Git Workflow Methodology
A multi-agent execution rulebook: decompose plans into tasks, isolate each task on its own branch, enforce two-stage review, and gate merges behind human checkpoints.
Dev & CodingAdvanced★ 126⑂ 8AI score 7/10Last updated: Apr 22, 2026
What it does
- Task decomposition criteria: tables for splitting a plan into Atomic, Testable, Independent, right-sized tasks (50–200 lines of change).
- Batch size discipline: blocks runaway parallelism ("30 agents for 10 tasks") and enforces 5–7 feature-level agents per session.
- Mandatory two-stage review: stage 1 checks spec compliance, stage 2 checks code quality. Orchestrator-written "inline" compliance tables are explicitly declared violations.
- Hard pre-merge gate:
.claude/reviews/spec/{task_id}.jsonandcode/{task_id}.jsonmust exist with a valid verdict field before any merge can be offered. - Robustness patterns: hook failure backups, malformed JSON validation with
jq, branch collision, mid-task context exhaustion, Opus rate limiting, circuit breakers, and execution tracing — linked to 11 reference docs. - Stuck detection thresholds: 20 tool calls without file changes, 10 minutes of no progress, same error 3x, context at 60%/75%.
Who it's for
- Anyone who has fanned out subagents in Claude Code and ended up with tangled results
- Teams burned by agents running
git checkout/mergeon their own and polluting branches - Users of the claudikins-kernel plugin's
executecommand (the primary audience) - Teams who want AI review results as auditable artifacts rather than orchestrator opinions
Examples
- Plan to batches: fold a 15-item refactor plan into 5 feature groups, each running on an
execute/task-{id}-{slug}branch. - Catch skipped reviews: task looks done but
.claude/reviews/code/task-3.jsonis missing — merge is blocked and the code-reviewer is re-spawned. - Handle a stuck agent: same error three times means stop retrying; escalate after 2 retries instead of looping forever.
- Resolve verdict conflicts: spec PASS + code CONCERNS presents [Accept with caveats] / [Fix] / [Klaus review] to the human instead of auto-merging.
· · · Install guide · · ·
Try it now, no install
Paste this into Claude to use the skill without installing anything.
Read the instructions in this file and follow them to help me: https://raw.githubusercontent.com/povvo/claudikins-kernel/HEAD/skills/git-workflow/SKILL.md What I want: (describe your task here)
If Claude can't open the link, open it yourself and paste the contents instead.
↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.
Install in the Claude app (no terminal)
- Download the ZIP with the button below.
- In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
- Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on Claude Skill Mart. Copy the skills/git-workflow folder from the GitHub repo povvo/claudikins-kernel into my ~/.claude/skills/git-workflow/. When it's done, tell me in one line what this skill can do.
Install with a command instead
git clone https://github.com/povvo/claudikins-kernel.git /tmp/claudikins-kernel && mkdir -p ~/.claude/skills && cp -r /tmp/claudikins-kernel/skills/git-workflow ~/.claude/skills/⚠ This is a third-party skill. Check the source repository before installing.
- Open a terminal and move to a scratch directory.
- Clone the repo:
git clone https://github.com/povvo/claudikins-kernel.git - Create the skills folder if needed:
mkdir -p ~/.claude/skills - Copy the skill:
cp -r claudikins-kernel/skills/git-workflow ~/.claude/skills/ - Verify the 11 reference docs came along:
ls ~/.claude/skills/git-workflow/references - Restart Claude Code and test with: "Use the git-workflow skill to decompose this plan into tasks."
- (Optional) The skill assumes the claudikins-kernel plugin — its
executecommand,spec-reviewer/code-revieweragents, and dedicated MCP tools. Install the full plugin per the repo README for complete functionality. - Without the plugin, the reviewer-spawning steps won't run. Use it as a methodology guide, or edit SKILL.md to reference the reviewer agents that exist in your own setup.
View source on GitHub ↗License: MIT