agentleFS
amrmrkjramrmrkjr/m3j-workspace/.agents/skills/m3j-workflow/SKILL.md

m3j-workflow

Skill0 starsChanged 3 months agoMITView on GitHub

Orchestrate a full AI coding workflow end-to-end — SPEC, ROADMAP, TASKS, test-first scaffolding, one-phase execution, quality gates, one small PR, and fresh-context review — free-tier models only. Use when the user says "m3j workflow", "AI coding workflow", wants a spec→roadmap→tasks pipeline, or asks to build something end-to-end with quality gates and one reviewable change per PR.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Check which of AGENTS.md, SPEC.md, ROADMAP.md, TASKS.md already exist.

    line 29
  2. Very specific:

    Run /finish (dispatches m3j-sentinel, 7-step quality gate).

    line 53
  3. Very specific:

    Keep SPEC.md, ROADMAP.md, TASKS.md each under ~200 lines.

    line 91
  4. Distinctive:

    If the user asks to skip a gate for cost, say so plainly: the gates are the point.

    line 94

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

M3j Workflow (free-tier port)

An end-to-end AI coding workflow for OpenCode — free-tier models only. The pipeline turns an idea into a reviewed, mergeable PR through small phases — each phase ships as one reviewable change.

Principles

The tool is replaceable, the gates are not. Any model can write code. The value lives in the gates — test scaffold first, /finish, fresh-context review, human merge check. Never weaken or skip a gate to save tokens.

Reuse, don't duplicate. This repo already has the agents and skills for every step. Your job is to dispatch them in the right order with the right inputs, not to reimplement their behavior. Map each step to the existing agent (table below).

Token economy. Free-tier context is precious. Small files, small phases, one phase per session, compact handoffs — pass paths and acceptance commands, never whole-file dumps or conversation history.

The pipeline

Working files: SPEC.md, ROADMAP.md, TASKS.md at the project root, from templates/. Run steps in order; each step's done-criterion must be met before the next starts.

Step 0 — Inventory

Check which of AGENTS.md, SPEC.md, ROADMAP.md, TASKS.md already exist. Create or refresh the missing ones. Done: all four files present, or a missing one is deliberately deferred and that call is recorded in the SPEC.

Step 1 — SPEC

Dispatch m3j-planner to write SPEC.md from templates/SPEC.md: problem, users, required behavior, architecture, security, PINNED dependency versions, non-goals, acceptance criteria. Done: every acceptance criterion is testable (a command or a human action), and every dependency has an exact pinned version — no ^ ranges for the risky ones.

Step 2 — ROADMAP

Dispatch m3j-planner to write ROADMAP.md from templates/ROADMAP.md: phases with exit criteria and dependencies, ≤3 tasks per phase. Done: every phase has a checkable exit criterion and its dependencies on other phases are explicit.

Step 3 — TASKS

Dispatch m3j-planner to write TASKS.md from templates/TASKS.md: vertical slices (each task delivers working value alone), an acceptance command + expected output per task, TDD checkbox steps. Done: every task has an observable acceptance command.

Step 4 — Test scaffolding first

Dispatch m3j-tester to set up the harness — test runner, lint, format, typecheck — BEFORE any feature code. Done: the harness runs green on an empty/trivial baseline and the TASKS.md acceptance commands are runnable.

Step 5 — Execute one phase

Dispatch m3j-executor to implement ONE phase from TASKS.md, TDD. One phase per session — never the whole roadmap. Done: the phase's tasks pass their acceptance commands and the phase exit criterion is met.

Step 6 — Local gate

Run /finish (dispatches m3j-sentinel, 7-step quality gate). Done: all 7 steps reported pass/fail, and every failure is fixed and re-verified.

Step 7 — One PR

Dispatch m3j-devops to open one small, reviewable PR for this phase. Done: the PR contains only this phase's change and its description names the phase plus its acceptance criteria.

Step 8 — Independent review, fresh context

Dispatch a NEW m3j-triage or m3j-sentinel session — a separate background agent with no memory of the implementation — to review the PR. Never the implementing session; you never grade your own work. Done: the fresh session returns a verdict with specific findings.

Step 9 — Fix/retest loop

For each finding, dispatch m3j-executor to fix, then re-run the affected acceptance commands and /finish. Route through m3j-triage (loads receiving-code-review): verify each suggestion is technically sound before implementing — don't blindly accept. Done: findings resolved or rejected with reasons, and all gates re-pass.

Step 10 — Human merge gate

Present the PR with this checklist. Nothing merges without human approval.

  • Diff contains only the intended change
  • Tests + CI pass on the latest commit
  • Review threads resolved
  • Docs match behavior
  • Real-world manual check done

Dispatch map (reuse-not-duplicate)

Pipeline stepAgentSkill it already covers
SPEC, ROADMAP, TASKSm3j-plannerbrainstorming, grilling, to-prd, to-issues, wayfinder
Test scaffoldingm3j-testertest-driven-development, webapp-testing
Phase executionm3j-executorexecuting-plans, subagent-driven-development, using-git-worktrees
Local gatem3j-sentinel/finish, code-review, security-review, verification-before-completion
PRm3j-devopsfinishing-a-development-branch, resolving-merge-conflicts
Fresh-context reviewm3j-triage / m3j-sentinelrequesting-code-review, receiving-code-review, triage
Fix loopm3j-executor + m3j-triagereceiving-code-review, verification-before-completion

Cost-conscious rules

  • Models are already pinned in the agent files — planning/gates on nemotron-3-ultra-free (m3j-planner, m3j-triage), execution on deepseek-v4-flash-free (m3j-executor, m3j-devops), tests on north-mini-code-free (m3j-tester). Don't override them.
  • Keep SPEC.md, ROADMAP.md, TASKS.md each under ~200 lines. If a phase outgrows that, split the phase.
  • One phase per execution session. Never feed the whole roadmap into one context.
  • Compact handoffs: give sub-agents file paths, acceptance commands, and only the phase's tasks — not full file dumps.
  • If the user asks to skip a gate for cost, say so plainly: the gates are the point. Skipping one is a human decision, made explicitly.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.