manage-ci

Governs inspection, editing, and validation of MeshLLM GitHub Actions CI workflows.

3.4k|405|Updated Feb 11, 2026
One-click install
npx skills add https://github.com/Mesh-LLM/mesh-llm --skill manage-ci
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: manage-ci
Source: https://github.com/Mesh-LLM/mesh-llm/tree/main/.agents/skills/manage-ci
Command: npx skills add https://github.com/Mesh-LLM/mesh-llm --skill manage-ci

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

MeshLLM's CI spans five PR entrypoints, five main entrypoints, reusable lanes, runner providers, caches, and release pipelines, and an uninformed edit can silently break required checks, trust boundaries, or artifact contracts. This Skill enforces the normative CI rules so every workflow change stays consistent, secure, and reviewable.

Core Features & Use Cases

  • Normative CI rule source: Defines the architecture contract for entrypoints, reusable slices, planning/routing, runners, caches, artifacts, and permissions that every CI change must follow.
  • Required change procedure: Prescribes a six-step workflow from inspecting workflows and classifying the change to validating contracts and reporting external state changes separately.
  • Trust and safety guardrails: Enforces pull_request over pull_request_target, least-privilege permissions, secret handling rules, and the bounded Depot PR cache-risk exception.
  • Use Case: When asked to add a new macOS smoke job, the Skill directs you to read the inventory and topology docs, classify the change as a reusable slice, add it to the typed lane graph, and run just ci-validate before reporting.

Quick Start

Use the manage-ci skill to review the failing PR Linux workflow and propose the smallest conforming fix.

Frequently Asked Questions about manage-ci

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I safely change a GitHub Actions workflow in MeshLLM?

Read the manage-ci skill, the current inventory, and ci/ci.md first, then classify the change, identify trust context and consumers, and make the smallest coherent edit. Validate with just ci-validate plus any applicable shellcheck, crate-list, or release checks before reporting.

What CI workflows validate pull requests in MeshLLM?

PR validation uses exactly five entrypoints: pr_quality.yml, pr_website.yml, pr_linux.yml, pr_macos.yml, and pr_windows.yml. Each computes the canonical plan and calls only its matching protected default-branch reusable lane, producing one stable PR result per lane.

Can PR workflows use pull_request_target or self-hosted runners?

No. PR content must be built with pull_request, never pull_request_target, and untrusted code must never run on persistent self-hosted runners. Only metadata-only cleanup workflows may use pull_request_target without executing PR-controlled content.

How are caches handled between PR and main CI runs?

GitHub-hosted PR jobs use restore-only or isolated cache modes, while trusted main and warmer jobs own shared publication. Cache keys must include every compatibility boundary, and restored contents are treated as untrusted and verified before reuse.

Why did my reusable workflow fail with a startup_failure and no jobs?

A called reusable workflow cannot request permission scopes its caller does not grant; GitHub rejects the run at creation with a zero-job startup_failure. Check that every caller grants scopes like packages: read needed by the callee.

When should CI changes update the inventory and topology docs?

Update references/current-inventory.md whenever checked-in workflow, runner, variable, or secret-name facts change, and update ci/ci.md when topology or artifact flow changes. Normative rule changes require updating the SKILL.md contract itself first.