agentleFS
shanselmanshanselman/openclaw-windows-hub/.agents/skills/windows-pr-prepare/SKILL.md

windows-pr-prepare

Skill2.2k starsChanged 12 days agoMITView on GitHub

Use before opening a Windows node PR, writing its description or updating its body. Prepare a local draft with validation and review evidence; publish only with explicit authority.

1 other project uses a version of this; 1 new project in the last 4 weeks, against 0 the 4 before.

Adopted by

Surging: 1 new project took it in the last 4 weeks, none the 4 weeks before.thin data

Week of Sep 20: 0 new projectsWeek of Sep 27: 0 new projectsWeek of Oct 4: 0 new projectsWeek of Oct 11: 0 new projectsWeek of Oct 18: 0 new projectsWeek of Oct 25: 0 new projectsWeek of Nov 1: 1 new projectWeek of Nov 8: 0 new projects
Sep 2026Oct 2026Nov 2026
New projects taking a version of this file each week, over the 8 weeks our latest run read: only 1 in all, so this is a hint, not a trend.

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. About 0 are common, 13 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use before opening a PR, writing its description or updating its body.

    line 8
  2. Distinctive:

    Read repository instructions, the affected production owner and existing tests.

    line 15
  3. Distinctive:

    Do not fetch, rebase or switch branches merely to make evidence identities agree.

    line 30
  4. Distinctive:

    Use the validation planner with all candidate changes, then the proof skill and proof pools for

    line 34
  5. Very specific:

    Read outputs from the exact invocation, not the newest file or exit zero alone.

    line 57
  6. Very specific:

    Check executed cases, pass/fail/skip counts and failures.

    line 58
  7. Distinctive:

    After edits, commits or rebase, reassess which results and reviews still apply.

    line 62
  8. Distinctive:

    Before publishing non-trivial UI, MCP, node-command, setup, pairing, security, permissions or diagnostics changes, obtain an independent …

    line 87
  9. Distinctive:

    Verify findings before fixing, then refresh affected checks and review coverage.

    line 96
  10. Distinctive:

    Read the current Windows PR template.

    line 101
  11. Distinctive:

    Follow its title convention.

    line 104
  12. Distinctive:

    Keep risks and blockers visible.

    line 111
  13. Distinctive:

    Put exact commands and actual outcomes/counts in Validation.

    line 112
  14. Distinctive:

    Keep supporting history in links or optional details, not an operational diary.

    line 119
  15. Very specific:

    Stop before publication without explicit authority.

    line 134
  16. Distinctive:

    Verify published content and links without treating publication as proof.

    line 139

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Windows PR preparation

Use before opening a PR, writing its description or updating its body. Follow this flow for the selected change, not a repository-wide audit. The preparation output is a local body and a ready-or-blocked handoff. Publication and merge require their own authorization.

1. Establish the claim, candidate and authority

Read repository instructions, the affected production owner and existing tests. State the concrete behavior being changed and the invariant to preserve. For instructions, identify the reader's task and authoritative guidance instead of inventing a product behavior claim.

Record the repository, task worktree, intended target, observed target tip, comparison base, HEAD and relevant staged, unstaged and untracked content. For uncommitted work, retain the test or proof runner's record of the source it used. If there is no such record, capture the diff and relevant file hashes when running the check and label that association as manually recorded. Do not infer an old run's source from today's files or require a commit just to obtain a SHA. Missing comparison context stays explicit.

Separate permission to inspect, implement and validate from permission to install, change shared runtimes, commit, push, upload, edit GitHub metadata or merge. Do not fetch, rebase or switch branches merely to make evidence identities agree.

2. Select the smallest sufficient validation

Use the validation planner with all candidate changes, then the proof skill and proof pools for applicable obligations. A planned lane or declared pool is not an executed result.

Ordinary prose uses documentation and skill checks, not full-product builds or native runtime proof. Code retains mandatory local closeout plus focused checks; the shipped winnode reference retains its command contract checks.

Select a runtime only when the claim needs one and authority permits it. A separate app profile does not authorize installing or registering a Gateway or changing a shared runtime. For MCP, follow the local MCP guide, including fresh-process profile/token/URL binding.

3. Check what actually ran

Apply the existing result rules. For inventory-bound evidence, use the validator and limits.

  • Read outputs from the exact invocation, not the newest file or exit zero alone. Check executed cases, pass/fail/skip counts and failures. Label manual observations as manual; do not fabricate receipts.
  • Distinguish the PR head, tested commit (including CI merge commits), uncommitted source used by the run, and loaded binary/runtime.
  • After edits, commits or rebase, reassess which results and reviews still apply. Source equality can establish correspondence, but cannot turn an earlier review or runtime execution into a new run.
  • Retain original failures. An instrumented pass does not erase them, and restoring source does not restore previously built binaries.
  • Ordinary allowed skips are not complete custom/native proof. Identify the CI run and attempt when citing results; do not present older results as newly executed tests. Successful dispatch is not completed testing or review approval.

Missing required review, UI or Gateway evidence stays Not verified / blocked. For an unexpected failure, preserve the attempt and identify the next bounded investigation or decision. Fix task-caused defects within scope; do not automatically repair unrelated tests, rerun until green or create new PR layers. A failure in another task blocks this task only if this task depends on that result. Otherwise continue the authorized work without borrowing its evidence.

4. Review the intended scope

Account for both source review and the required independent rubber-duck review:

  • For source review, when requested or required, use the bounded local-review recipe. Select mode/base explicitly, include committed and dirty work when intended, and retain Windows P0-P3 coverage. Follow its context, final-output and fallback rules.
  • Before publishing non-trivial UI, MCP, node-command, setup, pairing, security, permissions or diagnostics changes, obtain an independent rubber-duck review of the implementation and proof plan. Follow the proof skill's review route for available roles and explicit blockers. A source-review verdict alone does not establish that this review was performed.

Report actual scope, findings, dispositions and missing review. Self-checks are not independent review, and an unavailable required review remains blocked. Verify findings before fixing, then refresh affected checks and review coverage. Reviews are advisory, not runtime proof or authority to publish.

5. Write the local body from the actual template

Read the current Windows PR template. Copy it to a local session artifact; preserve headings, HTML comments, checklist items and ordering, including security, compatibility and review conversations. Follow its title convention. Check only supported claims, not aspirational work.

The proof package defines what counts as evidence. This step assembles that evidence into the current template:

  • Lead with the current behavior, strongest evidence and meaningful limits. Keep risks and blockers visible.
  • Put exact commands and actual outcomes/counts in Validation. Identify the tested source in Real Behavior Proof, including uncommitted work.
  • For prose-only work, label source/link checks as static assessment and runtime proof as not applicable.
  • Declare every applicable pool, or use - `none`: <reason> when no custom pool is required.

Keep supporting history in links or optional details, not an operational diary. Compare the completed body with the template; do not drop unanswered fields or claim pending checks passed. Apply the evidence/redaction boundaries and verify referenced artifacts. Private local paths are not public proof links.

6. Hand back readiness and stop

Report the exact candidate, local draft location, strongest evidence and remaining limits. Say ready for human review only when applicable gates are satisfied. Missing required source or rubber-duck review means blocked, even when tests pass. Name any other blocked, waiting for required CI, or candidate changed state and the next bounded decision. A complete local draft can still describe blocked work; it is not a ready-to-merge claim.

Stop before publication without explicit authority. If separately authorized, verify the intended repository/base/head and live PR state, then use the host's supported publication tools and their guard/failure contracts. Preserve concurrent human edits; after an uncertain write, inspect the result before retrying. Verify published content and links without treating publication as proof.

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.