/experiment — Track hypotheses, variants, and learnings
Stop re-testing what already failed. Stop forgetting what worked. This skill gives you a structured experiment tracking system that compounds knowledge across sessions — every test informs the next one.
For full schema, modes (new/log/review/suggest), and messaging-mode scoring → the premium reference.
Claude Code Triggers
Invoke this skill when user says:
- "/experiment new [description]"
- "/experiment log [experiment-id or description]"
- "/experiment review [domain]"
- "/experiment suggest [domain]"
- "New experiment:" or "I want to test..."
- "Log experiment result:" or "The results are in..."
- "What experiments have we run on...?"
- "What should I test next for...?"
Do NOT invoke when:
- User is doing A/B testing in code (development work)
- User wants to track bugs (use
/track-bug)
- User wants general research (use context skills)
Modes (quick reference)
| Mode |
Purpose |
Required arg |
/experiment new [description] |
Create experiment + variants |
description (optional) |
/experiment log [id or description] |
Record results + learnings |
id or description |
/experiment review [domain] |
Browse past experiments |
domain (optional) |
/experiment suggest [domain] |
Recommend next test |
domain |
For full mode workflows + output formats → the premium reference.
Domains
| Domain |
Use for |
content |
LinkedIn hooks, post formats, content angles, newsletter subjects |
messaging |
Value props, positioning variants, CTA wording, taglines |
workflow |
Skill parameters, MCP configurations, prompt engineering |
outreach |
Email subject lines, sequences, cold outreach approaches |
pricing |
Packaging variants, pricing page layouts, discount strategies |
Database
Location: .claude/experiments/experiments.db
Tables: experiments, variants, learnings, experiments_fts (FTS5 index)
For full SQL schema + init script → the premium reference.
Integration with Other Skills
/session-wrap — Auto-detects experiments discussed in a session, offers to log results/linkedin-content — Before generating new posts, checks for content experiments and applies learnings/outreach-emails — Before generating sequences, checks for outreach experiments/product-messaging — Before generating messaging, checks for messaging experiments/recall — Promoted experiment learnings are searchable via the recall index
Learning promotion
When a learning proves durable (validated by 2+ experiments), promote it:
- To MEMORY.md — Add to relevant section as a permanent rule
- To CLAUDE.md — If it's a formatting, voice, or workflow pattern
- To client CLAUDE.md — If client-specific
- Mark as
promoted = 1 in the learnings table
1---2name: experiment3description: Experiment4---56# /experiment — Track hypotheses, variants, and learnings78Stop re-testing what already failed. Stop forgetting what worked. This skill gives you a structured experiment tracking system that compounds knowledge across sessions — every test informs the next one.910For full schema, modes (new/log/review/suggest), and messaging-mode scoring → the premium reference.1112---1314## Claude Code Triggers1516**Invoke this skill when user says:**17- "/experiment new [description]"18- "/experiment log [experiment-id or description]"19- "/experiment review [domain]"20- "/experiment suggest [domain]"21- "New experiment:" or "I want to test..."22- "Log experiment result:" or "The results are in..."23- "What experiments have we run on...?"24- "What should I test next for...?"2526**Do NOT invoke when:**27- User is doing A/B testing in code (development work)28- User wants to track bugs (use `/track-bug`)29- User wants general research (use context skills)3031---3233## Modes (quick reference)3435| Mode | Purpose | Required arg |36|------|---------|--------------|37| `/experiment new [description]` | Create experiment + variants | description (optional) |38| `/experiment log [id or description]` | Record results + learnings | id or description |39| `/experiment review [domain]` | Browse past experiments | domain (optional) |40| `/experiment suggest [domain]` | Recommend next test | domain |4142For full mode workflows + output formats → the premium reference.4344---4546## Domains4748| Domain | Use for |49|--------|---------|50| `content` | LinkedIn hooks, post formats, content angles, newsletter subjects |51| `messaging` | Value props, positioning variants, CTA wording, taglines |52| `workflow` | Skill parameters, MCP configurations, prompt engineering |53| `outreach` | Email subject lines, sequences, cold outreach approaches |54| `pricing` | Packaging variants, pricing page layouts, discount strategies |5556---5758## Database5960**Location:** `.claude/experiments/experiments.db`61**Tables:** `experiments`, `variants`, `learnings`, `experiments_fts` (FTS5 index)6263For full SQL schema + init script → the premium reference.6465---6667## Integration with Other Skills6869- **`/session-wrap`** — Auto-detects experiments discussed in a session, offers to log results70- **`/linkedin-content`** — Before generating new posts, checks for content experiments and applies learnings71- **`/outreach-emails`** — Before generating sequences, checks for outreach experiments72- **`/product-messaging`** — Before generating messaging, checks for messaging experiments73- **`/recall`** — Promoted experiment learnings are searchable via the recall index7475---7677## Learning promotion7879When a learning proves durable (validated by 2+ experiments), promote it:80811. **To MEMORY.md** — Add to relevant section as a permanent rule822. **To CLAUDE.md** — If it's a formatting, voice, or workflow pattern833. **To client CLAUDE.md** — If client-specific844. Mark as `promoted = 1` in the learnings table8586---87
matteotitta