Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    ScrumAI team — scrum-master

    1

    Run a Scrum team of AI role agents over any backlog: sprint planning, dispatch, daily, review, retro, on the ex-workflow engine.

    Free

    Security scanned
    ScrumAI team — scrum-master

    ScrumAI team — scrum-master

    Example session with this skill installed

    Run sprint 2 on this backlog. Sprint Goal: "A host-agent user can install the workflowAI skill set with a reviewed install script that prints usage on --help and on unknown arguments, and can follow a README to do it." Commit INSTALL-SCRIPT, INSTALL-HELP and README; budget 400 000 tokens; dispatch workers on haiku, two at a time.

    • Read your context and instructions
    • Compiled the scrumai team

    Sprint 2 planning

    Sam (Scrum Master): Sprint Goal proposal (Nok's): "A host-agent user can install the workflowAI skill set with a reviewed install script that prints usage on --help and on unknown arguments, and can follow a README to do it." Nok, the backlog in order?

    Nok (Product Owner): INSTALL-SCRIPT (weight 1, priority 30, review only); INSTALL-HELP (weight 1, priority 25, after INSTALL-SCRIPT); README (weight 1, priority 20, after INSTALL-SCRIPT); CHANGELOG-FIX (weight 1, priority 5) — not this sprint.

    Sam: Budget { tokens: 400000 }. Committed: INSTALL-SCRIPT, INSTALL-HELP, README (3 points).

    wf ceremony plan .workflow/inbox/plan2.json --by scrum-master → sprint.opened S2

    Round 1

    • wf next --max 2 → one brief: INSTALL-SCRIPT (carried over; its sprint-1 handoff arrives as priorHandoffs, the three team agreements as invariants).
    • Kit (developer, haiku, 32.5 s, 71 191 tokens): verified all five acceptance criteria with its own runs (pnpm check 19/19, dry run, three bad-argument cases); lesson L5 (spec): the acceptance text lacks the --host/--scope defaults. Verdict done → item moves to review.
    • A fresh Kit reviews (25.7 s, 69 146 tokens): all five criteria hold; wf complete INSTALL-SCRIPT → done.

    Daily 1

    Sam: Day 1. 0 in progress, 0 in review, 1 done (INSTALL-SCRIPT), 0 blocked; INSTALL-HELP and README to do. Remaining weight 2; forecast at-risk. Spent 69 146 tokens of 400 000 in 2 completions.

    Sam: 1 lesson raised so far (Kit on INSTALL-SCRIPT, L5); we take it to the retro.

    • wf next --max 2 → INSTALL-HELP and README, dispatched to two Kits in parallel.
    • Kit on README (28.2 s, 67 191 tokens): README.md, 68 lines; flags and defaults read from install.mjs; pnpm check 21/21. Verdict done.
    • Kit on INSTALL-HELP (37.6 s, 71 228 tokens): usage() with every flag and default; --help → stdout exit 0; unknown argument → error + usage on stderr exit 1; two new tests. Lesson L6 (spec): the acceptance did not name the output streams. Verdict done.
    • Two fresh reviewers; one handoff block was malformed (/lessons/0/lesson unknown property) — the engine refused it with a fix, Sam sent the error back once, the corrected block arrived. Both items done.

    Daily 2

    Sam: Day 2. 0 in progress, 0 in review, 3 done, 0 blocked. Remaining weight 0; forecast on-track. Spent 337 221 tokens of 400 000 in 6 completions.

    Sam: 3 lessons raised so far; we take them to the retro.

    Sprint Review (human gate)

    Sam: Increment for sprint S2:
    — INSTALL-SCRIPT: done (test-output: pnpm check 21/21; reviewed)
    — INSTALL-HELP: done (test-output: --help stdout exit 0, unknown argument stderr exit 1; reviewed)
    — README: done (test-output: pnpm check 21/21; reviewed)
    Nok: Accept the S2 increment: the Sprint Goal is met; CHANGELOG-FIX is the remaining step.

    wf ceremony review .workflow/inbox/review2.json --by human → sprint.closed S2, reason review, done 3/3.

    Next steps

    • Retro: Nok's lessons (extend agreement 3 with streams and exit codes; review briefs list untracked files) become working agreements that every later brief carries.
    • wf metrics after close: velocity, commitment reliability, carryover ratio, budget utilisation, cost per point.
    • Plan sprint 3 for CHANGELOG-FIX.
    README.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    ScrumAI team runs a Scrum team of AI role agents inside your coding agent (Claude Code, Codex CLI, Cursor or any host that can run a prompt and node). The orchestrator is the scrum-master skill; the product-owner and developer role skills are nested under roles/, so one install gives you a working team.

    What it does

    • Seeds a backlog from a Product Goal: the Product Owner proposes items with acceptance criteria, weights, priorities and dependencies; you edit, the engine stores them.
    • Plans a sprint within a token budget (plan auto fills it in priority order and reports what it skipped).
    • Dispatches briefs to developer workers: each brief carries the acceptance criteria, the team's working agreements, upstream results and earlier attempts; a review always goes to a fresh worker, never the author.
    • Records handoffs with evidence; the engine refuses a completion that lacks the evidence the Definition of Done asks for, and refuses zero-token accounting.
    • Narrates daily, review and retro from real engine state only — every sentence traces to a wf command's output.
    • Learns: retro lessons become working agreements, Definition of Done changes, backlog items or team memory that every later brief carries.

    How it works

    The engine is scripts/wf.js, a bundled build of the ex-workflow state machine (Scrum rules, planner, budgets, gates; published on npm as ex-workflow, MIT). All state lives in .workflow/ in your project; nothing leaves your machine. The skill is prose: it tells the host which wf command to run, how to build each worker prompt, and what to say to you.

    What's inside

    SKILL.md (the orchestrator, ≤ 100 lines), policy.json (Scrum config template, personas, host model map), reference/ (every command with its JSON shapes, ceremony narration, the worker contract, team voice), roles/product-owner/, roles/developer/ (each a SKILL.md plus the engine role manifest), scripts/wf.js + scripts/package.json.

    Why not just prompt it yourself

    Prompted agents forget dependencies, skip verification and review their own work. Here the engine enforces who may move an item, what evidence it needs, when a sprint may close and who may approve — the orchestrator cannot talk its way around a rule.

    Known limitations

    Budgets are sized from the host's token counter; a host that reports no usage runs with estimates (--estimate) corrected later with wf account. The shipped Definition of Done (test-output) is the software preset; any domain names its own evidence kinds at init. Dogfooded on three real sprints; Node.js 18+ required.

    Support

    Message the creator on Agensi; updates ship as new listing versions.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Budgets are sized from the host's token counter; a host that reports no usage runs with estimates (--estimate) corrected later with wf account. Agreements and team memory live in .workflow/ in your repo. The shipped Definition of Done (test-output evidence) is the software preset; other domains name their own evidence kinds at init.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 18 days ago

    • Free with an Agensi account

    Software
    Node.js 18 or laterGit

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    .workflow/**

    The bundled scripts/wf.js is the MIT ex-workflow engine; it reads and writes only .workflow/ under the project root. No network.

    Listed24 days ago
    Updated15 days ago

    What's inside

    Frequently Asked Questions

    A prompted "Scrum Master" has no state: it forgets dependencies, lets a worker review its own work and closes a sprint whenever it feels done. Here the bundled ex-workflow engine (scripts/wf.js) holds the state machine: it decides which item may move, who may approve, what evidence the Definition of Done needs and when a sprint may close. The orchestrator can only run wf commands and narrate what they return.

    SKILL.md (the scrum-master orchestrator), policy.json (Scrum config template, team personas, host model map), reference/ (commands.md with every wf command and its JSON shapes, ceremonies.md narration scripts, handoff.md worker contract, team.md), roles/product-owner/ and roles/developer/ (a SKILL.md plus the engine role manifest each), and scripts/wf.js with scripts/package.json (the bundled MIT engine, CommonJS). 13 files, no network, all state under .workflow/.

    Any host that follows the SKILL.md standard, can run a prompt and has Node.js 18+ on PATH. Tested on Claude Code, where workers are dispatched as subagents; a host without subagents plays each role inline with the same prompt. The model map in policy.json names small/medium/large tiers (defaults never use the largest models); a non-Claude host maps them to its own models.

    Each sprint has a token budget in policy.json (sprintBudget.tokens). Every completion is recorded with the host's real token figures (wf complete --tokens-in/--tokens-out); zero figures are refused, and a host that reports nothing uses --estimate and corrects later with wf account. The planner prices candidates per call from the sprint's accounting history (estimateMode per-call) and stops dispatching when the remaining budget cannot cover the next item; with autoClose off the sprint waits for your review instead of closing itself. Budgets can be changed between sprints with wf config set.

    The engine is the npm package ex-workflow (MIT); each listing version bundles one engine version into scripts/wf.js. When the engine or the skill prose improves, a new listing version is published here and buyers get it as an update; your .workflow/state.json files load unchanged across engine versions (every new field is optional). For problems or questions, message the creator on Agensi.

    New in Agents & Orchestration

    Free