AgentFlow: Kanban-based autonomous dev pipeline with Claude Code skills

[UrRhb]

What problem are you trying to solve?

AI coding tools are either chatbots that need constant supervision or black-box agents that lack visibility. There's no reliable way to orchestrate multiple Claude Code workers across a full development lifecycle (build, review, test, integrate) with proper quality gates, cost tracking, and crash recovery — all visible from a standard project management tool.

Proposed idea

Consider adding AgentFlow (https://github.com/UrRhb/agentflow) to the collection. It's an orchestrator that turns your Kanban board into an autonomous AI dev pipeline with Claude Code.

It includes these skills/prompts:

• /spec-to-board — Decomposes a SPEC.md into atomic tasks on your Kanban board
• • /sdlc-orchestrate — Dispatches tasks to workers based on transitive priority
• • /sdlc-worker --slot T2 — Runs a worker that picks up tasks, builds code, creates PRs
• • /sdlc-health — Real-time pipeline status dashboard
• • /sdlc-stop — Graceful shutdown
    Key differentiators:
• Uses your Kanban board (Asana, GitHub Projects, Linear) as the state machine — no custom infra
• • Stateless crontab orchestrator (crash-proof)
• • Deterministic quality gates (tsc/eslint/tests) before AI review
• • Adversarial review: reviewer must find 3 issues before passing
• • Per-task cost tracking with automatic guardrails
• • Auto-revert on integration failure
• • 45 failure modes identified and addressed
    MIT licensed. Works as a Claude Code native plugin or standalone.

Real use cases

• Solo dev building a SaaS: Write a SPEC.md, run /spec-to-board, start 3-4 workers, and monitor progress from your phone while tasks flow through build, review, test, and merge stages
• • Team running multiple Claude Code projects: Use portfolio view in Asana to track multiple pipelines, with per-task cost tracking ensuring budgets aren't exceeded
• • Developer wanting CI-quality gates on AI-generated code: Deterministic tsc/eslint/test gates catch ~60% of issues before any AI review, with adversarial review catching the rest

Related skills, tools, or prior art

• Similar skill: Task orchestration, SDLC automation
• • Related doc: https://github.com/UrRhb/agentflow/blob/main/README.md
• • External reference: https://github.com/UrRhb/agentflow

Are you willing to help build this?

Yes, I can open a PR

Before posting

• I searched existing discussions and issues for similar ideas.

[sickn33]

thanks, make s PR.

[emanueleodierna729-ship-it]

Ricordati chi e l'artefice....