agentleFS
seankoji-comseankoji-com/.agents/skills/raise-pr/SKILL.md

raise-pr

Skill2 starsChanged 10 days agoMITView on GitHub

Use whenever creating or updating a pull request, including during implementation, review-loop, ship and PR maintenance.

  • Commits and pushes

7 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. 0 are common, 9 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 9

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the repository-designated template.

    line 24
  2. Distinctive:

    Keep the template's headings, prompts and required sections.

    line 28
  3. Distinctive:

    Use captions identifying before/after, flow and size.

    line 50
  4. Distinctive:

    If API tools lack uploads, use an authorized CLI attachment path or disclose upload unavailability.

    line 56
  5. Distinctive:

    When capture, upload or access is unavailable, a normal PR is allowed.

    line 61
  6. Distinctive:

    Before creating a PR, obtain a completed independent code review covering the proposed diff.

    line 74
  7. Distinctive:

    Treat changes to tests, CI workflows and configuration as material when they affect verification or runtime semantics, even if product code …

    line 95
  8. Distinctive:

    After a rebase, apply review reuse to merge-base checks, patch comparisons and changes in the base's assumptions.

    line 99
  9. Distinctive:

    Before pushing material changes with review pending, check existing auto-merge and merge-queue enrollment.

    line 105

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Raise a pull request

Apply these requirements to new PRs and updates to existing PRs. Follow the repository's delivery policy; a repository that requires direct publication does not get a PR. This skill adds no authority to commit, push, upload private artifacts, create PRs, merge or deploy.

Apply GitHub writing to descriptions and comments, including drafts and later edits.

Prepare the description

Inspect the actual repository, intended base, current head and scoped diff. Apply delivery preflight for effective rules, check producers, stacked-base validation, carried dependencies and credential-shaped examples before publication. Keep its unresolved gates in the PR and handoff. Find applicable PR templates and contribution instructions, including templates under .github, docs or the repository root and any organization defaults. Use the repository-designated template. If several apply and the choice is unclear, ask which one to use. If template discovery is unavailable, disclose that limitation rather than claim there is no template.

Keep the template's headings, prompts and required sections. Put the change summary, verification, screenshots and agent attribution in the matching sections; append missing sections when needed. Mark checklist items only when verified. Without a template, use a concise summary, verification, visual evidence and agent attribution. For an existing PR, retain relevant human text and evidence, updating stale claims rather than replacing the whole description blindly.

Include visual evidence

For changes affecting visible UI behavior, capture and inspect the affected flows at one mobile and one desktop size using supported browser or device tools. Record actual viewport dimensions or device, state, source revision and the tested build when it differs. Select the sizes appropriate to the application's supported layouts. For a platform that cannot run at one size, explain why.

Include after screenshots of the changed behavior at both sizes. Attempt matching before screenshots when practical, using the intended base or a verified baseline build. Keep state, viewport and test data comparable. Capture before editing when the task allows; otherwise use a safe isolated baseline without disturbing dirty work. Label the revision actually captured. New UI with no previous state can explain why a before image is not applicable.

Use captions identifying before/after, flow and size. Embed screenshots with PR-accessible image URLs through an authorized supported artifact mechanism and check that the images render for the intended reviewers. Local filesystem paths do not make images visible in a GitHub PR. If the installed CLI supports it, gh pr create --attach and gh pr edit --attach upload images and rewrite body references to the uploaded assets; check local help before relying on that flag. If API tools lack uploads, use an authorized CLI attachment path or disclose upload unavailability. Never move non-public captures to public third-party hosting as a workaround. Do not commit images to source solely to host PR evidence. Keep secrets and personal data out of captures and uploads.

When capture, upload or access is unavailable, a normal PR is allowed. Include the available screenshots, explicitly list each missing size/state or inaccessible image, explain why it is missing, and state the remaining verification. Explain missing before screenshots too. Never substitute mockups or old screenshots for observed results, or call incomplete visual evidence complete. Other repository release gates still apply.

For changes with no visible UI effect, explicitly state that screenshots are not applicable and explain why. Include useful checks or other evidence; screenshots of code or terminals are not a substitute for UI evidence.

Require independent review and disclose agents

Before creating a PR, obtain a completed independent code review covering the proposed diff. Reuse an existing review only when it covers the current behavior. Apply review for cross-lineage eligibility, reviewer selection, dispatch, evidence and findings. Self-review and a separate same-family agent do not qualify. Give the reviewer the diff, relevant context and acceptance criteria and require it to read and apply review in the assigned-reviewer role. Bound a standalone review/repair cycle to three rounds; report unresolved gates when exhausted. Honour explicit model choices. If an eligible reviewer is unavailable or delegation is prohibited, finish independent preparation and report the blocker; do not create the PR, including a draft. An explicit user waiver, in the original request or later, can override this skill's creation or renewed-review requirement for the stated scope. Disclose the missing review, waiver and remaining risk in the PR. A draft request alone is not a review waiver. Waivers do not bypass repository protections.

Address actionable findings within authorization and obtain review of material code changes made afterward. For existing PRs, material changes may be pushed for hosted checks, previews and independent review. Disclose pending review and require it before merge. Existing eligible agent reviewers can review after the push; this does not require the repairing agent to dispatch a reviewer. If no eligible reviewer is available, report the merge gate and who must supply review. Treat changes to tests, CI workflows and configuration as material when they affect verification or runtime semantics, even if product code is unchanged. Description-only corrections and changes demonstrated to preserve the reviewed semantics may reuse the review; explain reuse and rerun affected checks. After a rebase, apply review reuse to merge-base checks, patch comparisons and changes in the base's assumptions. A human review may satisfy repository approval rules; it does not replace this skill's required cross-lineage model review. Agent review cannot replace a required human approval, and no review replaces required CI.

Before pushing material changes with review pending, check existing auto-merge and merge-queue enrollment. Within authorization, hold automatic merging before the push so stale approvals cannot merge unreviewed repairs. If the hold cannot be verified or is outside authority, stop before pushing and report that blocker. Restore automatic merging only after current-diff review and other merge gates pass, or after an applicable explicit review waiver and all remaining gates.

Disclose every agent that materially wrote or reviewed the change, including the coordinating agent when it contributed code. For each, record its role, provider/model when known, contribution or review scope, and review outcome with remaining findings or limitations. Identify human contributions as human. State unknown identities explicitly rather than guessing. Record the reviewed revision or a precise identification of the reviewed dirty diff. A failed task, timeout or empty response is not a completed review.

Create or update and verify

Within existing authorization, use the host's GitHub API tools or authenticated CLI. Never use an interactive browser form for PR creation. Preserve the intended base, branch and draft status. Verify the returned PR URL and read back its description, template sections, evidence and attribution. Read the PR's closingIssuesReferences and confirm they match the issues it is meant to close; fix unintended references in the body and read them again. After a transport error, check whether the PR already exists before retrying. Partial attachment failure can create or update the PR despite a nonzero CLI exit; read back the returned URL and disclose missing images rather than creating a duplicate PR. Follow host PR linking rules.

On later updates, refresh affected screenshots, review coverage, agent attribution and checks to match the current change. Label retained historical evidence with its revision. Do not silently carry old evidence forward as current proof. Report the PR URL, head, checks, independent review and any visual evidence gaps. PR creation does not prove hosted CI, merge, deployment or live acceptance.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.