- Home
- Skills
- Agents & Orchestration
- Agent Foreman Orchestrator - Multi-Agent Task Pipeline
Works with the AI Agents you already use
Agent Foreman Orchestrator - Multi-Agent Task Pipeline
Give a coding agent five problems in one message and it usually solves three, merges two, forgets one and reports success.
$19
Agent Foreman Orchestrator - Multi-Agent Task Pipeline
Example session with this skill installed
Use foreman-orchestrator. Three things for the web shop: 1) the login button does nothing in Safari; 2) the CSV export loses the last row - screenshot says "Exported 49 of 50 orders"; 3) I'd like a dark-mode toggle.
- Read your context and instructions
- Compiled the agent foreman orchestrator
Summary
Three distinct problems -> three ledger rows (your exact words, screenshot transcribed) -> three task files -> three fresh workers, run one after another because they share one working tree. Every result went to a fresh reviewer; one failed review and was repaired automatically. All three are now done and verified.
Intake ledger
| # | Your words (verbatim) | Task | Mode |
|---|---|---|---|
| 1 | "the login button does nothing in Safari" | T1 Safari login | full |
| 2 | "the CSV export loses the last row" + screenshot: Exported 49 of 50 orders | T2 CSV export | precise (off-by-one visible in the export loop) |
| 3 | "I'd like a dark-mode toggle" | T3 dark mode | full |
Example acceptance (T2)
- Exporting 50 orders yields 50 rows.
- A test fails before the fix and passes after it.
- No other export format changes.
Status
| Task | Worker | Review 1 | Repair | Review 2 | Result |
|---|---|---|---|---|---|
| T1 Safari login | fresh | FAIL - the test mocked the browser check it was supposed to prove | fresh worker, repair task with the failed criterion and evidence | PASS | done |
| T2 CSV export | fresh | PASS (fail-first proven) | - | - | done |
| T3 dark mode | fresh | PASS | - | - | done |
Each row links its commit and the reviewer's evidence table. Nothing was pushed.
Handoff
Halfway through T3 the context passed the handoff threshold. The reply opened with
NEW SESSION NEEDED and the full prompt inline; the new session picked up exactly where the old one stopped, without dispatching anything twice.
Next steps
- Review the three commits and push when you are ready.
- Add more problems any time - each one gets its own ledger row and task file.
Connects securely to your tools. The creator never sees your data.
About this skill
Give a coding agent five problems in one message and it usually solves three, merges two, forgets one and reports success. Foreman Orchestrator makes your main agent a foreman instead of a worker.
- Intake ledger - every problem is written down verbatim the moment it is heard, and leaves the ledger only as done-and-verified or cancelled by you.
- One task file per distinct problem - goal, context (including transcribed screenshots), scope, non-goals, guardrails, dependency map, plan -> implement -> verify, measurable acceptance, commit rules and stop conditions. Each one is a complete brief a fresh agent can execute without the chat history.
- Fresh worker per task - unrelated problems never share a context. Different repositories run in parallel; the same working tree runs sequentially or in separate git worktrees, with simple locks.
- Independent QA and automatic repair for every task, with evidence instead of claims.
- Session handoff - a live copy-paste prompt for the next session, announced loudly before the context gets expensive.
- Backlog mode - dedupe, drop what is already fixed, cluster by root cause.
It works standalone and gets stronger with its companions: Foreman Git Guard, Foreman Verify & Repair and Foreman Token Meter.
Key capabilities
- Verbatim intake ledger; nothing heard is lost; rows close only on verified done or your cancel
- Decomposition rules for live requests, backlogs (clustering) and precise-mode fixes
- Rich task template with standing clause ("the repository outranks the task file") and pre-dispatch checklist
- Fresh worker subagent per task; write-safe parallelism with worktrees and locks
- Result template with an evidence-first header QA reads
- Built-in independent QA, repair tasks and stop conditions
- Git safety rules embedded in every task
- Live handoff prompt with context thresholds for 200k and 1M windows
- Optional scaffold of a control folder (INBOX, tasks, results) - created only with consent
Who it is for
Developers and tech leads who hand real, multi-part work to AI coding agents; teams running several agents on several repositories; anyone whose sessions run long enough to need a handoff.
Install / usage
Add the skill and say: "Use foreman-orchestrator. Set up the control folder here." Then describe your problems in any form. The foreman writes the ledger and task files, dispatches workers, verifies and reports a status table.
Limitations
Fresh workers and reviewers need subagent support; without it tasks run sequentially in one context with labelled self-review. It does not push or deploy unless you explicitly ask and your push policy allows it.
Changelog
- 1.0.0 - First release: ledger, task and result templates, dispatch and lock protocol, QA and repair, git safety, handoff template.
Fresh workers and reviewers need subagent support; without it tasks run sequentially in one context with labelled self-review. Parallel work in one repository needs separate git worktrees (git 2.5+) and no shared build state, otherwise tasks run sequentially. Task quality depends on the evidence you provide.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 4 days ago
Requires