ils15 Metis Gap Analysis — Pre-Plan Validation
Use this skill to analyze implementation plans BEFORE they are delivered to the user. Catches what the planner missed: hidden intentions, ambiguities, missing criteria, over-engineering, and edge cases.
Named after Metis, the Titaness of wisdom and deep thought — she forces externalization of implicit knowledge.
The Core Principle
The plan author has ADHD working memory. Force the gaps into the open.
When Athena generates a plan, implicit knowledge never makes it onto the page. Metis catches these gaps before the user approves the plan.
When to Run
Metis runs AFTER Athena generates the plan, BEFORE delivering to the user for approval.
Athena generates plan
↓
Metis analyzes (this skill)
↓
Gaps injected into plan
↓
Athena revises plan with gaps addressed
↓
Plan delivered to user for approval
Gap Categories
1. Hidden Intentions
What the user wanted but didn't say:
| Signal | Question to Ask |
|---|---|
| Vague scope ("improve performance") | "What metric defines 'improved'? Latency? Throughput? Memory?" |
| No rollback mentioned | "What's the rollback strategy if this breaks?" |
| No migration path | "How do we migrate existing data?" |
| No monitoring mentioned | "How will we know if this works in production?" |
2. Ambiguities
Terms that could derail implementation:
| Signal | Question to Ask |
|---|---|
| "Fast", "scalable", "robust" | "Define the target: <100ms? 10k req/s? 99.9% uptime?" |
| "Following existing patterns" | "Which pattern? There are 3 different auth patterns in the codebase." |
| "When ready", "as needed" | "What triggers readiness? What's the threshold?" |
| No scope boundaries | "What is explicitly OUT of scope?" |
3. Missing Acceptance Criteria
How to know it worked:
| Signal | Question to Ask |
|---|---|
| No test strategy | "What tests verify this works?" |
| No success metrics | "What does 'done' look like?" |
| No verification steps | "How does the user verify this in production?" |
| No error scenarios | "What happens when X fails?" |
4. AI Slop Patterns
Over-engineering and scope creep:
| Signal | Action |
|---|---|
| 10+ phases for a simple feature | "Can this be done in 3-5 phases?" |
| Abstract interfaces for single impl | "Do we need this abstraction?" |
| "Future-proof" without current need | "YAGNI — remove unless needed now" |
| Unnecessary microservices | "Can this be a module instead?" |
5. Edge Cases
What wasn't covered:
| Signal | Question to Ask |
|---|---|
| No error handling plan | "What happens on timeout? Network failure? Invalid input?" |
| No rate limiting | "What if this endpoint is hammered?" |
| No pagination for lists | "What if there are 1M records?" |
| No cache invalidation | "When does the cache stale?" |
Output Format
When gaps are found, inject into the plan:
## ⚠️ Gaps Identified (Metis Analysis)
### Hidden Intentions
- [ ] User said "improve performance" but didn't specify target metric
→ Recommendation: Define target (e.g., <200ms p95 latency)
### Ambiguities
- [ ] "Following existing patterns" — codebase has 3 different patterns
→ Recommendation: Specify which pattern (e.g., Repository pattern from users/)
### Missing Acceptance Criteria
- [ ] No test strategy defined
→ Recommendation: Add test requirements per phase
### AI Slop Patterns
- [ ] 8 phases for a simple CRUD feature
→ Recommendation: Consolidate to 3-4 phases
### Edge Cases
- [ ] No error handling for database connection failure
→ Recommendation: Add retry + fallback strategy
Clearance Checklist
Before declaring a plan complete, verify:
- Core objective is clearly defined
- Scope boundaries are established (what's IN and OUT)
- No critical ambiguities remain
- Technical approach is decided
- Test strategy is confirmed
- Rollback strategy exists
- Acceptance criteria are measurable
- Edge cases are addressed
If ANY item is unchecked → inject the gap into the plan.
Integration with Athena
Athena should run Metis analysis as a sub-step:
1. Athena generates initial plan
2. Athena applies Metis Gap Analysis (this skill)
3. Athena revises plan to address gaps
4. Athena delivers revised plan to user
This is NOT a separate agent — it's a skill that Athena applies to her own plan before delivery.