Skip to content
Back to skills

Ship It

ASecurity

Ship a change or coordinate an approved complex plan through risk-selected gates, PR feedback and merge. Accepts a task description, GitHub issue or Jira ticket. Use for end-to-end shipping.

  • 5 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 4, 2026
toolsbashgit

Works with

  • claude code
  • cli

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned October 7, 2026

npx -y skills add smykla-skalski/sai --skill ship-it --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ship It?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ship It
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/smykla-skalski-ship-it/badge)](https://www.skillsdirectory.com/skills/smykla-skalski-ship-it)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ship-it
description: Ship a change or coordinate an approved complex plan through risk-selected gates, PR feedback and merge. Accepts a task description, GitHub issue or Jira ticket. Use for end-to-end shipping.
license: MIT
compatibility: Works in Claude Code, Codex, opencode and Copilot CLI. Needs git and authenticated gh. Uses adversarial-review and adversarial-test when installed.
argument-hint: "[--issue] [--risk low|medium|high] <task description | github-issue-url | jira-url>"
allowed-tools: Agent Bash Edit Glob Grep Read Skill ToolSearch Write
user-invocable: true
metadata:
  short-description: Ship a change, issue, or Jira ticket to merge
---

# Ship It

Take a change to a merged PR, or coordinate independently shippable issues from an approved complex plan.

**Mode:** autonomous. Ask only when ambiguity cannot be resolved from the repository. Keep polling selected hosted gates until merged or a hard stop.

**Other agents:** Read [references/fallbacks.md](references/fallbacks.md) when not in Claude Code or an agent feature is missing.

Invocation: `/ship-it [--issue] [--risk low|medium|high] <task description | github-issue-url | jira-url>` (`$ship-it` in Codex). Runs only when invoked by name or asked to ship a change. Claude Code appends the arguments; if none are appended, take them from the user's request.

## Workflow contract

- Run phases below in order; do not skip a gate because a harness lacks a preferred tool.
- Before each phase, read [references/capabilities.md](references/capabilities.md) and [references/capabilities.json](references/capabilities.json), select its profile and satisfy the machine-readable preflight before any side effect.
- Read [references/roles.md](references/roles.md) and [references/roles.json](references/roles.json) before dispatch.
- Load each phase reference immediately before that phase, not during initial skill discovery.
- After resolving, maintain [references/checkpoint.md](references/checkpoint.md) before repository changes, GitHub [references/claims.md](references/claims.md) before branch or source changes, and [references/telemetry.md](references/telemetry.md) through completion.
- Read [references/evidence.md](references/evidence.md) when the first task revision is committed. Evidence due at each PR or merge gate must pass for the exact current revision.
- Read [references/risk.md](references/risk.md) after exploration. Select and report the revision's risk, policy source and required gates before validation.
- Read [references/release.md](references/release.md) after exploration. Resolve repository, GitHub and default release policy into the checkpoint before validation.
- Every source change invalidates completion evidence from the previous revision.
- Never bypass hooks, suppress checks, force-push after the first push, or force-merge.
- In Sail mode, missing worker or gate subagents pause the run; never replace them with inline work.

## Phase reference index

| Phase | Read immediately before starting |
| :-- | :-- |
| 1 — Resolve | [references/inputs.md](references/inputs.md) |
| 2 — Explore | [references/explore.md](references/explore.md) |
| 3 — Branch | [references/branch.md](references/branch.md) |
| 4 — Implement | [references/implementation.md](references/implementation.md) |
| 5 — Review | [references/review.md](references/review.md) |
| 6 — Test | [references/test.md](references/test.md) |
| 7–10 — PR loop | [references/pr-loop.md](references/pr-loop.md) |
| 11 — Complete | [references/completion.md](references/completion.md) |

## Phase 1 — Resolve the task

Classify the input, resolve its source and acceptance criteria, and create or resume its durable checkpoint outside the repository. No branch, edit or commit until resolution and checkpoint reconciliation succeed.

If the input is an approved complex plan or an umbrella issue with subissues, read [references/orchestration.md](references/orchestration.md) and follow its parent coordinator workflow. The parent never implements a child issue. An ordinary implementation issue, including a worker's assigned issue, follows the single-change phases below. In Sail mode, absent worker or review/test gate subagents pause the run; never use inline gate fallbacks.

## Phase 2 — Explore

Discover repository instructions, affected code and required quality gates. Resolve and checkpoint release policy before editing.

## Phase 3 — Branch

Start from the current default branch in an isolated conventional branch or assigned Sail worktree.

## Phase 4 — Implement

Implement the smallest complete change, add behavior tests, run relevant gates and create signed conventional commits.

## Phase 5 — Review

Run the selected review gates for the current committed revision.

## Phase 6 — Test

Run the selected test gates for the review-clean committed revision.

## Phases 7–10 — PR, wait, fix, merge

Push and open the PR, wait for selected hosted gates, resolve every required thread, revalidate changed revisions, then merge through the repository's documented convention.

## Phase 11 — Close, report, clean up

Verify delivery, close only the GitHub issue, report evidence and clean up when safe.

## Hard stops

Stop and name the exact next human action when: a phase capability preflight fails; input is empty, unrecognized or unreachable; the GitHub issue is closed or its claim conflicts; the Jira ticket is finished; branch protection needs approvals or admin action; a hosted requirement remains unsatisfied after ~30 min; a required review thread loops more than 3 times; a test requires disabling a check; a selected test gate returns BLOCKED; the review/test round cap is hit and the user can be asked; or the task needs a product/design decision the repository cannot answer.

Files in this skill

  • SKILL.md5.8 KB
  • agents/openai.yaml336 B
  • references/fallbacks.md4.2 KB
  • references/inputs.md5.7 KB
  • references/pr-loop.md4.5 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…