flow-next-land

Automates PR babysitting ticks that fix CI, resolve review threads, merge, and release.

692|55|Updated Dec 26, 2025
One-click install
npx skills add https://github.com/gmickel/gmickel-claude-marketplace --skill flow-next-land
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: flow-next-land
Source: https://github.com/gmickel/gmickel-claude-marketplace/tree/main/plugins/flow-next/codex/skills/flow-next-land
Command: npx skills add https://github.com/gmickel/gmickel-claude-marketplace --skill flow-next-land

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Open pull requests stall waiting on CI failures, unresolved review threads, and manual merge decisions. This Skill runs an autonomous tick that discovers build-loop-authored PRs, classifies each through a gate tree (CI tri-state, patience window, review signals, merge state), takes at most one action per PR, and emits a terminal LAND_VERDICT line for cadence drivers.

Core Features & Use Cases

  • Autonomous PR gate tree: Reads every CI check (never required-only), unresolved review threads, review signals, and mergeStateStatus to classify each PR before acting.
  • Bounded self-healing: Fixes failing CI within a strike budget, dispatches resolve-pr for review threads, and performs server-side catch-up via gh pr update-branch.
  • Safe merge and release: Merges only with --squash --match-head-commit after all gates pass, then closes the spec, runs release steps, and syncs the tracker.
  • Use Case: Run it on a /loop 30m cadence so every half hour it wakes, fixes a red CI check on one PR, merges another whose reviews converged, and prints LAND_VERDICT for the driver to decide whether to continue.

Quick Start

Ask the agent to run the flow-next land tick with --dry-run first to see how each open PR would be classified without mutating anything.

Frequently Asked Questions about flow-next-land

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automatically merge GitHub PRs after CI passes?

Run a land tick on a cadence such as /loop 30m. It gates each PR on green CI across every check, zero unresolved review threads, the configured review signal, and mergeStateStatus, then merges with --squash --delete-branch --match-head-commit, never via gh pr merge --auto.

How does the skill fix failing CI on a pull request?

When CI is red and the head is current with its base and thread-clean, land plans a ci-fix action: it checks out the PR branch, fixes the failure, stages only the edited files, pushes, and records a strike in its ledger. A configurable budget (default 3) stops endless fix loops.

Can I preview what the PR babysitter would do without changing anything?

Yes, pass --dry-run. The tick performs full discovery and per-PR gate classification, reporting the CI tri-state read, review-signal state, and would-be action for each PR, with zero mutations: no checkout, push, label, merge, dispatch, or ledger write.

Does it work with draft PRs and human reviewer requests?

Yes, via the opt-in land.requestReviewers config. When human review is the only missing merge input, land flips a draft PR to ready and requests the configured reviewers or codeowners, at most once per head SHA, recorded in the ledger to prevent duplicate requests.

Why does a land tick report NEEDS_HUMAN and stop acting on a PR?

NEEDS_HUMAN is the ambiguity escape hatch: it fires for a dirty working tree, failed gh probes, multiple open PRs on one branch, branch-only authorship without the make-pr marker, a durable flow-next:needs-human label, or a non-retryable review verdict. Land never asks the user questions mid-tick.

Can land run alongside the Ralph autonomous driver?

No. Land and Ralph are alternative autonomous drivers and must never be nested. If FLOW_RALPH or REVIEW_RECEIPT_PATH is set in the environment, land refuses to run and immediately emits LAND_VERDICT=NEEDS_HUMAN.