AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Github Workflows

skill-atlasomnia-donna-starter-github-workflows · by AtlasOmnia

github-workflows — Consolidate GitHub auth, repository management, PR lifecycle, issues, and code review workflows.

No reviews yet
0 installs
0 views
view→install

Install

$ agentstack add skill-atlasomnia-donna-starter-github-workflows

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access Used
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets Used
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-atlasomnia-donna-starter-github-workflows)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
5d ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Github Workflows? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

GitHub Workflows

Use this skill for the full GitHub operating loop: 1) authenticate, 2) manage repos and remotes, 3) open and review PRs, 4) work on issues, and 5) monitor CI.

Why this umbrella exists

The old project-specific skills (github-auth, github-repo-management, github-pr-workflow, github-issues, github-code-review) all target the same operational domain: repository setup, PR lifecycles, issue triage, and review. Consolidating them avoids fragmented workflows and makes the sequence from local repo state to merged code explicit.

Core entrypoints

  1. Authentication first
  • Set up GitHub access with gh where available, or token-based/SSH fallback.
  • Confirm auth method before attempting repo/PR operations.
  1. Repository foundation
  • Clone, create, or fork repositories and confirm remotes.
  • Configure defaults, visibility, topics, secrets, branch protection, and releases.
  1. Issue workflow
  • List/search/triage issues.
  • Create labels, assignments, and status updates.
  • Use templated issue creation for bug reports and feature requests.
  1. PR execution
  • Create topic branches with naming conventions.
  • Push changes and open PRs.
  • Monitor checks and handle failures.
  • Merge with explicit strategy once checks are green.
  1. Code review loop
  • Review local diffs and remote PR diffs.
  • Post general and inline comments.
  • Provide a structured review outcome: critical / warning / suggestion / looks good.

Recommended practical sequence

  • Project-idea incubation:
  • Use one private incubator repository for exploratory technical ideas.
  • Give each idea its own folder while it is being researched or lightly prototyped.
  • Promote a validated independent product into its own repository, leaving a compact link and decision record behind.
  • New repository work:
  • github-auth equivalent setup → github-repo-management operations.
  • Feature delivery:
  • Branch and commit changes → github-pr-workflow to open PR.
  • Track checks; on failure, fix and repush.
  • Post review comments and finalize via merge.
  • Maintenance and support:
  • Use issues to triage incoming work and route effort.
  • Use code review skill patterns before merging any externally contributed change.

PR + issue operating matrix

When a task spans both code and tracking, prefer this order:

  1. Ensure repo auth is valid.
  2. Confirm branch and remote are current.
  3. If needed, create/clone the repo and verify contributor permissions.
  4. For new work, open or update an issue first, then PR.
  5. Run or track CI before requesting final merge.
  6. Post review and close loop with state updates.

Curated staging and PR-body safety

Before the first commit in a public repository, verify git config user.name and git config user.email. If either is empty or would fall back to a machine-local identity such as user@hostname.local, set the repository-local identity to the authenticated GitHub account and its noreply address before committing. If caught after a local commit but before push, amend with git commit --amend --no-edit --reset-author, then verify both author and committer fields.

For large working trees, do not default to git add .. Stage an explicit path list, then:

  1. Run git status --short and investigate every MM entry. If a generator ran after staging, restage the final generated output so the committed file is the one that was actually tested.
  2. Run git diff --cached --check and scan the staged diff for secrets.
  3. Compare git diff --cached --name-only against a forbidden list covering credentials, result bundles, agent-session directories, build output, and known duplicate files.
  4. After push, verify local HEAD equals the remote branch SHA.

For multiline PR descriptions—especially Markdown containing backticks, $(), or shell metacharacters—write the body with the file tool and call gh pr create --body-file or gh pr edit --body-file . Do not pass rich Markdown inside a double-quoted --body shell argument: command substitution can execute text inside backticks and silently corrupt the PR description. Read the PR back with gh pr view --json body,url,state before reporting success.

If gh pr checks reports that no checks exist, say so explicitly; a successful push is not CI verification.

Public GitHub profile setup

Treat a GitHub profile as two separate deliverables:

  1. Account metadata — display name, bio, website, and optional location, updated through PATCH /user.
  2. Profile READMEREADME.md in a public repository whose owner and repository names exactly match the authenticated username.

Preflight with gh auth status and gh api user; do not assume repository scopes permit profile edits. The REST profile endpoint requires the OAuth user scope. If PATCH /user returns HTTP 404 with a scope hint, request only that scope using gh auth refresh -h github.com -s user -c. GitHub's device authorization must be approved by the user; do not enter credentials or approve permission dialogs on their behalf. Resume only after the refresh exits successfully and gh auth status confirms the scope.

For the profile README, use the established public/noreply Git identity, scan the staged content for secrets and PII, push it, then verify the rendered public profile rather than only the repository API response. Keep community-facing profiles credibility-first: concise project summaries, verified links, and an affiliation disclaimer when independent community work could be mistaken for an official vendor account.

Quick fallbacks when gh is unavailable

  • Keep GH token/env credentials ready from secure sources (GITHUB_TOKEN in .env or the OS keychain).
  • Use curl REST endpoints for auth-required actions and include explicit owner/repo extraction.
  • Preserve scope boundaries: issue endpoints and PR endpoints are intentionally separate and should not be mixed with repo creation commands.

Multi-repository sync and reconciliation

When asked to scan “all repos,” treat it as an inventory and classification task rather than repeating git pull blindly:

  • State the filesystem roots scanned and any exclusions.
  • Discover both .git directories and .git files, then deduplicate fetches with git rev-parse --git-common-dir so linked review worktrees do not trigger redundant network operations.
  • Fetch first, calculate exact ahead/behind counts, inspect tracking and no-upstream branches, and check live GitHub PR mergeability.
  • Separate canonical clones from duplicate clones, backups, malformed refs, detached campaign worktrees, and archival branches.
  • Never infer “needs merge” from ancestry counts or git branch -r --no-merged alone; compare patch IDs and final trees when squash merges or duplicate commits are possible.
  • Do not modify dirty or diverged repositories during a scan-only request.

When the user asks to continue into reconciliation:

  • preserve tracked and untracked work outside the repository plus a stash/archive branch before changing ancestry;
  • evaluate reapplied WIP against current source and gates rather than assuming every local change should be published;
  • refuse to merge stale work that reintroduces regressions merely to make the tree clean;
  • repair malformed refs only after recording reachability and preserving unique commits;
  • finish with repository gates, staged/public privacy scans, remote-SHA equality, clean canonical checkouts, and a fresh all-PR/all-branch scan;
  • distinguish Git blockers, product release gates, dependency advisories, and intentional archives in the report.

Destructive cleanup and archive retirement

Treat synchronization and archival cleanup as separate phases. A canonical repository can be fully reconciled even while duplicate snapshots, campaign worktrees, backup branches, or stashes remain intentionally preserved.

  1. Prove the canonical clone is clean and equal to its remote before touching any duplicate.
  2. Preserve unique branch tips with a verified git bundle, and preserve dirty work with an external patch/tar archive.
  3. Execute destructive operations serially, never in a parallel batch. Approval systems may allow one deletion and deny another, leaving partial state that must be re-read before continuing.
  4. After every approved deletion, verify the path/ref is absent and the canonical clone is unchanged.
  5. If an approval is denied, do not rephrase or route around it. Stop that cleanup item, leave the archive classified as noncanonical, and continue only with read-only verification.
  6. A directory containing .git files may be an unregistered review copy even when git worktree list shows only the canonical checkout. Classify it from both views before removal.

Staged privacy-scan precision

Do not scan staged diffs with an unbounded grep 'sk-': ordinary package-lock names such as task-* can match and create false alarms. Use token boundaries and realistic minimum lengths (for example (?<![A-Za-z])sk-[A-Za-z0-9_-]{20,}), scan without printing matched secret values, and keep local-path/PII checks separate from credential checks.

For explicitly authorized deletion of whole remote GitHub repositories, including complete git bundle backups, dirty-work sidecars, the delete_repo OAuth device flow, serial DELETE calls, and absence verification, follow the deletion steps in this skill.

Investigating “branched but not on main” reports

Treat this as a provenance question, not merely a branch-name search:

  1. Resolve the exact repository first. If the name is uncertain, search session history or the authenticated account’s repository list; once an exact owner/repo is known, query it directly with gh repo view owner/repo. Broad repository-enumeration filters can omit private, transferred, archived, or otherwise unexpectedly classified repositories.
  2. Inspect all three surfaces:
  • current branches: gh api --paginate repos/OWNER/REPO/branches?per_page=100
  • PR history: gh pr list --repo OWNER/REPO --state all
  • raw refs: git ls-remote https://github.com/OWNER/REPO.git
  1. Inspect repository events and main history when the expected branch is absent:
  • gh api --paginate 'repos/OWNER/REPO/events?per_page=100'
  • gh api 'repos/OWNER/REPO/commits?sha=main&per_page=30'

A local-only topic branch may have been fast-forwarded or pushed directly to main, leaving no remote branch and no PR. A deleted branch may still appear in repository events or refs/pull/*.

  1. Prove ancestry before concluding anything: record the old base, current main SHA, commit parents, and compare range. Distinguish “merged through a PR,” “fast-forward/direct push to main,” “branch exists but is unmerged,” and “local-only work not present on GitHub.”
  2. If the change is already on main, say plainly that nothing needs merging. Separately flag process concerns such as bypassing PR review; do not manufacture a merge task merely because the work began on a branch.

Self-hosted runner deployment

Use self-hosted runners to preserve GitHub PR/check workflows without consuming GitHub-hosted minutes.

  • Personal-account repositories only support repository-level runners. Organization-level runners are required for one runner pool shared across multiple repositories.
  • Restrict persistent self-hosted runners to private, trusted repositories. Route jobs with both self-hosted and a machine-specific custom label; never rely on an OS label alone.
  • Download runner packages from the official actions/runner release and verify the release asset SHA-256 digest before extraction.
  • Treat registration tokens as short-lived secrets. For remote Windows setup, transfer the token through a protected temporary file, consume it locally, and delete it immediately; command-line quoting can corrupt tokens and expose them in process listings.
  • On macOS, register as the normal user and install the provided LaunchAgent with svc.sh. Confirm both launchctl state and GitHub's runner API status.
  • On Windows, prefer the runner's NETWORK SERVICE service mode over LocalSystem. If tests require symlink creation, enable Windows Developer Mode rather than elevating the runner service. Match GitHub-hosted tooling explicitly (for example, checksum-verified portable PowerShell 7 when workflows use shell: pwsh).
  • A WSL user-systemd service may disappear when the distro idles after the initiating shell exits. If that occurs, use a Windows logon scheduled task that keeps wsl.exe ... run.sh in the foreground, then verify the runner remains online after the administration session disconnects.
  • Do not stop at an online status check. Route a real branch/PR workflow, fix environmental parity failures, merge only after all jobs pass, and verify the post-merge main run on the deployed workflow.

Verification checklist

  • auth method validated (gh auth status or equivalent token path)
  • repo and remotes inspected before creating PRs
  • issue/PR state checked before merging
  • branch/merge strategy chosen intentionally (merge, squash, rebase)
  • review outcomes stored with severity buckets
  • security-sensitive steps (secrets) handled through approved tools and approvals

Support-file equivalents from merged skills

Legacy skill assets now live in the archived originals under:

  • github-auth: inline auth setup block
  • github-code-review: inline review checklist
  • github-issues: templates/bug-report.md and templates/feature-request.md
  • github-pr-workflow: PR templates
  • github-repo-management: repo setup checklists

When not to use this skill

  • If the request is only a one-off gh command, use the direct gh call.
  • For non-GitHub issue/PR tasks, use the project-specific skill that matches the actual domain

(e.g., local code review logic, testing, or architecture workflows).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.