Skip to content
Agentmart
Submit a tool

agent-loop-auto

Autonomous variant of /agent-loop for ONE repo: thin coordinator + parallel implementer subagents + a coordinator-dispatched INDEPENDENT reviewer, plus an…

Skill by PlatformRelay

IndexedMIT★ 1
Write a review

Chalk Labs

Need help deploying agent-loop-auto?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/platformrelay-agent-loop-auto

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Autonomous variant of /agent-loop for ONE repo: thin coordinator + parallel implementer subagents + a coordinator-dispatched INDEPENDENT reviewer, plus an Integrator subagent that auto-merges each PR (rebase) without operator input once gates + independent APPROVE + CI are green — then loops to the next lanes until the backlog is exhausted or a stop condition hits. Invoking this skill is the operator's explicit authorization to merge autonomously. Use for hands-off progress on your OWN backlog; never for external-contributor PRs.

Use cases: devops, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/agent-loop-auto

What's inside

Skill name
agent-loop-auto
Description
Autonomous variant of /agent-loop for ONE repo: thin coordinator + parallel implementer subagents + a coordinator-dispatched INDEPENDENT reviewer, plus an Integrator subagent that auto-merges each PR (rebase) without operator input once gates + independent APPROVE + CI are green — then loops to the next lanes until the backlog is exhausted or a stop condition hits. Invoking this skill is the operator's explicit authorization to merge autonomously. Use for hands-off progress on your OWN backlog; never for external-contributor PRs.
License
MIT
Instructions
80 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# agent-loop-auto — autonomous parallel loop (auto-merges)

Everything `/agent-loop` does (thin coordinator + **parallel** implementers + **coordinator-dispatched
independent review**, per-repo, TDD → gates → review), **plus**: it **auto-merges** each lane's PR when
it passes, then keeps picking new lanes until done. Read `/agent-loop` for the implementer procedure,
the independent-review step (3b), and the role schemas — this file only states what's different: the
auto-merge gate and a dedicated **Integrator subagent** that keeps merge work off the coordinator.

> **Autonomy is operator-authorized by invoking this skill.** It overrides the base loop's
> "operator merges" step — but only within the guardrails below. This is *your own* backlog being
> advanced hands-off; it is **not** a path to merge other people's code.

## When to use
- You want the harness to work through a repo's backlog and land the results without babysitting.
- **Not** for external-contributor PRs — those always need operator-visible review (`/merge-open-prs`).

## Auto-merge gate (ALL required, per lane)
A lane's PR auto-merges only when **all** hold:
1. Full gate matrix green (repo config) — `task verify/lint/test/coverage/scrub` or `pnpm build/export`.
2. The **coordinator-dispatched independent reviewer** (base Step 3b — a fresh subagent that saw only
   the diff, never the implementer's context) returned **APPROVE** with **no P0/P1** findings.
3. CI is **green on the PR's head commit** (`gh pr checks`), and the PR is **MERGEABLE** (no conflicts).
4. The lane's author is the harness/operator (never an external contributor's PR).

Then the coordinator hands the lane to an **Integrator subagent** (below) — it does not merge inline.

> **Permission-layer veto (learned 2026-07-16, mkurator):** in auto-permission/background environments
> the classifier may categorically deny `gh pr merge` to the default branch (public-deploy-triggering
> merges especially) — even when this gate is fully satisfied. Do not route around a denial. Fallback
> per gate-clean lane: record the PR in the repo INBOX 🟢 with the exact merge command, and ask the
> operator directly if they are live. True auto-merge requires the operator to add a `gh pr merge`
> allow rule to the repo's `.claude/settings.local.json` (mkurator has one since 2026-07-16).

## Integrator subagent (keeps merge work off the coordinator)
For each gate-clean lane the coordinator dispatches **one Integrator subagent** — the coordinator does
not rebase/merge inline. The integrator:
1. **Rebase onto current `main` first — isolation worktrees can come up on a STALE base.** An
   `isolation:"worktree"` worktree may be created at an earlier session HEAD, not the live `main` (seen
   this run: a worktree 7 commits behind). `git rebase main` the lane head; **never ff-merge an
   unrebased head**. (Tell implementers to `git merge --ff-only main` at lane start too.)
2. **Re-run the full gate matrix on the rebased head** (file-disjoint lanes rebase clean — verify anyway).
3. `gh pr merge <n> --rebase --delete-branch`. **If the classifier denies the merge (veto note above),
   do not route around it** — return `{merged:false, error:"classifier-veto"}` so the coordinator runs
   the INBOX-record fallback.
4. **Return** `{ lane, merged, mainSha, error? }`. On success the coordinator moves the board row to
   **Integrated**, clears any matching INBOX task, and updates `SESSION-HANDOFF`; on `error` it records
   and skips. The coordinator routes on this schema — it does not read the merge output itself.

Integrators run **serialized** (one merge to `main` at a time); implementers stay parallel.

## Stop conditions (halt the loop and hand back)
Stop and surface via the repo's INBOX (a **DECISION** with options) when any occur:
- Backlog empty, or no remaining **non-overlapping** lanes to parallelize.
- A lane hits **BLOCK / REQUEST CHANGES**, red or pending CI, or a merge conflict — leave that PR open,
  don't merge, record why.
- A change touches **security, public API/CRD, or release** — surface for a human look before merging
  even if gates pass (don't silently auto-merge high-blast-radius changes).
- `main` CI goes red after a merge — stop starting lanes, diagnose.
- A configured ceiling is reached: max lanes, max iterations, or the turn's token budget.

## Flow
1. **Steps 1–3b of `/agent-loop`** — orient, pick N non-overlapping lanes, dispatch parallel
   implementers, then dispatch the **independent reviewer** per lane; collect the structured returns.
2. For every lane meeting the auto-merge gate, **dispatch an Integrator subagent** (above); skip +
   record the rest. The coordinator routes on each integrator's `{merged, error?}` return — it does not
   rebase or merge inline.
3. **Loop:** pick the next batch of independent lanes and repeat, until a stop condition.
4. **Report + `/retrospective`:** what merged, what was skipped and why; log notable decisions to
   `decisions.md` (including any dissent). Leave the repo green.

## Do not
- Auto-merge on anything less than **gates green + independent APPROVE + green CI + mergeable**.
- Merge inline in the coordinator, or run integrators in parallel — merges to `main` are serialized.
- Auto-merge an **external contributor's** PR, or a security/API/release change without surfacing it.
- Run past the stop conditions / ceilings, or force-merge / use `--admin` / bypass branch protection.
- Parallelize machine-locked suites (e2e/integration), or coordinate across repos.
- Merge partial or self-unreviewed work — every merge still passes the same independent review gate as
  `/agent-loop` (a fresh reviewer that never saw the author's context).

From PlatformRelay/Kaddy, MIT.

More from PlatformRelay/Kaddy

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Jul 20, 2026 (3 months ago)
Last release
Jul 21, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from PlatformRelay

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get agent-loop-auto running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying agent-loop-auto?

CallGet help
Cite this page

agent-loop-auto skill. Agent Mart. https://agentmart.dev/skills/platformrelay-agent-loop-auto (last updated 2026-10-24)