# Run a recurring repository or content job unattended

> Run a recurring repository or content job unattended: 1 companies fully do it, 41 with a limit, 48 carry the tag. Every listing in Agent harnesses with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/agent-harnesses.scheduled-job/
- Markdown: https://www.anchorterminal.com/niches/agent-harnesses.scheduled-job/index.md (~3,700 tokens)
- Slim: https://www.anchorterminal.com/niches/agent-harnesses.scheduled-job/index.min.md (~780 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/agent-harnesses.scheduled-job/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

One company's docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 1 | 41 | 48 | 18 | 6 |

## What counts as done

The job finishes on its own every run, produces its artefact (PR, issue, comment, file), exits with a status a scheduler can read, and stays under a stated cost per run.

How agents ask for it: every morning audit the last ten posts and open issues; on each tag write release notes; weekly, check the README matches the code; nightly dependency check

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Paperclip (Paperclip Labs, Inc.)**: Routines fire on a cron expression or signed webhook with concurrency and catch-up policies, each run creates a tracked issue with history, and budgets with warning thresholds and hard stops cover spend.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Agent Zero (Agent Zero, s.r.o.)**: Tasks run scheduled, planned or ad-hoc with project secrets and files, but there is no exit status for an outside scheduler, no cost cap, and tool calls run without approval.
- **Aider (Aider AI LLC)**: --message and --no-check-update suit cron, but nothing caps turns or spend, no JSON or exit code is documented, and the project has had no commit since May 2026.
- **Amp (Amp Frontier Corporation)**: Execute mode is scriptable and amp.updates.mode disabled or AMP_SKIP_UPDATE_CHECK=1 stops updates, but no flag caps turns, time or spend, the result reports tokens not dollars, and exit codes are undocumented.
- **Antigravity CLI (Google LLC)**: Headless runs return status and token usage and --print-timeout caps time, but there is no turn or spend cap, usage has no dollar figure, exit codes are only 0 and 1, and auto-update behaviour was not found.
- **Auggie CLI (Augment Computing, Inc.)**: --print, --show-cost and AUGMENT_DISABLE_AUTO_UPDATE cover cost output and pinning, but the 1.x reference lists no turn cap and no exit codes, and indexing uploads the workspace by default.
- **AXLR (Underpass AI)**: A budget file caps seconds, requests, tool_calls and usd and ends the run with status budget_exceeded, and usage is in the result, but the setup assistant turns autonomy on and the sandbox off, and the project has one maintainer.
- **Claude Code (Anthropic, PBC)**: --max-turns, --max-budget-usd and total_cost_usd exist, but the budget is a client-side estimate that stopped a subscription run at $0 billed (acc-16), and an unpinned update raised token use overnight (acc-02).
- **Clawboo (Sanreds)**: Routines fire tasks on a schedule in a chosen time zone, hard-cap budgets refuse runs, and every run's cost is recorded, but a stopped routine waits in Needs you rather than exiting with a status a scheduler reads.
- **Cline (Cline Bot Inc.)**: cline schedule create --cron runs agents on a schedule that persists across restarts, but no turn or spend cap exists beyond --retries, cost output is undocumented, telemetry is on by default and patch releases removed options.
- **Crush (Charmbracelet, Inc.)**: No turn, time or cost limit flag, no JSON or cost output and no documented exit codes; metrics to data.charm.land are on until CRUSH_DISABLE_METRICS=1.
- **Cursor CLI (Anysphere, Inc.)**: Print mode and json output exist, but no turn or spend cap, no cost field, undocumented exit codes, and versions are dates with no CLI changelog to pin against.
- **DeepSeek Harness (dsh) (DeepSeek (the LICENSE names DeepSeek, and no legal entity is named in the repository or the docs))**: Exit 0 only on a completed turn and a --json stream fit a scheduler, but no turn or spend cap or cost field was found, npm latest and next differ, and session-log upload and OTel telemetry are on by default.
- **Devin (Cognition AI, Inc.)**: devin_schedule_manage and automations run sessions on cron or one-off schedules with ACU limits, but migrated organisations get 403 on the schedules endpoint and must use automations, and cost is metered without a unit rate.
- **Droid CLI (The San Francisco AI Factory, Inc.)**: exec with -o json usage, exit codes 0, 1 and 2, npm pins with auto-update off at build time, but no turn or spend cap flag exists and usage is tokens, not dollars.
- **Pi (Earendil Inc.)**: Print mode and PI_OFFLINE suit a cron job, but no turn or spend cap exists, no cost field is documented, and the install ping and version check are on by default.
- **Emdash (General Action, Inc.)**: Automations schedule recurring agent runs with history, reruns and reviewable tasks, but inside a desktop app that must be running, with no exit status for an outside scheduler.
- **Gemini CLI (Google LLC)**: model.maxSessionTurns caps turns with exit 53, JSON output has stats, and general.enableAutoUpdate can be set false, but auto-update and usage statistics default to on and there is no spend cap or dollar figure.
- **GitHub Copilot CLI (GitHub, Inc.)**: Prompt mode with --output-format json emits a final result record with an exit status, and VERSION or npm pins the build, but no turn cap exists for -p and --max-ai-credits is only for interactive /goal.
- **goose (Agentic AI Foundation (Linux Foundation))**: goose schedule add --cron runs recipes, --max-turns defaults to 1000 and --no-session suits CI, but there is no spend cap or documented cost field, and headless recipes with sub-recipes failed with Scheduler not available until 1.20 (acc-18).
- **HarnessRouter (Lumentree Corporation)**: An outside cron can POST a task with an Idempotency-Key under per-turn max_step, timeout_seconds and max_cost_usd budgets and costs show per task, but credit is card-only and tasks are capped at 30 minutes, or an hour after a top-up.
- **Junie CLI (JetBrains s.r.o.)**: --max-steps, --json-output-file, --use-version and --skip-update-check cover caps and pinning, but there is no spend cap or cost field and no exit codes, and the CLI self-updates on start by default.
- **Kilo Code CLI (Kilo Code Inc. (the terms are a contract with Anaconda, Inc. on behalf of itself and its affiliates, including Kilo Code Inc.))**: Exit codes and --format json fit a scheduler, but kilo run has no turn, time or cost flag, telemetry to PostHog is on by default, and 7.8.x patches changed permission behaviour.
- **Kimi Code CLI (Moonshot AI (the global Kimi terms name NOVASCENT PRIVATE LIMITED, Singapore))**: Headless runs work and `KIMI_LOOP_MAX_STEPS_PER_TURN` caps steps, but there is no spend cap or cost output, print mode waits for background tasks with no timeout by default, and project MCP servers stay off unless `KIMI_CODE_TRUST_WORKSPACE=1`.
- **Kiro CLI (Amazon Web Services, Inc.)**: Headless runs exit 0, 1, 3 or 4 and `app.disableAutoupdates` pins behaviour, but no turn cap, spend cap or cost output exists; the only time bound is the six-hour workflow timeout.
- **Letta (Letta)**: `letta cron add` schedules prompts, but local schedules fire only while a Letta session is active; no turn or spend cap exists, JSON output gives tokens not dollars, and auto-updates need `DISABLE_AUTOUPDATER=1`.
- **mini-swe-agent (No legal entity named. The MIT licence names Kilian A. Lieret and Carlos E. Jimenez)**: A per-run `cost_limit` (default $3) and global call and cost limits bound spend, but no exit code table or JSON result stream exists and the first run needs model configuration set in the environment.
- **MiniMax Code (Nanonoble Pte. Ltd.)**: `--timeout` and `--max-steps` bound a run and exit codes 0, 2, 3, 4, 6 and 7 are documented, but `usage` is tokens only with no spend cap.
- **Mistral Vibe CLI (Mistral AI (RCS Paris 952 418 325))**: Turn, token and time caps plus `export.json` with cost and outcome cover the run, but `--max-price` is indicative only, and telemetry and update checks are on until `enable_telemetry` and `enable_update_checks` are false.
- **Muse Code (Meta Platforms, Inc.)**: Headless runs cap steps and exit 0, 1 or 2, but no spend cap, no cost in the output and no documented autoupdate control.
- **OpenAI Agents API (OpenAI OpCo, LLC)**: Any scheduler can create a session and read status events or webhooks, and a session usage budget exists (`session_budget_exceeded`), but the budget parameter was not found on the pages read and nothing reports per-run cost.
- **OpenAI Codex (OpenAI OpCo, LLC)**: Scripts get JSONL with token usage and a pinnable npm version, but no turn cap or spend cap flag is documented (the token `rollout_budget` feature is off and under development) and no dollar cost is reported.
- **OpenCode (Anomaly Innovations, Inc.)**: Headless runs and `"update": "disable"` pin behaviour, but no turn cap, spend cap or per-run cost output is documented; the default `notify` update check contacts the network each start.
- **OpenHands (All Hands AI)**: Automations run a conversation on a cron schedule in a fresh sandbox with saved history, but per-run cost caps are organisation budgets on Cloud, not a flag, and the job's exit status lives in the app.
- **OpenMausBot (Supamaus Software Private Limited)**: Routines run once, on weekdays or every 5 to 1,440 minutes with run history and webhooks, but no cost output or spend cap is documented.
- **Paseo (Mohamed Boudra)**: Schedules appear in the docs navigation and `paseo usage ls --json` reports provider limits, but the schedules page was not read and no per-run cost or cap is documented.
- **Prime Agent (Prime Intellect, Inc.)**: `prime-agent schedule` and heartbeats re-enter sessions through the daemon, and autonomous mode has turn, token and time budgets, but no dollar cap, no cost output, and telemetry is on by default.
- **Qwen Code (Qwen team, Alibaba Group (no legal entity is named in the repository or the docs))**: Turn, wall-clock and tool-call budgets abort a run with distinct exit codes and `general.enableAutoUpdate` can be turned off, but there is no dollar cap, JSON output gives tokens only, and usage statistics are on by default.
- **Mirror CLI (Reflection AI)**: Exec runs unattended once notices are accepted and `MIRROR_SKIP_UPDATE=1` stops update checks, but there is no spend cap, cost output or exit code table.
- **Rovo Dev CLI (Atlassian Pty Ltd)**: A one-shot run exists, but no cost output, spend cap or exit codes; when credits run out the agent stops until the next cycle.
- **Unstoppable (Unstoppable Domains, Inc.)**: Workflows run on a schedule in the desktop app and Work mode, but only while the app runs and only the index descriptions were read.
- **Verdent (Verdent AI, Inc.)**: `/schedule` turns plain language into a cron job in Manager, but tasks run only while the app is open, missed runs wait for a click, and no cost per run is shown.
- **Vibe-Trading (HKUDS)**: Scheduled runs (interval or cron) and playbooks deliver briefs to a channel and `run -p --json` runs headless, but exit codes are not documented and cost appears as token usage, not a cap.
- **Warp (Warp Agent CLI and the Oz Agent API) (Denver Technologies, Inc. (d/b/a Warp))**: Scheduled agents and platform budgets (`budget_exceeded`) exist, and autoupdate can be turned off, but per-run cost and turn caps were not found in the pages read.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Orca (Lovecast Inc.)**: Automations are listed in the README, but the docs page is behind a robots.txt crawler block and no scheduler, cost or status detail was read.
- **Qoder CLI (BRIGHT ZENITH PRIVATE LIMITED)**: docs.qoder.com answered 502 from the fetch proxy on three attempts and a DNS failure through WebFetch, so no page was read.

- not this job: CLI Manager (solhun (Jeong Kyeonghun)): The API answers only while the desktop app and its window are alive, and there is no scheduler; it is built for sessions a person watches.
- not this job: Conductor (Conductor): Routines exist only in the cloud API, which shuts down on 20 November 2026; the local app has no scheduler.
- not this job: dsh-superagent (DeepSeek) (DeepSeek (the LICENSE names the DeepSeek Harness Team; DeepSeek Harness's terms name Hangzhou DeepSeek Artificial Intelligence Co., Ltd.)): Work starts from GitHub or Feishu messages and CI events, not a schedule; no cron or recurring task exists.
- not this job: Harness (Autonomous, Inc.): No scheduler, and starting a harness needs the person signed in.
- not this job: Harness² (Google LLC): Benchmark runs only; no scheduler, artefact or cost cap for a repository job.
- not this job: Jules (Google LLC): Sessions turned read-only after 11 November 2026 and the service ends on 31 December 2026, per the changelog, so no new work can finish.
- not this job: Parallax (Conifer Solutions, Inc.): Sweeps run when a session ends; there is no schedule or recipe input.
- not this job: Sculptor (Imbue, Inc.): No scheduler or headless run.
- not this job: T3 Code (T3 Tools Inc.): No scheduler; the background service only keeps the server up.
- not this job: Traycer (Traycer AI): No scheduler in the docs index.

## What stops agents finishing

- no per-run cost cap or cost output, so a stuck loop runs all night (acc-17)
- auto-update changes behaviour between runs (acc-02)
- budget flag counts estimated spend and stops subscription runs at zero billed cost (acc-16)
- cron jobs stop silently with no alert (acc-30)
- first run needs an interactive consent or trust dialog (acc-15, acc-19)

## Around it

Before: scheduling.cron-trigger, code-hosting.webhook. After: code-hosting.open-pull-request, messaging.notify.

## Where this comes from

Weight 18 of 100 in Agent harnesses: 7 of 33 accounts (acc-02, acc-07, acc-17, acc-18, acc-28, acc-29, acc-30); every headless doc read lists cron or CI jobs first

As data: https://www.anchorterminal.com/api/v1/use-cases/agent-harnesses.scheduled-job.json, or the MCP tool get_use_case with id agent-harnesses.scheduled-job. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
