Ship

/ship

Optimize the delivery path from completed work to working install. When execution is cheap, delivery is the work.

Ship is the final step of the Execution phase. Code that isn't in users' hands isn't delivering value.

When to Use

Invoke /ship when execution is complete, tests pass, review is done, and you need to get changes to users. Also usable as a standalone diagnostic: run /ship against a stalled delivery pipeline to surface delivery-path tax without deploying anything.

Do not use when: You're still building. Ship is for completed work.

JIT References

Load [references/reality-contact.md](references/reality-contact.md) only when shipping reveals a different target environment, user behavior, ownership failure, or delivery-path tax than expected.

The Ship Process

Step 1: Identify the Delivery Path

Map the path from code to working install:

> "The delivery path for this change is: [local] → [PR/review] → [merge] → [CI/CD] → [staging/prod] → [user install]."

Be specific. Name each step and who/what owns it.

Step 2: Assess the Delivery-Path Tax

The delivery-path tax is friction that slows velocity. Identify where time is lost across: review queue, merge conflicts/stability, gate/scan time (serial vs parallel, flaky), approval bottlenecks, deploy duration, rollback readiness, and manual steps.

Step 3: Execute the Ship

With CI/CD Context

1. Reference the pipeline — point to the specific build/deploy configuration
2. Trigger the appropriate workflow — label PRs, push to deploy branches, etc.
3. Monitor the pipeline — watch for failures, report status
4. Verify deployment — confirm the change is live

Without CI/CD Context

1. List deployment steps — what needs to happen in order
2. Execute each step — or provide commands to execute
3. Verify each stage — confirm success before proceeding
4. Report final status — confirm user-facing change is live

Step 4: Post-Ship Verification

Shipping isn't done until you verify:

• Is it running? — deployed and executing
• Is it working? — intended behavior is present
• Is it reachable? — users can access/use it
• Were the promised characteristics delivered? — shipped result matches the declared success criteria from /execute
• Did reality match the frame? — users/operators encountered the problem we thought we solved; hidden constraints or ownership failures are routed to the right learning artifact

Create ad-hoc verification tests if needed — the point is confidence, not permanent infrastructure. Verification ties back to what the work claimed it would deliver, not only whether deployment completed.

Ship remains delivery-focused: route failed verification or rollback learning to /salvage; route broader reality-contact learning to /aim, /problem-space, guardrails, or metis. Do not reopen shipped work unless verification fails.

Output Format

## Ship

### Ship Report

**Shipped:** [what was shipped]
**Target:** [where it was shipped to]
**Path:** [the delivery path taken]

### Delivery-Path Tax Identified
- [Friction point 1]: [time/effort cost]
- [Friction point 2]: [time/effort cost]

### Ship Actions Taken
1. [Action 1 - result]
2. [Action 2 - result]

### Verification
- [x] Deployed successfully
- [x] Running in target environment
- [x] User-facing functionality confirmed

### Delivered Characteristics Validation
- [x] [Declared criterion/characteristic 1]
- [ ] [Declared criterion/characteristic 2, if not yet verified]

### Reality Contact Check
- Frame match: [matched | partially matched | missed]
- Different problem encountered: [none | describe]
- Hidden constraints / ownership failures: [none | describe]
- Learning routed to: [`/salvage` | `/aim` | `/problem-space` | guardrail | metis | none]

### Recommendations
[Any suggestions for reducing delivery-path tax in future]

Session Handoff

If a session file is in use, replace or append ## Ship. Preserve delivery-path tax, verification status, delivered-characteristic validation, reality-contact result, and learning route.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: open-horizon-labs
• Source: open-horizon-labs/skills
• License: MIT

Install and usage instructions live in the source repository linked above.