📦 Job Types
Five job types for different workloads. Each has its own prompt template, timeout, and default model.
At a Glance
| Type | Purpose | Timeout | Default Model |
|---|---|---|---|
| session | General-purpose coding session | 10 min | opus / gemini-3.1-pro |
| research | Deep-dive investigation → published report | 10 min | opus / gemini-3.1-pro |
| workflow | Run a specific slash command | 10 min | sonnet / gemini-3-flash |
| clara | Clara inbox sweep & classification | 10 min | sonnet / gemini-3-flash |
| autopilot | Long-running autonomous program loop | 60 min | opus / gemini-3.1-pro |
session Session
The most flexible type. Your prompt is passed directly to the CLI agent with an autonomous execution suffix. Use for any coding task: building features, fixing bugs, refactoring, deploying.
| Field | Detail |
|---|---|
| Prompt template | {your prompt}. Do not ask for confirmation. Execute ALL steps fully and autonomously. |
| Required payload | prompt (string) |
| Optional payload | resumeSessionId — resume a previous session |
| Typical output | Code changes, git commits, deployment logs |
| CWD | E:\Agentic (Hub root) |
resumeSessionId. The agent picks up from where it left off.
research Research
Triggers the research skill to produce a structured HTML report. The agent reads
.agent/skills/research/SKILL.md, investigates the topic, and publishes to /reports/.
| Field | Detail |
|---|---|
| Prompt template |
Read the research skill... research: "{topic}". Depth: {depth}. Produce a full report and publish.
|
| Required payload | topic or prompt (string) |
| Optional payload | depth — "standard" or "deep" |
| Typical output | HTML report at /reports/{slug}/ + D1 report record |
workflow Workflow
Runs a specific slash command (e.g. /status, /commit, /log). The agent
executes the workflow steps autonomously.
| Field | Detail |
|---|---|
| Prompt template | Run the workflow: {command}. Do not ask for confirmation. |
| Required payload | command or prompt (string) |
| Typical output | Varies — git commits, log entries, status reports |
clara Clara
Runs the Clara secretary agent to process the inbox — classify feedback, route items, update ongoing, sync D1. No prompt needed.
| Field | Detail |
|---|---|
| Prompt template |
Run /clara to process the inbox. Execute ALL steps (classify, route, update ongoing, update D1, log, commit).
|
| Required payload | None (prompt is optional context) |
| Typical output | Task/feedback updates, ongoing.md changes, git commits |
autopilot Autopilot
For long-running autonomous programs that loop over multiple iterations. The agent reads a program file, executes the defined loop, and saves progress checkpoints.
| Field | Detail |
|---|---|
| Prompt template |
Read the autopilot program at "{path}" and execute it. After EACH iteration, save checkpoint state.
|
| Required payload | programPath or prompt |
| Optional payload | checkpoint — JSON object to resume from |
| Timeout | 60 minutes (6× longer than other types) |
| Typical output | Multiple iterations, checkpoint files, staged commits |
Remote vs Local Context
Remote jobs run differently from local interactive sessions. Key differences to keep in mind:
| Aspect | Local (interactive) | Remote (Bifrost) |
|---|---|---|
| CWD | Current workspace | Hub root (E:\Agentic) |
| User prompts | Interactive back-and-forth | Single prompt, autonomous |
| Permissions | Manual approval | --dangerously-skip-permissions |
| Output | Terminal | SSE stream + channel messages |
| Timeout | None | 10–60 min based on type |
| Rate limit | Manual retry | Auto-retry with countdown |
| Browser | Available | No browser/UI access |