← Back to Skills Marketplace
martok9803

CI Whisperer

by MartinKirov · GitHub ↗ · v0.2.0
cross-platform ✓ Security Clean
632
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install martok9803-ci-whisperer
Description
Analyze GitHub Actions failures and propose fixes. Use when a user shares a failing GitHub Actions run URL/id, says 'CI is failing', asks 'why did this workf...
README (SKILL.md)

CI Whisperer

Fetch GitHub Actions run details, pinpoint the failure, and propose a minimal fix.

This skill is meant to feel like a senior engineer doing a fast “CI autopsy”.

Modes

Read-only mode (default)

  • Collect evidence, explain root cause, propose fixes.
  • No pushes, no PRs, no branch creation.

PR fix mode (opt-in)

PR fix mode is allowed only when both are true:

  1. The user explicitly asks to open a PR.
  2. A local toggle is enabled (the “on/off button”):
    • env var: CI_WHISPERER_WRITE=1

If the toggle is not enabled, refuse politely and explain how to enable it.

Workflow

1) Identify target run

Accept any of:

  • workflow run URL
  • run id
  • PR number (then locate latest run)

Always determine:

  • owner/repo
  • run id

If the user didn’t specify the repo, ask for it (or infer from context).

2) Gather evidence (tool-backed)

Prefer deterministic tooling. Use /usr/bin/gh when the system has multiple gh binaries.

Suggested commands:

  • gh run view \x3Crun-id> --repo owner/repo --json status,conclusion,createdAt,updatedAt,event,headBranch,headSha,url,name
  • gh run view \x3Crun-id> --repo owner/repo --log-failed
  • gh run view \x3Crun-id> --repo owner/repo --log (only if needed; can be noisy)

If gh is not authenticated, stop and ask the user to run:

  • /usr/bin/gh auth login

3) Produce a "CI Autopsy" report

Return:

  • failing job(s) and step(s)
  • the exact error excerpt (short; redact secrets)
  • likely root cause(s) ranked
  • minimal fix options
  • confidence level

4) (Optional) Open a PR (only with explicit approval + write toggle)

If the user asks to fix it and CI_WHISPERER_WRITE=1:

  • create a branch
  • apply minimal changes
  • run local lint/tests if available
  • open PR with a clear description and link to the failing run

If the user asks but write mode is OFF:

  • provide the patch/diff instructions, but do not push.

Safety

  • Never print tokens.
  • Don’t open PRs or push changes unless explicitly requested.
  • If logs contain secrets, redact before quoting.

Bundled scripts

Use scripts for repeatable fetching and parsing:

  • scripts/ci_autopsy.py (fetch run metadata + failed logs)
Usage Guidance
This skill is coherent with its stated purpose, but review the following before installing: 1) It relies on the GitHub CLI (`gh`) and your gh-authenticated account — the metadata didn’t declare that dependency, so ensure `gh` is present and authenticated. 2) Logs can contain secrets; the SKILL.md tells the agent to redact secrets but the helper script prints raw logs. Don’t enable automated PR creation (set CI_WHISPERER_WRITE=1) unless you trust the session and are prepared for the agent to create branches/PRs; enable it only for sessions where you want that behavior. 3) If you will allow PR creation, consider reviewing the generated diff/patch locally before pushing. If you want higher assurance, request the maintainer add automatic log-redaction and update metadata to declare `gh` and the optional env var.
Capability Analysis
Type: OpenClaw Skill Name: martok9803-ci-whisperer Version: 0.2.0 The OpenClaw AgentSkills bundle 'ci-whisperer' is classified as benign. The `SKILL.md` clearly outlines a legitimate purpose (analyzing GitHub Actions failures) and implements robust safety measures, such as requiring explicit user approval and an environment variable (`CI_WHISPERER_WRITE=1`) for any write operations (e.g., creating PRs). The `scripts/ci_autopsy.py` script uses the GitHub CLI (`gh`) in a read-only capacity, explicitly preferring `/usr/bin/gh` to prevent path hijacking, and uses `subprocess.run` with a list of arguments, which is a safer method against shell injection. There is no evidence of malicious intent, data exfiltration, unauthorized execution, or prompt injection designed to bypass security controls.
Capability Assessment
Purpose & Capability
The name/description match the included instructions and the bundled script: both rely on the GitHub CLI to fetch run metadata and failed logs and to optionally create branches/PRs. However, the registry metadata declares no required binaries or env vars even though SKILL.md and the script explicitly rely on the `gh` CLI and an optional env var `CI_WHISPERER_WRITE` — those runtime dependencies are not reflected in the metadata.
Instruction Scope
Instructions are narrowly scoped to fetching run details/logs and producing a report; PR creation is gated behind explicit user approval and the CI_WHISPERER_WRITE toggle. However, SKILL.md instructs the agent to redact secrets while the included script simply prints raw `gh run view --log-failed` output (no automatic redaction). That creates an operational risk: logs can contain secrets or tokens and the tool does not sanitize them automatically — the agent/user must do the redaction.
Install Mechanism
No install spec (instruction-only plus a small helper script). No downloads or external installers. Low risk from install mechanism.
Credentials
The skill does not request secrets in metadata and relies on the host's `gh` authentication (user's GitHub credentials) — which is expected. The SKILL.md references an optional `CI_WHISPERER_WRITE` env var to enable write/PR mode, but this env var is not declared in the registry metadata (minor inconsistency). The bundled script itself does not read environment variables aside from preferring /usr/bin/gh.
Persistence & Privilege
always:false and no modification of other skill/system configurations. The skill does not request persistent or elevated privileges. Autonomous invocation is allowed (platform default) but is not combined with other red flags.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install martok9803-ci-whisperer
  3. After installation, invoke the skill by name or use /martok9803-ci-whisperer
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.0
Add PR fix mode toggle docs + improved README.
v0.1.0
Initial public release.
Metadata
Slug martok9803-ci-whisperer
Version 0.2.0
License
All-time Installs 0
Active Installs 0
Total Versions 2
Frequently Asked Questions

What is CI Whisperer?

Analyze GitHub Actions failures and propose fixes. Use when a user shares a failing GitHub Actions run URL/id, says 'CI is failing', asks 'why did this workf... It is an AI Agent Skill for Claude Code / OpenClaw, with 632 downloads so far.

How do I install CI Whisperer?

Run "/install martok9803-ci-whisperer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is CI Whisperer free?

Yes, CI Whisperer is completely free (open-source). You can download, install and use it at no cost.

Which platforms does CI Whisperer support?

CI Whisperer is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created CI Whisperer?

It is built and maintained by MartinKirov (@martok9803); the current version is v0.2.0.

💬 Comments