By ahlerjam
Universal GitHub-issue-driven agent workflow: AI-groomed backlog, autonomous runner with hard budgets, model tiering, five-layer verification stack.
Use when a pull request needs the cross-vendor critic pass — Codex reviews the diff against the issue spec and AGENTS.md, including test-manipulation checks; falls back to a same-family Claude review (or a skip) when Codex isn't configured — normally invoked by the implement runner, usable standalone with a PR number.
Use when a defined scope of this repository should be built autonomously — an epic number, a milestone, an issue list, "the next N issues", or "find gaps in area X" — and the operator wants no further interaction until done or stopped.
Use when creating or structuring GitHub issues, epics, or user stories for this repository before implementing, when grooming the backlog (gap scan), or when refining a user impulse into a spec issue. Does NOT implement code — that is the implement skill.
A Claude Code plugin for a universal, GitHub-issue-driven agent workflow: the AI writes and grooms the backlog, an autonomous runner works it off with hard per-issue budgets, model tiering and a five-layer verification stack — built for solo developers who want autonomy and working results.
Skill prompts are currently written in German. The workflow itself (labels, commands, config) is language-neutral.
/plugin marketplace add ahlerjam/flowkit
/plugin install flowkit@flowkit
Then, inside each target repository:
/flowkit:setup
The setup is idempotent: it checks companion plugins (superpowers, browser-use,
context7 — optional, reported if missing), creates labels, a per-repo config
(.claude/workflow.config.json), permission/hook templates and (optionally)
a generalized PR-deep-review CI gate. Nothing lands on your default branch
directly — the install arrives as a pull request.
| Action | Command | Result |
|---|---|---|
| Idea → issue | /flowkit:issue impuls "<one sentence>" | Full spec issue (What/Why/Scope/AC + labels); low-risk issues become agent-ready automatically |
| AI finds work | /flowkit:issue gaps <area> [max N] | Spec issues labeled needs-triage — flipping the label to agent-ready is your only mandatory touchpoint |
| Work it off | /flowkit:implement next N | issues A,B | epic N | milestone "X" [max X] | Autonomous run: Planner → Builder (TDD, isolated worktree) → fresh AC verifier → critic → PR deep review → auto squash-merge → post-merge smoke |
| Standalone second opinion | /flowkit:critic <PR> | Cross-vendor review via Codex CLI; without Codex access a narrowly-scoped Claude fallback takes over (clearly marked) |
size/S|M|L label): overruns abort cleanly
with a budget-exceeded label and a draft PR instead of burning tokens.maxFixRounds): exactly one model
escalation, then needs-human — the run moves on, never loops.onSmokeFailure policy (revert by default).Everything lives in .claude/workflow.config.json per repository — commands
(test/lint/typecheck/smoke), protected areas, parallelism, budgets, models,
auto-ready policy, markers and the merge check. See
templates/workflow.config.json.template and the JSON schema next to it.
MIT
Based on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub ahlerjam/flowkit --plugin flowkitModulares akademisches Forschungs-Toolkit mit 28 selbstaktivierenden Skills, Vault-MCP-Server (SQLite + FTS5 + sqlite-vec), Universal Book Fetcher (8-Tier-Pipeline), humanizer-de Anti-KI-Audit, LaTeX-Export, Per-Uni-Profile, 5D-Scoring. Durchsucht Semantic Scholar, CrossRef, OpenAlex, BASE, Google Scholar und 9 weitere wissenschaftliche Quellen.
Harness-native ECC plugin for engineering teams - 67 agents, 279 skills, 94 legacy command shims, reusable hooks, rules, MCP conventions, and operator workflows for Claude Code plus adjacent agent harnesses
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
v9.54.1 — Reliability wave: tangle contextual review correction loop with hard round ceiling, progress-supervised review rounds (per-agent stall watch, descendant-tree kills), council diversity and agy pin fixes, marketplace generator source-of-truth fix, provider troubleshooting runbook and cost-expectations docs. Run /octo:setup.
Persistent file-based planning for AI coding agents. Crash-proof markdown plans (task_plan.md, findings.md, progress.md) that survive context loss and /clear, with an opt-in completion gate and multi-agent shared state. Manus-style. Works with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and 60+ agents via the SKILL.md standard. Includes Arabic, German, Spanish, and Chinese (Simplified and Traditional).