Index of every page on this site: llms.txt
All skills
e2e-inc avatar

/ship-pr

@a5d1788
by e2ee2e-inc/e2e14k stars
782
Run it yourself

Take finished work in e2e-inc/e2e to a PR a human can review without fighting CI or bots - checks, verification against the built packages, a fresh-context self-review, the PR itself, then babysitting to the Ready for Human Review label. Use whenever you are asked to open, create, ship, or submit a PR, or when implementation is done and the next step is review.

Use this Skill: https://skilld.dev/gh/e2e-inc/e2e/ship-pr

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈93 tokens for metadata: the name and description. ≈1.1k when used: this file. ≈497 more on demand in 1 file.

Description uses 4.7% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈373 characters, or 4.7%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Ship a PR

Opening a PR is not the finish line. The finish line is a PR labeled Ready for Human Review: verified against the built packages, reviewed by a fresh pair of eyes, green, and every bot thread handled. A human should only ever spend time on judgment.

Run every step. If one cannot run, say which and why in the handoff instead of skipping it quietly.

1. Checks

Important: first run the shortening pass from authoring-docs on every guide page the change touches (docs/**/*.mdx outside docs/reference/), even for a one-paragraph edit. A reference page the change touches follows authoring-docs, Reference pages instead: ## Reference first, minimal summary tables, every fact kept. The checks below then catch an anchor either pass broke.

Run what CI runs for what you touched, and fix everything:

pnpm check                    # lint, dead code, typecheck, error codes, peer ranges, docs
pnpm test                     # or the touched package: pnpm --filter e2e run test:unit
pnpm --filter @e2e-dev/testbed test
pnpm test:web-benchmark       # when the runner, web engine, or benchmark changed

Then the repo rules pnpm check cannot see (AGENTS.md, "Contracts"):

  • A user-visible change to a published package adds a .changeset/ entry, bumped patch on 0.x even for a feature. minor only when the change is breaking (summary opens **Breaking:**) or the maintainer calls a big release (**Release:**).
  • Behavior changes update the matching docs/**/*.mdx page in the same change, "not implemented yet" callouts included, and skills/e2e/ when it describes the changed surface.
  • A public API change reviews the emitted .d.ts and updates packages/e2e/tests/types/sdk-types.ts; a wire change edits the schema, both fixtures, and the producer together.
  • A new build input or benchmark dependency goes into .github/filters.yml.
  • A new integration, CI service, or migration guide gets the vendor's logo on every card that links to it, never a Font Awesome placeholder, and a card on its section's overview (authoring-docs, Vendor logos). A package PR needs this as much as a docs PR.

2. Verify

Prove the change with the verify skill on the surface it touches: the built CLI on the testbed or a benchmark, the MCP server, a scratch project for init, or the docs site. Bug fixes capture main against the branch. Keep the output and any video for the PR.

3. Self-review in a fresh context

The agent that wrote the change does not get to judge it. Spawn a reviewer with no memory of this conversation (a subagent, or a new session) and give it references/review-prompt.md with the base branch filled in. Fix every critical and important finding you agree with, then rerun steps 1 and 2 for what the fixes touched. For each finding you reject, keep the one-line reason for the PR body or handoff.

No way to spawn a reviewer: re-read the whole diff cold against the same prompt yourself, and say so in the handoff.

4. Commit and open

  • Branch off an up-to-date origin/main. Commits follow Conventional Commits.
  • Title and body with the writing-pr skill. The body carries the verification evidence from step 2.
  • Open it ready, not draft: gh pr create --base main --title ... --body-file <file> --attach <each screenshot and video> (gh 2.99+). The body must carry the ## Verified section writing-pr describes; fix a thin one with gh pr edit --body-file <file> --attach ....
  • Multi-PR change: open every layer, each based on the one below, before babysitting.

5. Babysit

Hand the PR to the babysit skill and run it to the end: green, threads handled, Ready for Human Review label. For a stack, babysit bottom-up; a layer is labeled only when it is ready on its own.

Handoff

Reply to the human with:

  • the PR URL and its label state
  • how it was verified (surface, what you ran, what you saw, the media)
  • what you fixed from review, what you dismissed and why
  • anything left for them: escalated threads, parts you could not verify, scope you declined

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    This skill implements a comprehensive workflow for shipping pull requests in the tester-army/e2e repository, covering testing, verification, and code review. It uses standard development tools like pnpm, git, and the GitHub CLI. A low-severity risk of indirect prompt injection exists because the skill processes arbitrary code changes, but this is addressed by instructing a fresh-context review session.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: LOW · No issues

Signed by skilld at a5d1788. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 weeks ago
metadata
{
  "internal": true
}

README badge

README badge for e2e-inc/e2e/ship-pr