/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:

1. To MEMORY.md — Add to relevant section as a permanent rule
2. To CLAUDE.md — If it's a formatting, voice, or workflow pattern
3. To client CLAUDE.md — If client-specific
4. Mark as promoted = 1 in the learnings table

---