Simplified workflow
Lightweight workflow for small routine changesFrom its SKILL.md
npx -y skills add statsclaw/statsclaw --skill simplified-workflowAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
6.2 KB, as published. Nobody here has run it
Skill: Simplified Workflow (Workflow 10)
A lightweight workflow for small, well-understood changes where the full two-pipeline architecture adds overhead without proportional safety benefit.
When to Offer
Leader MUST evaluate every incoming request against these smallness criteria before committing to a full workflow. If ALL of the following are true, leader SHOULD offer the simplified workflow:
| Criterion | Test |
|---|---|
| Few files | Change touches ≤3 files (estimated during planning) |
| Low risk | No algorithmic changes, no numerical methods, no public API changes |
| Clear scope | Request is unambiguous — no uploaded files, no papers, no formulas |
| Well-understood | The fix/change pattern is routine (typo, config tweak, simple bug fix, dependency bump, adding a parameter, formatting) |
| No theoretical work | No mathematical derivation, no statistical methodology, no specification needed |
Examples of Small Requests
- Fix a typo in a function name or docstring
- Bump a version number in DESCRIPTION/package.json/Cargo.toml
- Add a missing
@exporttag - Fix a linting error
- Add a simple parameter with a default value
- Update a URL in documentation
- Fix a broken import/require path
- Add
.Rbuildignoreentries
Examples That Are NOT Small (use full workflow)
- Implement a new estimator or algorithm
- Fix a numerical bug that changes outputs
- Refactor a module's public API
- Any change where the user uploaded reference material (PDFs, papers)
- Changes that affect test outcomes or coverage
How to Offer
When leader detects a small request, it MUST ask the user before proceeding:
Leader uses AskUserQuestion with:
question: "This looks like a small change (≤3 files, routine pattern). Use simplified workflow?"
options:
- label: "Simplified (faster)"
description: "Skip planner/scriber. Direct: plan → build → validate → ship."
- label: "Full workflow"
description: "Run the complete two-pipeline architecture with all teammates."
If the user chooses "Full workflow" or provides a custom answer suggesting thoroughness, leader proceeds with the standard workflow (1–9).
If the user chooses "Simplified", leader uses Workflow 10.
If leader is uncertain whether the request is small, it MUST ask. The default is always to offer the choice — never silently downgrade to simplified.
Workflow 10: Simplified Pipeline
Agent Sequence
leader → builder → tester → shipper?
No planner, no scriber, no reviewer. Leader writes a lightweight spec directly in request.md (extended with acceptance criteria).
State Model (Simplified)
CREDENTIALS_VERIFIED → PLANNED → PIPELINES_COMPLETE → REVIEW_PASSED → DONE
Skipped states: SPEC_READY, DOCUMENTED. No comprehension.md, spec.md, test-spec.md, ARCHITECTURE.md, docs.md, or review.md from reviewer.
Steps
- CREDENTIALS — Same as full workflow (step 4 in CLAUDE.md). Hard gate.
- PLAN — Leader writes
request.mdwith:- Clear description of the change
- Acceptance criteria (what "done" looks like)
- Affected files (the write surface)
- Expected validation outcome
Leader writes
impact.mdas in the full workflow. Status →PLANNED.
- BUILD — Dispatch builder with
isolation: "worktree". Builder receivesrequest.mdas its spec (no separatespec.md). Builder implements and writesimplementation.md. Status →PIPELINES_COMPLETE. - VALIDATE — Dispatch tester. Tester runs profile validation commands and verifies acceptance criteria from
request.md. Tester writesaudit.md. If BLOCK → respawn builder (max 3). Tester acts as the quality gate (replaces reviewer for simplified workflow).- For audit pass: status →
REVIEW_PASSED.
- For audit pass: status →
- SHIP (if requested) — Dispatch shipper. Status →
DONE.
Preconditions (Simplified)
| Target State | Precondition |
|---|---|
PLANNED | request.md with acceptance criteria AND impact.md exist |
PIPELINES_COMPLETE | implementation.md exists |
REVIEW_PASSED | audit.md exists with verdict PASS |
DONE | shipper.md exists (if ship requested) |
What Is Skipped
| Component | Reason |
|---|---|
| planner | No theoretical analysis needed for routine changes |
| scriber | No architecture/process-record needed for small changes |
| reviewer | Tester provides sufficient quality gate |
spec.md / test-spec.md | Builder uses request.md directly; tester validates against acceptance criteria |
comprehension.md | No uploaded material to comprehend |
ARCHITECTURE.md | Small change doesn't warrant architecture documentation |
| Log entry | No process record for routine changes |
Escalation
If at any point during the simplified workflow:
- Builder raises HOLD (spec ambiguity) more than once
- Tester blocks more than once on the same issue
- The change turns out to affect more files than estimated
Leader MUST escalate to the full workflow. This means:
- Dispatch planner with all context accumulated so far
- Continue from
SPEC_READYonward using the full pipeline - Update
status.mdto reflect the escalation
Progress Bar for Simplified Workflow
┌─────────────────────────────────────────────────────────────────────┐
│ StatsClaw Progress (simplified) [2/4] │
│ │
│ [✔] Plan ── [▶] Implement ── [ ] Validate ── [ ] Ship │
│ │
│ ▶ Active: builder implementing changes... │
└─────────────────────────────────────────────────────────────────────┘
See skills/progress-bar/SKILL.md for full progress bar specification.