implementation-loop

Drives GitHub issues through autonomous Developer and QA sub-agent loops until merge.

96|11|Updated Feb 8, 2026
One-click install
npx skills add https://github.com/thefrederiksen/devthrottle --skill implementation-loop-thefrederiksen
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: implementation-loop
Source: https://github.com/thefrederiksen/devthrottle/tree/main/.claude/skills/implementation-loop
Command: npx skills add https://github.com/thefrederiksen/devthrottle --skill implementation-loop-thefrederiksen

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Manually shepherding a GitHub issue from implementation through QA review to merge consumes developer attention and context. This Skill supervises the full cycle autonomously, spawning fresh sub-agents for each phase so one session can drain an entire queue of issues without context exhaustion. ## Core Features & Use Cases - Autonomous DEV/QA loop: Spawns a fresh Developer sub-agent to implement an issue, then an independent QA sub-agent to verify it, bouncing between them up to 3 times before escalating to a human. - Duplicate-claim prevention: Claims issues via flow:* labels with verify-after-claim comment arbitration, sweeps stale claims, and enforces a leave-clean invariant (no dirty trees, stashes, or orphaned branches). - Multi-tracker support: Drives GitHub issues by default or Azure DevOps work items via --source devops, emitting a machine-readable IMPL-LOOP-TERMINAL sentinel on every terminal path for external queue watchers. - Use Case: Run /implementation-loop --all to drain every flow:ready-dev issue in the repo, with QA squash-merging each passing PR to main and parking weak specs as flow:needs-human. ## Quick Start Ask the agent to run /implementation-loop --all to drive every flow:ready-dev GitHub issue through development and QA until merged or escalated.

Frequently Asked Questions about implementation-loop

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

FAQPage Schema
How do I autonomously drive a GitHub issue from implementation to merge?▼

Invoke /implementation-loop with an issue number to run it through Developer and QA sub-agents until QA passes and squash-merges the PR to main. Use --all to drain the entire flow:ready-dev queue one issue at a time.

How does the loop prevent two sessions from working the same issue?▼

It uses select-then-claim with flow:in-progress labels plus a CLAIM comment; the oldest CLAIM comment wins the race and losers back off. A stale-claim sweep reclaims issues whose claim is older than 60 minutes.

Can the implementation loop work with Azure DevOps work items?▼

Yes, use /implementation-loop --source devops <workItemId>. Claim and write-back happen via az boards against the work item, while branch, PR, and squash-merge mechanics stay GitHub-based in the named code repo.

What happens when QA keeps failing an issue?▼

Each QA failure increments a bounce counter and spawns a fresh Developer sub-agent with the defect comment. After 3 bounces the loop labels the issue flow:needs-human, comments a defect summary, and stops autonomously.

Why does the loop stop on a dirty working tree before starting?▼

The Step 0a pre-flight requires an empty git status --porcelain, no leftover stashes, and the correct base branch so work never mixes with unrelated changes. On any violation it stops and emits a failed terminal signal.

What is the IMPL-LOOP-TERMINAL sentinel block?▼

It is a machine-readable block printed as the final output for every issue, reporting signal (done, needs-human, or failed), PR number, merge status, and a one-line reason. External queue watchers parse it instead of prose.