agentleFS
agentik-osagentik-os/claude-code-skills/stepper-os/SKILL.md

stepper-os

Skill1 starsChanged 55 days agoMITView on GitHub

Drive a project through Stepper OS (AgentikOS suite) - the execution operating system around coding agents. Compiles a Blueprint into a dependency-aware graph of modules/epics/slices/steps, then executes it step by step through the `omega-stepper` CLI: plan -> start -> implement -> verify -> done, where DONE is gated by a deterministic verifier (never self-report). Use when the user says "/stepper-os", "/omg-stepper-os", "stepper", "execute the blueprint", "run the steps", "next step", "step plan", or in French "execute le blueprint", "etape par etape", "prochaine etape", "lance le stepper". NOT for designing the blueprint itself (that is /blueprint-os or /omg-blueprint-os) and NOT a generic todo list (that is the harness task tool).

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When the project has a Blueprint (e.g.

    line 59
  2. Very specific:

    Read the Blueprint documents (blueprint/) AND, when the project went through Design OS, the Design Handoff (design/).

    line 61
  3. Very specific:

    Write one YAML file per spec under stepper/modules/, stepper/epics/, stepper/slices/, stepper/steps/.

    line 68
  4. Distinctive:

    Every step must be independently executable: contract, context files, acceptance checks, definition of done.

    line 75

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Stepper OS - execute a Blueprint step by step, verification-gated

Stepper OS is NOT the coding agent. It is the execution operating system around you: it owns the sequence (planner), the truth (tracker), and the definition of done (verifier). You implement; Stepper decides what is next and what is proven.

The full doctrine lives in the pack: OS/stepper-os/pack/ in the OmegaOS repo (installed at ~/.omega/os/stepper-os/pack/). Load 12_AGENT_OPERATING_PROTOCOL.md as your operating rules for the whole session. The CLI is omega-stepper (auto-installs its venv on first run).

The loop you follow

  1. Locate or create the project.
    • Existing Stepper project: a stepper.yaml at the project root.
    • New project: omega-stepper init --name <name> then compile the Blueprint (see "Compiling a Blueprint" below).
  2. Recover exact state first (never trust conversation memory):
    omega-stepper resume     # reconcile interrupted attempts
    omega-stepper status     # weighted + raw progress
    omega-stepper plan       # ranked READY candidates + safe wave
  3. Claim the top wave step: omega-stepper start <STEP-ID> - it prints the full agent brief (contract, blueprint refs, invariants, forbidden changes, commands, acceptance checks). READ every context file it names before editing.
  4. Implement the contract. One step, one contract - never widen scope. Include the required tests, errors, security, docs the brief names.
  5. Close through the verifier: omega-stepper done <STEP-ID>.
    • PASS -> step is DONE, go back to plan and keep going (protocol rule 13: do not stop at arbitrary milestones).
    • FAIL -> repair against the printed evidence and run done again. The attempt ceiling (max_fix_attempts) is a hard stop: when start refuses with "escalate", stop and hand the failure to the operator.
  6. Review gates: steps with review_roles need a recorded review: omega-stepper review <STEP-ID> <role> PASS --by <name>. Never record a review you did not actually perform; ask the operator when a human gate (security, architecture) is required.
  7. Finish = release gate, not vibes: omega-stepper release-check must PASS. Report progress with omega-stepper report.

Compiling a Blueprint into steps

When the project has a Blueprint (e.g. from /blueprint-os) but no steps yet:

  1. Read the Blueprint documents (blueprint/) AND, when the project went through Design OS, the Design Handoff (design/). Declare both in stepper.yaml under sources: (blueprint.root/handoff and design.root/handoff) so every step's references resolve to real files.
  2. Derive modules -> epics -> vertical slices -> atomic steps per pack/00_MASTER_SPEC.md and the schemas in pack/03_STEP_CONTRACT_SPEC.md (full example: pack/10_EXAMPLE_STEP.yaml).
  3. Write one YAML file per spec under stepper/modules/, stepper/epics/, stepper/slices/, stepper/steps/. Each step names the exact upstream docs that govern it — blueprint_references (product truth: WHAT/WHY, requirement/decision/capability ids) AND design_references (UX/UI truth: flow / screen / surface / state ids from the Design Handoff). A UI-touching step with no design_references is almost certainly missing its design docs.
  4. Step granularity: one focused agent cycle (~15 min to ~2 h human-equivalent). Every step must be independently executable: contract, context files, acceptance checks, definition of done. A vague step is invalid - refine it.
  5. omega-stepper validate must pass — it checks schema + the acyclic graph AND audits that every step's Blueprint/Design references resolve to real docs under their source roots (and warns when a UI step cites no design ref). The agent brief (omega-stepper start <id>) then loads BOTH the Blueprint and Design references so the coding agent reads the right source of truth.

Hard rules (from the pack, enforced by the engine)

  • Tracker owns progress; planner owns sequence. Never freelance the order.
  • DONE only through omega-stepper done - the verifier has final say.
  • Dependencies are authoritative; start refuses steps that are not READY.
  • Repair against evidence; never pivot to unrelated work after a failure.
  • Preserve existing user work; never reset the repo to simplify your task.
  • Blueprint drift is forbidden: a canonical decision change is a decision request to the operator, not a silent redesign.

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.