cleanexpo Skill: audit-mode-classifier
Name: audit-mode-classifier Purpose: Classify a repository as Framework, Application, or Hybrid before applying any readiness scoring Type: Capability Uplift Version: 1.0.0 Locale: en-AU
Triggers
Activate this skill when ANY of the following are true:
/auditcommand is invoked- A readiness score is being calculated
- "Finished", "Ready", "Done", "Production ready", or "Launch it" is used
- A "pathway to finished" report is being generated
- A Definition of Done is being assembled
HARD RULE: Classification must happen BEFORE scoring. This skill always runs first.
Inputs
- Repository root directory
- README.md (or equivalent entry document)
- CLAUDE.md if present
- package.json (for project description)
- Directory listing: root, docs/, .claude/, .skills/, apps/
Classification Algorithm
Step 1: Collect Signals
Scan for Framework signals (max 14 pts):
| Signal | Points | Check |
|---|---|---|
| CLAUDE.md with agent routing table | 3 | Grep CLAUDE.md for "agent" |
.skills/ directory with custom SKILL.md files |
3 | Glob .skills/custom/**/*.md |
.claude/agents/ with agent definitions |
3 | Glob .claude/agents/**/*.md |
adopt-project.* or bootstrap.* script in scripts/ |
3 | Glob scripts/adopt*, scripts/bootstrap* |
| "starter" or "template" in README.md title or first paragraph | 2 | Read README.md lines 1-10 |
CONSTITUTION.md or memory.md present |
2 | Glob memory.md, .claude/memory/CONSTITUTION.md |
Hooks system in .claude/hooks/ or .claude/settings.json |
2 | Glob .claude/hooks/ |
| No production deployment configured (intentional absence) | 1 | Absence of vercel.json, netlify.toml, Dockerfile at root |
Scan for Application signals (max 9 pts):
| Signal | Points | Check |
|---|---|---|
| Live production URL referenced in README | 2 | Grep README.md for https:// |
| Business-specific models (orders, billing, CMS, LMS) | 2 | Grep apps/ for "order", "billing", "payment", "cms" |
| Customer-facing feature documentation | 2 | Grep docs/ for "user guide", "customer", "onboarding for end-users" |
| Production deployment guide specific to this repo | 1 | Read docs/production-deployment.md intro |
| apps/web AND apps/backend present as primary project output | 2 | Glob apps/web/, apps/backend/ — present as non-demo |
Step 2: Score
Framework Score ≥ 8 AND Application Score ≤ 3 → FRAMEWORK MODE
Application Score ≥ 8 AND Framework Score ≤ 3 → APPLICATION MODE
Both ≥ 4 → HYBRID MODE
Step 3: Identify Mode-Appropriate "Finished" Criteria
Framework Mode — use criteria from audit-mode-classifier.md §Framework Mode
Application Mode — use criteria from memory.md §2 (default SaaS checklist)
Hybrid Mode — score both layers separately:
- Framework Readiness: framework criteria only
- Template Readiness: template quality criteria only
- Downstream responsibilities: listed but NOT scored as defects
Output Format
## Repository Classification
| Field | Value |
| ----------------------- | -------------------------------- |
| **Mode** | FRAMEWORK / APPLICATION / HYBRID |
| **Primary Purpose** | [one sentence] |
| **Framework Signals** | [Xpts / 14pts max] |
| **Application Signals** | [Xpts / 9pts max] |
### Signals Detected
**Framework:**
- [signal detected — Xpts]
- [signal detected — Xpts]
- ...
**Application:**
- [signal detected — Xpts]
- ...
### Scoring Approach Applied
[For Framework/Application]: [which criteria set will be used]
[For Hybrid]:
- Layer 1 (Framework): scored against framework criteria
- Layer 2 (Template): scored against template quality criteria
- Downstream Items: listed but NOT scored as defects
### Downstream Project Responsibilities
The following items will NOT be scored as defects against this repository.
They are the responsibility of downstream projects that clone/adopt this framework:
- [item 1]
- [item 2]
- ...
Validation Gates
Before producing any readiness score, verify:
- Classification has run (mode is FRAMEWORK, APPLICATION, or HYBRID)
- Scoring criteria match the mode
- Downstream responsibilities are explicitly listed and excluded from scoring
- If HYBRID, two separate scores are generated — never merged into one
Gate failure: If classification was skipped, halt scoring. State: "Classification required before scoring. Running classifier now."
Failure Modes
| Failure | Cause | Recovery |
|---|---|---|
| False "not production ready" | Application Mode applied to Framework repo | Re-run classifier, recategorise |
| Merged score in Hybrid mode | Two layers merged into one score | Split into Framework + Template scores |
| Downstream items marked as defects | Mode not detected before scoring | Re-run classifier, re-categorise items |
| Inflated score | Framework items marked as app-ready without proof | Verify each criterion against framework checklist |
Eval Examples
Good Example (Hybrid Classification)
Repository: NodeJS-Starter-V1
Framework Signals: 14/14 (CLAUDE.md with agents, .skills/71 custom, .claude/agents/23, adopt-project.mjs, "starter" in README, CONSTITUTION.md, hooks system, no production URL)
Application Signals: 4/9 (apps/web + apps/backend present)
Mode: HYBRID
Framework Layer: AI governance engine, 23 agents, 71 skills, adoption pipeline
Template Layer: Next.js 15 + FastAPI starter template
Downstream (NOT defects):
- Production deployment → downstream project responsibility
- SSL/TLS certificates → downstream project responsibility
- Database backups → downstream project responsibility
- Legal pages → downstream project responsibility
- Payment flows → downstream project responsibility
Bad Example (Category Error)
❌ WRONG: "NodeJS-Starter-V1 is 38% production ready because it lacks:
- live production deployment [CRITICAL]
- SSL/TLS certificates [CRITICAL]
- database backups [CRITICAL]
- legal pages [CRITICAL]"
✓ CORRECT: "NodeJS-Starter-V1 Framework Readiness: 90%
Template Readiness: 75%
Items below are downstream project responsibilities (not defects):
- production deployment, SSL, backups, legal pages"