Awesome GitHub Copilot

What It Does

awesome-copilot is GitHub's official community collection of Copilot customizations. It holds 338 agent skills (SKILL.md folders with optional bundled assets), 212 custom agent definitions (.agent.md files), 183 coding instructions (.instructions.md files applied by file pattern), 6 hooks, 8 agentic workflows, and 67 installable plugins. Resources cover languages (Clojure, Apex, ColdFusion, Rust), platforms (Azure, AWS, Shopify, Salesforce), and workflows (security review, architecture diagramming, codebase onboarding). The repo ships with a build pipeline (eng/ scripts), a marketplace.json generator, a full Astro website, and 33 CI workflows for validation and automation. It follows the agentskills.io specification for skill frontmatter.

The Good

Massive breadth with real depth at the top. The quality-playbook skill weighs in at 2,738 lines. The azure-architecture-autopilot skill bundles 13 reference files, 3 Python scripts, and PNG screenshots. The acquire-codebase-knowledge skill includes a scan.py script and 7 template files. These are not thin stubs. The top 20 skills average over 600 lines each.

Enforced frontmatter validation at PR time. The skill:validate and plugin:validate npm scripts check name format (lowercase-hyphen), description length (10-1024 chars), folder-name-to-name-field match, and bundled asset sizes (max 5MB per file). CI runs these on every PR. All 338 skills pass both name and description checks. The plugin validator checks spec paths, name uniqueness, version format, and keyword constraints.

33 CI workflows covering every angle. PR validation (skill-check, check-plugin-structure, validate-readme), publishing (publish.yml, deploy-website.yml), quality monitoring (resource-staleness-report, skill-quality-report), duplicate detection (pr-duplicate-check, duplicate-resource-detector), contributor management (contributors, contributor-check), and even external plugin intake with a two-stage approval pipeline. This is the most thorough CI setup I have seen in a skill collection.

Plugin marketplace with one-line install. The copilot plugin install @awesome-copilot command installs curated bundles. The marketplace.json is auto-generated from plugin folders. Plugins group related agents, skills, and commands around themes (azure-cloud-development, java-development, frontend-web-dev, context-engineering). This is the only skill pack reviewed so far with a working marketplace and a discoverable install path.

Multi-format resource types serve different use cases. Instructions auto-apply by file pattern (applyTo: '**.ts'). Agents are role-specific system prompts with tool lists. Skills are self-contained folders with optional scripts, references, and assets. Workflows are GitHub Actions automation written in markdown. Hooks trigger during agent sessions. Each format targets a different Copilot customization surface.

Progressive disclosure and LLM-friendly discovery. The repo provides llms.txt for agent consumption, a machine-readable AGENTS.md, and a website with full-text search. The agentskills.io spec is explicitly referenced. This repo treats LLM discoverability as a first-class concern.

The Bad

Wide quality variance between entries. The smallest skill (azure-role-selector) is 6 lines: a frontmatter block and a single paragraph. The largest (quality-playbook) is 2,738 lines. Median skill size is 151 lines, which is solid, but the long tail of thin entries dilutes the average. Some skills are just a prompt template with no scripts, no references, and no bundled assets. Others are 800-line monolithic prompt engineering documents. The collection does not surface this variance in its directory listing or website.

15 instructions missing the description field. The CONTRIBUTING.md mandates a description field for all instructions, but 15 of 183 instruction files omit it (14 are dataverse-python-* variants plus memory-bank). Seven instructions also lack the required applyTo field. One agent (declarative-agents-architect.agent.md) is missing its description. These gaps suggest the validation CI only covers skills and plugins, not agents or instructions.

Heavy Azure and Microsoft vendor weighting. A large proportion of the collection targets Azure, Bicep, Power Platform, Dataverse, and other Microsoft ecosystem tools. This makes sense given GitHub's ownership but tilts the collection away from language-agnostic or non-Microsoft developers. Arize (Phoenix/LLM observability) has 10 separate skills for different aspects of the same tool, which feels like it could be consolidated.

npm cache permissions blocked local build. Running npm ci failed on macOS with EACCES errors on the npm cache directory. This is a local environment issue, not a repo bug, but it prevented running the full validation suite. The build requires js-yaml, vfile, and vfile-matter as dependencies, which is lightweight. The failure mode is worth noting for anyone trying to contribute locally.

No quality scoring or tier system. The repo does not tag skills as "tested," "community-vetted," or "experimental." Every entry appears at the same weight in the README and website. With 338 skills, a quality tier system would help users prioritize. The skill-quality-report CI workflow exists but its output is not surfaced in the README or website.

Smoke Tests

Structural validation (host-based, macOS aarch64)

Ran structural checks directly against the shallow-cloned repo in /tmp/awesome-copilot-review.

$ git clone --depth 1 https://github.com/github/awesome-copilot /tmp/awesome-copilot-review
Cloning into '/tmp/awesome-copilot-review'...

$ ls agents/ | wc -l
 212

$ ls instructions/ | wc -l
 183

$ ls skills/ | wc -l
 338

$ ls plugins/ | wc -l
 67

$ ls hooks/ | wc -l
 6

$ ls workflows/ | wc -l
 8

$ find . -type f | wc -l
 2313

$ du -sh .
 135M .

Result: 814 total resources across 6 categories. 2,313 files. 135MB total.

Frontmatter completeness check:

$ grep -rl "^name:" skills/*/SKILL.md | wc -l
 338
✅ PASS: All 338 skills have name field

$ grep -rl "^description:" skills/*/SKILL.md | wc -l
 338
✅ PASS: All 338 skills have description field

$ grep -rl "^description:" agents/*.agent.md | wc -l
 211
❌ FAIL: 1 of 212 agents missing description (declarative-agents-architect.agent.md)

$ grep -rl "^description:" instructions/*.instructions.md | wc -l
 168
❌ FAIL: 15 of 183 instructions missing description

$ grep -rl "^applyTo:" instructions/*.instructions.md | wc -l
 176
❌ FAIL: 7 of 183 instructions missing applyTo field

Pass rate: 3 of 5 structural checks passed. The skill validation is solid. Agent and instruction validation has gaps.

Asset bundling check:

$ find skills/ -type d -name "references" | wc -l
 97
✅ PASS: 97 skills ship reference documents

$ find skills/ -type d -name "scripts" | wc -l
 21
✅ PASS: 21 skills ship executable scripts

$ find skills/ -type d -name "assets" | wc -l
 14
✅ PASS: 14 skills ship template/image assets

$ find skills/ -type f \( -name "*.py" -o -name "*.sh" -o -name "*.js" \) | wc -l
 42
✅ PASS: 42 executable files bundled across skills

Pass rate: 4 of 4 asset checks passed.

Size range check:

$ wc -l skills/*/SKILL.md | sort -rn | head -3
 2738 skills/quality-playbook/SKILL.md
 1045 skills/cosmosdb-datamodeling/SKILL.md
 916 skills/markdown-to-html/SKILL.md

$ wc -l skills/*/SKILL.md | sort -n | head -3
 6 skills/azure-role-selector/SKILL.md
 13 skills/remember-interactive-programming/SKILL.md
 14 skills/dataverse-python-quickstart/SKILL.md

$ wc -l skills/*/SKILL.md | sort -n | awk 'NR==170{print}'
 151 skills/gsap-framer-scroll-animation/SKILL.md
✅ PASS: Median skill is 151 lines. Healthy distribution.

CI pipeline check:

$ ls .github/workflows/ | wc -l
 33
✅ PASS: 33 CI workflows covering validation, publishing, quality monitoring

What the runs tell you

The structural validation confirms the repo's infrastructure is mature and well-enforced for skills (100% frontmatter compliance). Agent and instruction validation is less strict, with 16 total files missing required fields. The asset bundling numbers show that about 29% of skills (97 of 338) ship extra reference material, and 6% ship executable scripts. The collection is primarily prompt-driven rather than tool-driven, which matches its Copilot customization focus. npm ci failed locally due to cache permissions, so the build validation scripts could not be executed. The validation logic was inspected manually in eng/validate-skills.mjs and eng/validate-plugins.mjs and is thorough.

Setup Walkthrough

1. To install the meta-plugin (discovery agent): copilot plugin install awesome-copilot@awesome-copilot
2. If the marketplace is not registered: copilot plugin marketplace add github/awesome-copilot then install.
3. Browse individual resources at https://awesome-copilot.github.com with full-text search and filtering.
4. To install a specific plugin like the Azure development pack: copilot plugin install azure-cloud-development@awesome-copilot
5. To contribute a skill: fork, run npm run skill:create -- --name my-skill, edit SKILL.md, run npm run skill:validate, open a PR. CI validates frontmatter, naming, and asset sizes automatically.

No local build step needed for end users. The plugin system handles installation. Contributors need Node.js for the build pipeline.

Alternatives

1. agentskills/agentskills -- the open specification and reference library that awesome-copilot follows. Better for building cross-platform skills, less Copilot-specific.
2. obra/superpowers -- a smaller, more opinionated Hermes Agent skill collection with deep subagent coordination. Higher per-skill density, fewer total resources.
3. vercel-labs/agent-skills -- a vendor skill pack focused on Next.js and Vercel ecosystem. Narrower scope, more cohesive quality.