Skip to content
Back to skills

4d Spec

ASecurity

Orchestrator that merges the 4D Paradigm with Spec-Driven Development (SDD). Classifies task complexity and activates SDD phases only when warranted. Use at the START of any implementation task to determine the right workflow depth.

  • 17 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 2, 2026
ai-agentsgogitapidatabase

Works with

  • api

Security analysis

A100/100

Scanned October 1, 2026

npx -y skills add CarlosCaPe/octorato --skill 4d-spec --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of 4d Spec?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for 4d Spec
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/carloscape-4d-spec/badge)](https://www.skillsdirectory.com/skills/carloscape-4d-spec)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: 4d-spec
description: >
  Orchestrator that merges the 4D Paradigm with Spec-Driven Development (SDD).
  Classifies task complexity and activates SDD phases only when warranted.
  Use at the START of any implementation task to determine the right workflow depth.
triggers:
  - new feature
  - implementation task
  - build feature
  - spec driven
  - 4d+s
  - complex change
---

# 4D+S: Spec-Driven 4D Orchestrator

Merges the Octopus 4D Paradigm with SDD (Spec-Driven Development) to get the best of both worlds.

## When to Activate

This skill activates at the START of any task that involves writing or modifying code.
It classifies complexity and routes to the appropriate workflow depth.

## Complexity Classifier

Assess the task against these criteria:

| Signal | Points |
|--------|--------|
| Touches 1-3 files | 0 |
| Touches 4-10 files | +2 |
| Touches 10+ files | +4 |
| New feature (not a fix) | +2 |
| Architectural decision required | +3 |
| Multiple modules/services affected | +2 |
| User explicitly requests spec | +5 |
| Database schema changes | +1 |
| New API endpoints | +1 |

**Score → Workflow:**

| Score | Level | Workflow |
|-------|-------|---------|
| 0-2 | TRIVIAL | 4D only (Describe → Gate → Execute → Diligent → Disclose) |
| 3-5 | MEDIUM | 4D + `plan.md` (task checklist before Gate) |
| 6+ | LARGE | 4D + full SDD (`feature.md` + `plan.md` + analyze + implement ⇄ converge + `review.md` + archive) |

## Workflow by Level

### TRIVIAL (score 0-2)

Standard 4D: no SDD artifacts needed:
1. **1D Describe**: state what and why (1-3 sentences)
2. **2D Delegate**: run delegate-check
3. **4D Gate**: Change Manifest table
4. **Execute**
5. **3D Diligent**: build/lint/test
6. **4D Disclose**: impact + side effects

### MEDIUM (score 3-5)

4D + task checklist:
1. **1D Describe**: state what and why
2. **2D Delegate**: run delegate-check
3. **2S Plan**: generate `plan.md` with numbered tasks (cap: 20 tasks max)
   - Use `/sdd-plan` format but lighter: tasks need no `[AC-##]` brackets
   - Plan lives in working directory, deleted after completion
4. **4D Gate**: Change Manifest + plan.md summary
5. **Execute**: follow plan tasks in order, mark done
6. **3D Diligent**: build/lint/test
7. **4D Disclose**: impact + side effects
8. Clean up: delete plan.md (git tracks the actual changes)

### LARGE (score 6+)

Full 4D+SDD. "Done" is a verdict computed against the spec by a pass that did not write
the code, never a claim by the builder:
1. **1D Describe + Spec**: `/sdd-feature` produces an ears-1 `feature.md` (EARS criteria,
   Glossary, at most 3 open markers)
   - Resolve markers with `/sdd-refine`, one question at a time
2. **2D Delegate**: run delegate-check (loads relevant agents/skills)
3. **2S Plan**: `/sdd-plan` refuses while markers remain, then writes `plan.md` in the task
   grammar `T## [AC-##] <path>: <action>`
4. **2S Analyze**: `/sdd-analyze` as an independent verifier subagent. `FIX-FIRST` sends you
   back to step 1 or 3
5. **4D Gate**: Change Manifest + spec summary + plan summary + analyze verdict
6. **Execute**: `/sdd-implement` (follows plan, verifies each layer, grades nothing)
7. **3D Converge**: `/sdd-converge` as an independent verifier subagent. `GAPS` appends a
   Convergence section to `plan.md`; go back to step 6. Loop until `CONVERGED`
8. **3D Review**: `/sdd-review` (7 quality dimensions, reads the converge verdict)
9. **4D Disclose**: impact radius + converge and review verdicts
10. **Archive**: `/sdd-archive` requires `CONVERGED` and updates `docs/project.md`; the spec stays in its directory

## Integration with Existing 4D

| 4D Phase | SDD Enhancement | When |
|----------|----------------|------|
| 1D Describe | Becomes `feature.md` with ACs and edge cases | LARGE only |
| 2D Delegate | No change: delegate-check still runs | Always |
| 4D Gate | Manifest now INCLUDES plan.md task list | MEDIUM+ |
| 3D Diligent | Adds the converge loop (done = verdict) + 7-dimension quality review | LARGE only |
| 4D Disclose | Adds archive for institutional memory | LARGE only |

## Key Adaptations for Solo Operator

- **Max 20 planned tasks** in any plan.md (tasks above the first `## Convergence` section): if SDD generates more, consolidate
- **No docs/project.md required**: we already have `.claude/CLAUDE.md` per arm
- **A LARGE spec is born in `docs/specs/<yyyymmddHHMM>-<feature-name>/` and never moves**: the converge verdict and the receipt are keyed on that path
- **review.md is optional for MEDIUM**: only mandatory for LARGE
- **`/sdd-yolo`** maps to our "hazlo directo" exception: full pipeline with single gate

## Output Format

At task start, always report:

```
4D+S Classification: [TRIVIAL/MEDIUM/LARGE] (score: N)
  Signals: [list matched signals]
  Workflow: [which phases activate]
  SDD skills: [which /sdd-* commands will be used, or "none"]
```

## Quick Reference

| Want to... | Command |
|------------|---------|
| Classify a task | Read this skill's classifier |
| Write a spec | `/sdd-feature` |
| Refine a spec | `/sdd-refine` |
| Generate plan | `/sdd-plan` |
| Check spec against plan | `/sdd-analyze` |
| Implement from plan | `/sdd-implement` |
| Decide whether it is done | `/sdd-converge` |
| Review code quality | `/sdd-review` |
| Archive completed work | `/sdd-archive` |
| Full pipeline (one gate) | `/sdd-yolo` |

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.