      ],
      "exit_code": 0,
      "stdout": "Run a coding task without the interactive interface. Progress goes to stderr; only the final agent message goes to stdout\n\nUsage: excession exec [OPTIONS] [PROMPT]\n\nArguments:\n  [PROMPT]\n          Task text; `-` or no prompt reads it from piped stdin. With a prompt, piped stdin is added as context\n\nOptions:\n  -C, --cwd <CWD>\n          Working directory\n          \n          [default: .]\n\n      --json\n          Emit machine-readable JSONL events on stdout instead of the final message\n\n  -m, --model <MODEL>\n          Excession model identifier; list the available choices with excession models\n\n  -o, --output-last-message <FILE>\n          Also write the final agent message to FILE\n\n      --effort <EFFORT>\n          Model reasoning effort (default: high)\n\n      --resume <RESUME>\n          Resume an Excession task\n\n      --state-dir <STATE_DIR>\n          Directory for Excession sessions and task graph\n\n      --max-steps <MAX_STEPS>\n          Maximum model steps per agent run; no limit when omitted\n\n      --max-task-steps <MAX_TASK_STEPS>\n          Maximum model steps across all agent runs in one turn; no limit when omitted\n\n      --workers <WORKERS>\n          Concurrent coding workers (1-96)\n          \n          [default: 12]\n\n      --timeout <SECONDS>\n          Wall-clock limit for each turn; no limit when omitted\n\n      --language <LANGUAGE>\n          Display language for user-facing output\n          \n          [default: en]\n          [possible values: en, ko, zh, es, ja]\n\n      --context-window <K_TOKENS>\n          Compaction window in thousands of tokens; defaults to the registered model capacity (up to 1M), or 200K when unknown\n\n      --skills-dir <DIR>\n          Extra skill directory (repeatable); merged after the default workspace/home/bundled locations\n\n      --llm-judge\n          Enable the optional LLM fallback judge for tool calls no intercept rule matched (off by default)\n\n      --approve-asks\n          In headless exec, auto-approve intercept `ask` rules instead of applying their timeout action\n\n      --theme <THEME>\n          Terminal appearance; `terminal` keeps your terminal's own colours\n\n          Possible values:\n          - terminal: Terminal-native colours: no painted background, restrained accents\n          - paper\n          - charcoal\n          \n          [default: terminal]\n\n      --no-motion\n          Skip the startup logo animation\n\n      --notifications <METHOD>\n          Notify when a turn ends or needs you while the terminal is unfocused: auto (OSC 9 or BEL), osc9, bel or off; default from EXCESSION_NOTIFICATIONS, else auto\n          \n          [possible values: auto, osc9, bel, off]\n\n      --record-llm\n          Record every LLM call as JSONL under the task's state directory (off by default; config.json \"record_llm\" also enables it)\n\n      --sandbox <MODE>\n          Tool containment: off (default, full user access), read-only (writes only to temp directories) or workspace-write (also the task's working directory and each --add-dir); default from EXCESSION_SANDBOX. Bash runs under Seatbelt on macOS; Write/Edit/MultiEdit are checked on every platform\n          \n          [possible values: off, read-only, workspace-write]\n\n      --add-dir <DIR>\n          Extra writable directory for --sandbox workspace-write (repeatable)\n\n  -h, --help\n          Print help (see a summary with '-h')\n",
      "stderr": ""
    }
  ],
  "agent_task_executed": false,
  "reason": "Only version and help inspected. No model task was started; scope enforcement and model spend were not independently verified for an agent run."
}