mattgierhart/PRD-driven-context-engineering / .claude/skills/ghm-harvest
ghm-harvest
VerifiedExtracts durable insights from temp/ files to SoT during EPIC Phase E. Triggers at EPIC completion or explicit `/ghm-harvest` invocation. Outputs new SoT entries and archive manifest.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.5 KB |
| assets/harvest-log-template.md | markdown | 4.2 KB |
| references/examples.md | markdown | 5.2 KB |
| references/harvest-patterns.md | markdown | 7.6 KB |
references/harvest-patterns.md
View on GitHub# Harvest Patterns
**Purpose**: Decision framework for extracting temp/ insights to SoT, archiving context, and deleting noise.
---
## Harvest Philosophy
**Core Principle**: "Temp is scratch, SoT is durable"
**Why Harvest**:
- Prevent SoT contamination (methodology in templates)
- Preserve insights (don't lose learnings)
- Clean workspace (fresh start for next EPIC)
**When to Harvest**: End of EPIC Phase E (before closing EPIC)
---
## Three-Way Sort: Extract | Archive | Delete
### Extract → SoT (Durable Knowledge)
**Decision Rule**: "Will this be useful beyond this EPIC?"
**Examples**:
- ✅ User pain points (CFD)
- ✅ Architecture decisions (ARC, BR)
- ✅ API contracts (API)
- ✅ Customer quotes (CFD)
- ✅ Competitive insights (CFD)
**Anti-Examples**:
- ❌ "Tried approach X, didn't work" (contextual, archive instead)
- ❌ Daily progress notes (contextual, archive)
- ❌ Debugging session logs (noise, delete)
---
### Archive (Contextual Value)
**Decision Rule**: "Useful for retrospective, but not general SoT"
**Examples**:
- ✅ EPIC timeline (what happened when)
- ✅ Blockers and resolutions (how we solved issues)
- ✅ Team discussions (context for decisions)
- ✅ Experiment results (A/B test data)
**Archive Location**: `archive/YYYY-MM/epic-XXX-description.md`
**Purpose**: Post-mortem, team learning, future retrospectives
---
### Delete (No Future Value)
**Decision Rule**: "Will anyone ever need this again?"
**Examples**:
- ✅ Debug logs (issue resolved)
- ✅ Scratch calculations (final number in SoT)
- ✅ Duplicate notes (consolidated version in SoT)
- ✅ Superseded drafts (final version in SoT)
**Safety Check**: Review once before deletion (can't recover)
---
## Extraction Patterns
### Pattern 1: User Research → CFD
**Temp File**:
```markdown
# User Interview 5
Date: 2026-01-10
User: Sarah (Design Lead at startup)
Pain Points:
- "I spend 2 hours/day searching for design files" (frustrated)
- "Our team uses Dropbox, Figma, and Slack - files everywhere" (chaotic)
Needs:
- Unified search across tools
- Quick preview without downloading
```
**Extract to SoT**:
```markdown
# CFD-092: File Search Friction
**Type**: Pain Point
**Source**: User interview (Sarah, Design Lead, 2026-01-10)
**Evidence Tier**: Tier 1 (Direct quote)
**Pain**: "I spend 2 hours/day searching for design files across Dropbox, Figma, and Slack"
**Impact**:
- Time: 2 hours/day wasted (25% of workday)
- Emotion: Frustrated ("files everywhere")
**Implication**: Unified search is high-value feature
```
**Delete**: `temp/user-interview-5.md` (insights extracted)
---
### Pattern 2: Technical Decision → BR/ARC
**Temp File**:
```markdown
# Database Choice
Evaluated: PostgreSQL vs MongoDB
PostgreSQL Pros:
- Strong ACID guarantees
- Mature ecosystem
- Team familiar
MongoDB Pros:
- Schema flexibility
- Faster for document storage
Decision: PostgreSQL (data integrity > schema flexibility for our use case)
```
**Extract to SoT**:
```markdown
# BR-095: Database Technology Choice
**Decision**: PostgreSQL
**Alternatives Considered**: MongoDB
**Rationale**:
- ACID guarantees critical (financial data)
- Team expertise (faster development)
- Schema stability (known data model)
**Trade-off**: Less schema flexibility, but acceptable (data model stable)
```
**Delete**: `temp/database-choice.md` (decision captured in BR)
---
### Pattern 3: Experiment Result → Archive
**Temp File**:
```markdown
# A/B Test: Onboarding Flow
Variant A (Current): 40% activation
Variant B (New flow): 55% activation
Winner: Variant B
Rolled out: 2026-01-12
```
**Archive**:
```markdown
# archive/2026-01/ab-test-onboarding.md
[Same content]
Archived because:
- Experiment complete, winner chosen
- Useful for future onboarding changes (learn what worked)
- Not general SoT (specific to this time period)
```
**Do NOT Extract to SoT**: Time-bound experiment, not durable principle
---
### Pattern 4: Daily Notes → Archive or Delete
**Temp File**:
```markdown
# EPIC-012 Daily Log
Day 1: Started API design, blocked on auth token format
Day 2: Resolved auth, drafted endpoints
Day 3: Code review, updated schema
Day 4: Tests written, all passing
Day 5: Deployed to staging
```
**Archive**: `archive/2026-01/epic-012-daily-log.md`
- Useful for retrospective (understand blockers, timeline)
- Not SoT (too specific to this EPIC)
**Alternative**: Delete if no retrospective planned
- Only keep if post-mortem will reference it
---
## Common Mistakes
### Mistake 1: Extracting Noise to SoT
**Bad**:
```markdown
# CFD-093: User Said They Like Blue
User mentioned they prefer blue color scheme.
```
**Why Bad**: One user opinion ≠ pattern. Noise, not insight.
**Fix**: Only extract when pattern (3+ users) or high-impact single user
---
### Mistake 2: Deleting Insights
**Bad**: Delete temp file with unique customer quote before extracting to CFD
**Fix**: Always review temp files for extractable insights BEFORE deletion
---
### Mistake 3: Over-Archiving
**Bad**: Archive everything "just in case"
**Why Bad**: Archive becomes junk drawer (hard to find useful content)
**Fix**: Be selective. Only archive if future retrospective value.
---
### Mistake 4: Contaminating SoT with Context
**Bad**:
```markdown
# BR-096: API Rate Limit
We chose 1000 req/min after discussing with team on Slack.
Sarah thought 500, but John argued for 1000 because...
[3 paragraphs of deliberation]
Final decision: 1000 req/min
```
**Fix**:
```markdown
# BR-096: API Rate Limit
**Decision**: 1,000 requests/min
**Rationale**: Balance between UX (responsive) and cost (infrastructure)
**Alternatives Considered**: 500 req/min (too restrictive for power users)
```
**Archive team discussion separately** (if useful for retrospective)
---
## Harvest Workflow
### Step 1: Inventory (List All Temp Files)
```bash
ls temp/
```
Output:
```
epic-012-notes.md
user-interview-3.md
api-draft.md
debug-log.txt
```
---
### Step 2: Categorize (Extract | Archive | Delete)
| File | Category | Reason |
|------|----------|--------|
| `epic-012-notes.md` | Archive | Timeline useful for retro |
| `user-interview-3.md` | Extract → Delete | Extract pain point to CFD, then delete |
| `api-draft.md` | Extract → Delete | Migrate to API-045, then delete |
| `debug-log.txt` | Delete | No value (debug session) |
---
### Step 3: Extract Insights to SoT
- Create CFD-089 from user-interview-3.md
- Update API-045 from api-draft.md
- Document in harvest log
---
### Step 4: Archive Contextual Files
- Move `epic-012-notes.md` to `archive/2026-01/`
- Update archive manifest
---
### Step 5: Delete Noise
- Delete `debug-log.txt`
- Delete `user-interview-3.md` (after extracting CFD-089)
- Delete `api-draft.md` (after updating API-045)
---
### Step 6: Validate
- [ ] temp/ directory empty (clean slate)
- [ ] All insights in SoT (nothing lost)
- [ ] Archive manifest updated
- [ ] Harvest log complete
---
## Archive Organization
### Directory Structure
```
archive/
2026-01/
epic-012-notes.md
ab-test-onboarding.md
team-discussion-pricing.md
2025-12/
epic-008-notes.md
user-research-summary.md
```
### Naming Convention
`YYYY-MM/epic-XXX-description.md` or `YYYY-MM/topic-description.md`
---
## Harvest Log Purpose
**Why Document Harvest**:
- Audit trail (what was deleted, why)
- Prevent accidental loss (reviewed before deletion)
- Team learning (patterns for next EPIC)
**Log Sections**:
1. Inventory (what files existed)
2. Categorization (extract/archive/delete decisions)
3. SoT updates (what was created/changed)
4. Archive manifest (what was preserved)
5. Deletion log (what was removed)
---
*Reference: Use these patterns when harvesting temp/ insights in `ghm-harvest` skill.*
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |