New: 97,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

PR Create

by vfarcic in vfarcic/dot-agent-deck

Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off.

MITAuto-check passedDevelopment

Install PR Create

skills CLI
$ npx skills add vfarcic/dot-agent-deck --skill pr-create -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install vfarcic/dot-agent-deck pr-create --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/vfarcic/dot-agent-deck.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/pr-create .claude/skills/pr-create && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
pr-create
GitHub stars
109
Token cost
~3k tokens
SKILL.md length
1,771 words
Files
1
Skills in repo
23
Repo updated
First seen
Licence
MIT

At a glance

Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off.

  • Works in 5 steps: Before the PR → Open it → Settle CI and the automated review → …
  • Work on a branch is finished and needs to become a reviewed PR
  • SKILL.md covers 1. Before the PR, 2. Open it, 3. Settle CI and the automated… and 4. Answer and resolve every…, plus 2 more sections
  • Calls gh and cargo

What it does

PR Create is an agent skill from vfarcic/dot-agent-deck. Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off. Never merges; arms auto-merge only when nothing downstream gates the merge. Use when work on a branch is finished and needs to become a reviewed PR, whether or not a PRD started it.

Its SKILL.md is about 3k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Development, covering Pull requests and PRD writing. The repository describes itself as: A rich terminal dashboard for monitoring and controlling multiple AI coding agent sessions. The licence is MIT.

When your agent uses it

  • Work on a branch is finished and needs to become a reviewed PR
  • Not a PRD started it

Example prompts

  • “/pr-create”

Workflow steps

5 steps, taken from the step headings in SKILL.md.

  1. Before the PR
  2. Open it
  3. Settle CI and the automated review
  4. Answer and resolve every finding
  5. Hand off

What it can do on your machine

Read from SKILL.md and the folder at commit c5d24e7. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • gh
    • cargo

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use gh, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

PR Create loads about 3k tokens when it runs. Until then it costs about 89 tokens; SKILL.md has 1,771 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~89
When it runs · the whole SKILL.md, loaded when a task matches
~3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from vfarcic/dot-agent-deck at commit c5d24e7, republished under its MIT licence (© vfarcic). 1,771 words, ~3,013 tokens.

Download SKILL.mdSave it as .claude/skills/pr-create/SKILL.md (or your agent's skills folder).
name
pr-create
description
Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off. Never merges; arms auto-merge only when nothing downstream gates the merge. Use when work on a branch is finished and needs to become a reviewed PR, whether or not a PRD started it.
user-invocable
true

Create a pull request and drive it to verified

Owns one arc: committed work on a branch → a PR that is green, reviewed, and ready for someone to merge.

Out of scope on purpose:

  • Merging. Nothing here merges. Arm auto-merge and hand off — CLAUDE.md rule 8: nobody merges their own unapproved PR, and for an admin that succeeds silently rather than failing, which is how the gate decays into ceremony.
  • Closing the issue. Closes #<n> in the PR body does it on merge.
  • Implementation. If a finding needs a code change, make it; if it needs a decision you cannot make, say so on the thread.

This file is project-local and owned by this repository — it was forked out of the dot-ai mirror precisely so corrections survive (CLAUDE.md rule 13). Edit it freely.

1. Before the PR

  • CLAUDE.md is the authority on the gates (rules 2 and 5). Run cargo xtask affected-checks and run what it prints (issue #1575): for a branch with any Rust, build input or unmapped path in it, that is cargo fmt --check, cargo clippy --workspace --all-targets --features e2e,e2e-live -- -D warnings and cargo test-fast; for a branch that is only mapped text (docs, skills, changelog.d/, .github/, PRDs, CLAUDE.md and the like), it is the xtask tests plus the root-package tests that read those files. cargo xtask affected-checks --run prints the plan and runs it, stopping at the first failure — prefer it to piping the printed plan into a shell, which reports success when the helper itself fails to build or run. Either way add the tests covering what you touched — name those in the PR body, since part of the tier runs on no runner anywhere. Read the rules rather than trusting this summary.
  • A red you met along the way is yours (CLAUDE.md rule 6), whoever caused it and even if it passed on a retry: after rule 6's isolation rerun, fix it in this PR or quarantine it (a named owner, an expiry issue, and #[ignore = "quarantined: <owner>, #<issue>"] on the test), and say in the PR body which, for each one. Rerunning it until green and mentioning it is neither; .claude/skills/verify-pr/SKILL.md Phase 5 has the mechanics. Before fixing a red your change did not cause, check whether an open PR already fixes it (gh pr list --search '<test name>'); if one does, name that PR in the body and leave the red to it — #1460 and #1462 both changed selectRow in desktop/driver/harness.ts to fix terminal_002.
  • Changelog fragment: changelog.d/<issue>.<type>.md, type one of breaking|feature|bugfix|doc|misc. Release notes are built from these, not from PR labels.
  • User docs (rule 21): a user page under docs/ the PR touches, and the changelog fragment, say what the user does, sees and configures; implementation detail and issue history go to docs/develop/ or nowhere, fixed in this PR.
  • Rule 12 if the change touches the daemon, protocol, orchestration or hooks: answer the PROTOCOL_VERSION-vs-.breaking.md question explicitly in the PR body, including the cross-version manual test. The cross-version CI job runs that test for any PR that changes the binary (docs/develop/cross-version-harness.md, "In CI"): once it passes on the head, link its run in the PR body as the record. When it reports a PROTOCOL_VERSION move instead, record that refusal as the outcome. Run cargo xver yourself only to diagnose a red job, or for a pairing the job does not set up. A .breaking.md fragment also needs its CONTRACT_BREAKS entry in src/daemon_protocol.rs (issue #801) — xtask/linkage-check fails the build without it.
  • Working tree clean, branch pushed. Never push to main — it is protected and returns GH013.

2. Open it

gh pr create, with a body that says what changed and why, how it was verified (name the tests), and Closes #<n> — or Refs #<n> for a PR that ships only part of a PRD, which leaves the PRD open (the orchestrator template's step 5 in .dot-agent-deck.toml has the rule).

Then gh pr edit <n> --add-reviewer <the other maintainer> — but request review last, after CI and the automated review have settled and you have pushed the fixes. dismiss_stale_reviews_on_push voids an approval on any later push, so asking early buys a guaranteed second round trip.

3. Settle CI and the automated review

Wait for the check-runs. gh pr checks <n> reports both CI and the reviewer's own check-run.

The wait must be bounded. An automated reviewer that is out of quota, uninstalled, or broken produces no check-run at all — there is no message and no failed state, so "wait until it appears" never terminates. Measured on this repo 2026-08-23: an exhausted Greptile quota produced zero comments and zero check-runs, indistinguishable from the app being gone.

So: give the reviewer a budget (~15 minutes from PR creation is ample; it normally lands in 3–5). If no reviewer check-run exists when the budget expires, proceed and say so explicitly in your report — "no automated review was obtained" is a result. Do not hang, and do not report the gate as passed. Never block on a reviewer that is not configured here at all.

A check that goes red here falls under the same rule as one met locally (step 1): a test that fails in CI and then passes on gh run rerun is a flaky test you have now met, so it gets a fix or a quarantine in this PR, not just the green rerun. The PR body was written before this wait, so add the red and its exit to it (gh pr edit <n> --body-file <file>) once you have taken one — the body, not only your report, is what step 1 asks to carry it.

A green check-run is not the review. The findings live only in the inline comments:

sh
gh api repos/{owner}/{repo}/pulls/<n>/comments --paginate

Keep --paginate — replies count toward the page, so a busy PR silently truncates the findings you are about to certify as read. The summary comment and the review object carry none of them, and a COMMENTED review with a passing check can still carry real defects.

Show full SKILL.md (779 more words)Show less

4. Answer and resolve every finding

For each one: fix it, or reply saying why not. Then resolve the thread.

Resolving is not bookkeeping — it is half the job, and skipping it blocks the merge twice over:

  • required_review_thread_resolution is on, so an unresolved thread blocks the merge button;
  • the agent PR reviewer skips any PR carrying one, so it also blocks the approval that merge needs.

Measured: #1035 sat a full day with 14 green checks and every finding already fixed, and #1019 sat two days on a finding the reviewer itself had retracted 34 seconds after the author rebutted it.

sh
# unresolved thread ids — PAGINATED. `first:100` is a page, not the total, and
# resolved threads stay in the connection, so a second unpaginated call returns
# the same first 100 and silently hides the rest.
gh api graphql --paginate \
  -f query='query($o:String!,$r:String!,$n:Int!,$endCursor:String){repository(owner:$o,name:$r){
    pullRequest(number:$n){reviewThreads(first:100,after:$endCursor){
      pageInfo{hasNextPage endCursor} nodes{id isResolved path}}}}}' \
  -f o=<owner> -f r=<repo> -F n=<pr> \
  --jq '.data.repository.pullRequest.reviewThreads.nodes[] | select(.isResolved==false) | .id'

gh api graphql -f query='mutation($t:ID!){resolveReviewThread(input:{threadId:$t}){thread{isResolved}}}' -f t=<id>

Re-run the query after resolving and confirm it returns nothing. "I resolved the ones I saw" is not the same claim as "none is unresolved".

Resolve only what you actually addressed. Leave a thread open when you are waiting on the commenter, or when they asked something they still need answered there — a thread closed over unaddressed feedback is worse than one left open.

Push fixes before requesting review (step 2). Greptile does not re-review — greptile.json sets triggerOnUpdates: false — but Qodo, evaluated beside it, re-reviews on every push (.pr_agent.toml). So after your last push, wait for Qodo too, and repeat this step for anything new it raised; a finding raised on your final push and left unanswered is still an unresolved thread blocking the merge. Do not wait the way step 3 waits: that wait keys on a check-run, and Qodo creates none — measured 2026-09-24 on #1235, #1257 and #1271, where the head commit's check-runs carry Greptile Review and nothing from Qodo, so "until its check appears" never returns. Wait on its summary comment instead: it edits one comment in place rather than posting a new one, and the body names the head SHA it reviewed, so the review is done for this push when that body cites your head. Bound it the same way regardless, and proceed reporting that no re-review was obtained. Its medium and above findings arrive at the same endpoint as Greptile's and are answered and resolved the same way; its informational tier stays in that summary comment (inline_comments_severity_threshold = 2), so read it as well as the inline endpoint.

5. Hand off

Report and stop: PR URL, check status, each finding and what you did about it, each red you met and whether it was fixed or quarantined, and whether an automated review was obtained at all.

Arm auto-merge only when nothing downstream gates the merge. gh pr merge <n> --auto --squash waits for exactly what a manual merge needs — the approval, the required checks, every thread resolved — so against a bare hand-off it is not a bypass.

It is not a way to wait for a check that is not required. The required contexts here are build, build-macos, build-windows, security and e2e-deterministic (verified 2026-09-22 against gh api repos/vfarcic/dot-agent-deck/rules/branches/main, ruleset rules only; the classic protection on main requires none); everything else on the page — renovate/stability-days, Greptile Review, notify-main-red — holds nothing, on either path. Armed at hand-off, before the approval, the PR reads BLOCKED, gh arms it, and GitHub merges it the moment the approval lands, pending unrequired check or not — autoMergeRequest is set, so nothing looks wrong. Run after the approval with only an unrequired check outstanding, the PR is already mergeable (UNSTABLE) and gh drops the flag and merges on the spot, because isImmediatelyMergeable counts UNSTABLE (pkg/cmd/pr/merge/merge.go:593 and :826, at gh v2.100.0): exit 0, silent when stdout is not a terminal, and gh pr view <n> --json autoMergeRequest reading null after the call is the tell. #1208, already approved, landed a lockfile update inside Renovate's stability window that second way (#1222). So read gh pr checks <n> before arming; if an unrequired check is pending and matters, do not arm — poll that check and merge when it passes, or leave the PR disarmed for a person. CLAUDE.md rule 8.

It is a bypass when your caller has a gate of its own after this step. The orchestrated PRD workflow is exactly that shape: .dot-agent-deck.toml runs this skill at step 5, then builds a demo reel at step 6, then asks the user for an explicit merge go-ahead at step 7. Arming here would land the PR the moment the approval arrived — during step 6, with nobody present — and nothing later can un-arm it, because re-running this skill only arms it again.

So the default is do not arm; arm only when you are the last gate. If you were told to stop before merge, that is your answer: report and leave it disarmed.

Reference

  • docs/develop/governance.md — the ruleset, bypass actors, who may merge, the emergency override.
  • CLAUDE.md rules 2, 5, 6, 8, 12, 13.

© vfarcic, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/pr-create of vfarcic/dot-agent-deck.

Open the folder on GitHubat commit c5d24e7

Compare with similar skills

PR Create next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

PR Create compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
PR Create this skillvfarcic/dot-agent-deck109—~3kAutomated safety check: PassMIT
Codebase Modernizerluongnv89/skills131—~5kAutomated safety check: PassMIT
Astro Code Reviewwithastro/astro63k—~7.6kAutomated safety check: PassCustom licence
Specwarpdotdev-demos/cloud-factory-demo320—~2.1kAutomated safety check: PassMIT
Panel Reviewstacklok/mecatl218—~5.8kAutomated safety check: PassApache-2.0
Schematicblader/schematic239—~2.2kAutomated safety check: PassMIT

Similar skills

  • Codebase Modernizer

    luongnv89/skills

    Audit a stale, inherited, or messy codebase — deps, bugs, security, tests, CI, docs, UI/UX — then emit a phased, testable modernization plan.

    131 GitHub stars~5k tokensUpdated yesterday
    DevelopmentAuto-check passed

  • Astro Code Review

    withastro/astro

    Official

    Perform a static, read-only code review of an Astro pull request or of a local branch, commit range, diff, patch, or working tree being prepared as a pull request.

    63k GitHub stars~7.6k tokensUpdated today
    DevelopmentAuto-check passed

  • Spec

    warpdotdev-demos/cloud-factory-demo

    Coordinate spec-driven development for a GitHub, Jira, Linear, or other issue-tracker issue marked ready-to-spec by using write-product-spec and write-tech-spec, creating PRODUCT.md and TECH.md…

    320 GitHub stars~2.1k tokensUpdated 1 mo ago
    DevelopmentAuto-check passed

  • Panel Review

    stacklok/mecatl

    Review completed non-trivial code across four independent axes: Spec, Standards, Test adequacy, and installed Domain specialists.

    218 GitHub stars~5.8k tokensUpdated yesterday
    DevelopmentAuto-check passed

  • Schematic

    blader/schematic

    Reverse engineer a detailed product and technical specification document from a git branch's implementation.

    239 GitHub stars~2.2k tokensUpdated 7 mo ago
    Product & Project ManagementAuto-check passed

  • Prp Issue

    Wirasm/prp

    Autonomously owns one workstream from an issue, PRD, document, existing plan, or free-form request through planning, implementation, pull request, independent review, corrections, and green CI.

    2.3k GitHub stars~2.5k tokensUpdated 5 days ago
    Product & Project ManagementAuto-check passed

More from vfarcic/dot-agent-deck

All 23 skills in this repo

  • Dispatch Shape

    vfarcic/dot-agent-deck

    Choose the shape of a unit you are about to dispatch in this repo — one agent (--single) or a team (--orchestration '<name') — from divisibility criteria instead of asking, and report the shape you…

    109 GitHub stars~2.8k tokensUpdated yesterday
    Auto-check passed

  • Docs Screenshots Review

    vfarcic/dot-agent-deck

    Check that a change to the user-facing docs covers both clients (the TUI and the desktop app) unless the feature exists in only one, and decide whether it needs a new or updated screenshot, then…

    109 GitHub stars~1.8k tokensUpdated yesterday
    Auto-check passed

  • Dot AI Request Dot AI Feature

    vfarcic/dot-agent-deck

    Generate a feature request prompt for another dot-ai project.

    109 GitHub stars~679 tokensUpdated yesterday
    Auto-check passed

  • Publish Docs

    vfarcic/dot-agent-deck

    Publish the docs site to GHCR with a main-<sha tag and bump site/helm/values.yaml so Argo CD picks it up — without cutting a SemVer release.

    109 GitHub stars~1.7k tokensUpdated yesterday
    Auto-check passed

  • Run Dot Agent Deck

    vfarcic/dot-agent-deck

    Run, build, smoke-test, and screenshot the dot-agent-deck binary against an isolated sandbox.

    109 GitHub stars~2.1k tokensUpdated yesterday
    Auto-check passed

  • Demo Reel

    vfarcic/dot-agent-deck

    Stitch a manifest of terminal recordings into one narrated MP4 (title/description card, then clip, repeated) and optionally upload it privately to YouTube.

    109 GitHub stars~5.2k tokensUpdated yesterday
    Auto-check: notes

Categories

Questions about PR Create

What does PR Create do?

Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off. PR Create is an agent skill from vfarcic/dot-agent-deck. Take committed work from a branch to a verified pull request — push, open the PR, settle CI and the automated review, answer and resolve every finding, and hand off.

When should I use PR Create?

PR Create fits situations like: work on a branch is finished and needs to become a reviewed PR; not a PRD started it.

How do I install PR Create in Claude Code?

Run `npx skills add vfarcic/dot-agent-deck --skill pr-create -a claude-code`. Or copy the skill folder (.claude/skills/pr-create in vfarcic/dot-agent-deck) into .claude/skills/pr-create in your project. Claude Code loads it when a task matches its description.

How do I install PR Create in Codex?

Run `npx skills add vfarcic/dot-agent-deck --skill pr-create -a codex`. Or copy the skill folder (.claude/skills/pr-create in vfarcic/dot-agent-deck) into .agents/skills/pr-create in your project. Codex loads it when a task matches its description.

Can I use PR Create in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add vfarcic/dot-agent-deck --skill pr-create -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/pr-create, .gemini/skills/pr-create, .github/skills/pr-create and .opencode/skills/pr-create in your project.

What does PR Create need to run?

Going by SKILL.md and its folder, PR Create needs the command-line tools its instructions call (gh and cargo).

Does PR Create access the network?

SKILL.md contains no URLs. Its commands use gh, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is PR Create safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does PR Create use?

PR Create is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does PR Create use?

About 3k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to PR Create?

Skills that share tags, products or a category with PR Create: Codebase Modernizer (luongnv89/skills, 131 stars), Astro Code Review (withastro/astro, 63k stars), Spec (warpdotdev-demos/cloud-factory-demo, 320 stars) and Panel Review (stacklok/mecatl, 218 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains PR Create?

vfarcic (a GitHub user) maintains it in vfarcic/dot-agent-deck, which has 109 GitHub stars. The repository holds 23 skills in this directory. The repository was last updated on October 7, 2026.

Source: vfarcic/dot-agent-deck on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.