Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    GitHub Actions CI Fixer and Secure Workflow Builder

    1

    Finds the real root cause of a failing GitHub Actions run and returns a minimal diff, or writes a fast least-privilege workflow. Verdict SHIP / HARDEN / BLOCK.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    GitHub Actions CI Fixer and Secure Workflow Builder

    GitHub Actions CI Fixer and Secure Workflow Builder

    Example session with this skill installed

    Our test workflow went red after we moved to actions/upload-artifact v7. The tests pass but 2 of the 3 matrix legs fail at the upload step. Run 18342, can you find the cause and give me the fix?

    • Read your context and instructions
    • Compiled the github actions ci

    Verdict: HARDEN

    Mode: FIX · Workflow: .github/workflows/test.yml · Run: 18342 · Linted: yes (0 errors)

    Root cause

    Category: Action version change (upload-artifact v4+ behavior)
    Evidence (first error, job test (ubuntu-24.04, 22)):

    ##[error]Failed to CreateArtifact: Received non-retryable error: Failed request: (409) Conflict: an artifact with this name already exists on the workflow run
    

    Why: all three matrix legs upload with name: coverage. Since v4, artifacts are immutable and names must be unique per run, so the first leg wins and the other two fail. The npm test steps passed in every leg; the red X is only the upload.

    Fix

           - uses: actions/upload-artifact@<sha> # v7.0.2
             with:
    -          name: coverage
    +          name: coverage-${{ matrix.os }}-node${{ matrix.node }}
               path: coverage/
    

    In the report job, download all legs at once:

           - uses: actions/download-artifact@<sha> # verify current major
             with:
    -          name: coverage
    +          pattern: coverage-*
    +          merge-multiple: true
    

    Checklist

    ItemStatus
    Top-level permissions: contents: readmissing, add
    Third-party actions pinned to SHA2 of 4 on tags
    timeout-minutes per jobmissing
    concurrency cancels superseded PR runsmissing
    setup-node cache: npmpresent
    Untrusted ${{ }} in run:none
    pull_request_targetnot used
    actionlint0 errors

    Also noticed (not changed)

    • Hidden files are skipped by default in v4+; if the report needs coverage/.nyc_output, add include-hidden-files: true.
    • Add permissions, timeout-minutes: 15 and a concurrency group in a separate PR to move this to SHIP.

    Verify

    actionlint
    git push && gh run watch
    

    Green looks like: three coverage-* artifacts on the run and the report job succeeding.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A red GitHub Actions run usually ends with "Process completed with exit code 1", which tells you nothing. The real cause is often higher up: a runner image that moved under ubuntu-latest, an action major that changed behavior, a token missing one scope, or a matrix leg that collided with another. And new workflows copied from old examples tend to ship with write-all tokens, unpinned actions and no timeouts.

    This skill gives your coding agent a fixed procedure for both jobs.

    • FIX mode: pulls the failed log with the gh CLI (or reads a pasted one), compares against the last green run, quotes the first causal error, and classifies it into one category such as permissions, OIDC, cache, runner image, artifact behavior, matrix, path filters, YAML expressions or flaky test
    • Returns the smallest unified diff that fixes that cause, and lists anything else it noticed separately instead of rewriting your file
    • BUILD mode: detects your stack from lockfiles and writes a workflow with built-in setup-* caching, concurrency groups, least-privilege permissions, timeouts, SHA-pinned actions and Dependabot for action updates
    • Security rules for pull_request_target, workflow_run, script injection and fork secrets, including the actions/checkout v7 fork-checkout block
    • Runs actionlint and reports the result
    • One verdict for both modes: SHIP, HARDEN or BLOCK

    A reference file holds the current action majors, runner image mappings, the Node 20 removal timeline and a table of log signatures with fixes, dated October 2026 with sources.

    For developers, maintainers and platform engineers who own CI on GitHub.

    Honesty note: it does not invent versions, SHAs or speedup numbers. It tells the agent to re-check versions on the web, resolve SHAs live, and say "not linted" or "cause not proven" when that is the case.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for DevOps & Cloud.

    It reads logs and workflow files only, so problems in repository settings, branch protection, secret values or cloud trust policies are flagged as decisions for you rather than fixed. Version and runner facts are dated October 2026. It does not cover CI systems other than GitHub Actions, and it will not rerun or dispatch workflows without your approval.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Software
    PythonGit
    Optional
    GitHub CLIactionlint

    Without GitHub CLI, paste the failed job log.

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    A SKILL.md with the FIX and BUILD procedures, verdict rules, output template and checklist, plus references/actions-facts.md with current action versions, runner image mappings, the Node 24 runtime timeline, a failure-signature table, security rules, a workflow template and actionlint usage, with sources.

    Any agent that loads SKILL.md skills, such as Claude Code, Codex CLI, Cursor, VS Code Copilot and Gemini CLI.

    Unzip the folder into your agent's skills directory (for Claude Code, ~/.claude/skills/ or .claude/skills/ in a repo). The gh CLI and actionlint are optional but let the agent fetch logs and lint workflows itself; without them you paste the log and it reports the workflow as not linted.

    It works from logs and workflow files, so it cannot see repository or organization settings, branch protection or secret values. Version facts are dated October 2026 and should be re-checked with web access. It focuses on GitHub-hosted Actions, not other CI systems.

    It forces the agent to quote the first causal error instead of the last line, pick one category, keep the diff minimal, check versions instead of guessing them, and apply a fixed security checklist before giving a SHIP, HARDEN or BLOCK verdict.

    No. If gh is installed and signed in, the skill pulls the failed log itself with gh run view --log-failed. Otherwise paste the failing job log and your workflow file. actionlint is optional but recommended for checking the result.

    New in DevOps & Cloud