# Ship Learnings

> Ship-learnings — post-ship compound-learnings capture

- Skill: `matteotitta/ship-learnings` (Agent Skill)
- Install (CLI): `npx skillmds@latest add matteotitta/ship-learnings`
- Raw SKILL.md: https://api.skillmd.com/api/skills/matteotitta/ship-learnings/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: matteotitta (https://skillmd.com/u/matteotitta)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/matteotitta/ship-learnings

---


# Ship-learnings — post-ship compound-learnings capture

Capture what we learned from a completed ship so the next strategy refresh has accumulated evidence. Adapted from `/ce-compound` in [EveryInc/compound-engineering-plugin](https://github.com/EveryInc/compound-engineering-plugin) v3.5.0 (MIT).

Per Moretti's framing (K1): **everything that ships is an experiment.** Each ship tests a hypothesis from the strategy doc. Capture the result.

## When to run

Invoke when the user says:
- "We just shipped [feature] — what did we learn?"
- "Run ship-learnings on [release]"
- "Post-ship review for [experiment]"
- "Compound learnings"

Do NOT invoke when:
- The session is the unit (use `/session-wrap` for session-scoped wrap)
- The ship is broken / incident (use `engineering:incident-response`)
- No ship has occurred — this is post-ship only

## Distinct from /session-wrap

| Skill | Scope | When |
|---|---|---|
| `/session-wrap` | One Claude Code session | End of session |
| **`/ship-learnings`** | One ship / release / experiment | After feature ships, regardless of how many sessions it took |

A ship may span 5 sessions; one ship-learnings record captures all of them.

## Inputs

**Required:**
- Locked `strategy-doc` (the hypothesis being tested by this ship)
- Ship description: name, date shipped, scope summary

**Recommended:**
- Latest `product-pulse` (the metric movement post-ship)
- User feedback / quotes / support tickets from the ship period
- Original strategy track this ship belongs to

## Steps

1. **Phase 1 — Load context.** Read locked `strategy-doc` + latest `product-pulse`. Identify which strategy track this ship belonged to.
2. **Phase 2 — Pull ship signals.** Metric deltas (from pulse), user quotes, support volume changes, anomalies.
3. **Phase 3 — Run interview.** Walk the user through the 7-section structure. The hardest section is **§3 Result** — push back if the user says "kind of worked" without evidence.
4. **Phase 4 — Compose record.** Apply length discipline: ≤ 600 words total.
5. **Phase 5 — Self-roast.** Run checks below.
6. **Phase 6 — Push.** Save to ship-learnings folder + flag in next strategy refresh review.

## Self-roast (run before push)

- [ ] Hypothesis is quoted directly from strategy-doc (not paraphrased)
- [ ] §3 Result has explicit verdict (validated / partial / invalidated) — not vague
- [ ] §3 Result cites specific evidence (metric delta from pulse, user quote, ticket count) — not opinion
- [ ] §4 What worked and §5 What didn't are concrete (specific decisions / patterns) — not generic ("communication was good")
- [ ] §6 Action items have owners
- [ ] §7 Follow-on experiments are testable hypotheses, not vague ideas
- [ ] Total word count ≤ 600 (compounding-learnings should be easy to skim across many ships)
- [ ] Framing K1 applied: "shipping is an experiment" — verdict is data, not judgment

# Ship-learnings: {feature name} ({YYYY-MM-DD})

**Strategy track:** {track name from strategy-doc} · **Pulse ref:** {latest pulse path}

## Ship summary
{1-2 lines}

## Hypothesis (from strategy-doc)
> {direct quote from strategy-doc}

## Result: {validated | partially validated | invalidated}
{paragraph with evidence}

## What worked
-...

## What didn't work
-...

## What to do differently next time
- **{action}** — owner: {name}

## Follow-on experiments
-...
```

## Composition rule reference

Ship-learnings is the feedback node in the **PM closed loop** (P3). Strategy → pulse → ship → ship-learnings → strategy refresh. See [.claude/rules/pm-loop.md](../../../../../rules/pm-loop.md).

## Attribution

Adapted from [EveryInc/compound-engineering-plugin](https://github.com/EveryInc/compound-engineering-plugin) v3.5.0 (MIT). Source pattern: `/ce-compound`. Framing basis: K1 "everything that ships is an experiment" from Marcus Moretti's AI PM Guide.