{"slug": "capstone-compose-your-skill-library", "title": "Capstone: Compose Your Skill Library", "summary": "Capstone: Compose Your Skill Library, a course module from Agentic AI, instructs learners to combine three of eight previously built skill packs into a single higher-order agent, such as an infrastructure change reviewer, code quality gate, or release readiness check. The module emphasizes testing the composed capability, not just individual skills, and publishing a complete, versioned, multi-pack skill library. It provides routing rules to determine which skills trigger based on request content and advises presenting combined findings in clearly labeled sections.", "body_md": "· Agentic AI · 5 min read\n\n### 📋 Prerequisites\n\n- All eight previous skill packs in this course\n\n### 🎯 What You'll Learn\n\n- Compose three independently-built skills from this course into one higher-order agent\n- Test a composed capability, not just its individual skills\n- Publish a complete, versioned, multi-pack skill library\n\n## Eight Packs, Sixteen Skills — Now Combine Some\n\nYou’ve built eight complete skill packs across this course, each independently useful. This capstone does what [Skill Composition](/courses/production-agent-skills-engineering/skill-composition) taught in the companion course: combine several of them into one capability that’s bigger than any single pack, while keeping every piece independently testable and reusable — composition, not a God Skill wearing a library’s worth of features.\n\n## Pick a Composition\n\nThree natural combinations from the packs you’ve built, though your own is welcome if a different one fits your actual work better:\n\n**Infrastructure change reviewer** — `terraform-plan-review`\n\n+ `k8s-manifest-review`\n\n+ `aws-iam-least-privilege-check`\n\n. A single request like “review this infra change before I merge it” routes to whichever of the three applies based on what’s actually in the change — a Terraform plan, a Kubernetes manifest, an IAM policy, or some combination.\n\n**Code quality gate** — `python-convention-check`\n\n+ `sql-query-review`\n\n+ `github-pr-description`\n\n. A pull request gets its Python reviewed, any SQL in the diff reviewed separately, and a structured description drafted — three skills, each doing one job, combining into “review this PR” as a single request.\n\n**Release readiness check** — `python-test-scaffold`\n\n+ `github-pr-description`\n\n+ `terraform-plan-review`\n\n. Before a release: confirm test coverage exists for new code, generate a clear PR description of what’s shipping, and check any infrastructure changes bundled with the release.\n\n## Build the Composition\n\n**Confirm each skill’s output contract still holds independently.** Per[Skill Composition](/courses/production-agent-skills-engineering/skill-composition), composition only works cleanly when each piece’s output is something the next stage — or, here, the person reading the combined result — can actually consume without re-interpreting it from scratch. Re-check each chosen skill’s output format before combining.**Decide what triggers which skill.** This is a Decision pattern applied one level up — not inside any single skill, but in how a request routes to one, two, or all three of your chosen skills. Write this down explicitly, the same way`github-issue-triage`\n\n’s rules were explicit rather than left to be inferred:\n\n```\n## Routing\n\n- Request mentions a Terraform plan or `.tf` file → terraform-plan-review\n- Request mentions a Kubernetes manifest or `.yaml` deployment spec →\n  k8s-manifest-review\n- Request mentions an IAM policy or permissions → aws-iam-least-privilege-check\n- Request is a general \"review this change\" with more than one of the\n  above present → run all that apply, and present findings grouped by\n  which skill produced them\n```\n\n**Combine the output.** Decide how findings from multiple skills get presented together — a single response with clearly labeled sections per skill is usually clearer than an interleaved summary that obscures which finding came from which review.\n\n## Test the Composition, Not Just the Pieces\n\nTesting each individual skill (which you already did, pack by pack) doesn’t guarantee the *composition* works — that’s a distinct claim, and it’s the one this capstone is actually about proving. Specifically test:\n\n**A request that should trigger only one skill**— confirm the others correctly stay silent, the same should-not-trigger discipline from the free course’s[Testing Skills](/courses/agent-skills-mastery/testing-skills)lesson, now applied across a set of related skills instead of one skill against unrelated prompts.**A request that should trigger more than one**— confirm findings from each are clearly attributed, not blended into an ambiguous combined summary.** A request that’s a near-miss for the whole set**— something that sounds infrastructure-related but isn’t actually a Terraform plan, Kubernetes manifest, or IAM policy — confirm nothing fires incorrectly.\n\n## Publish the Whole Library\n\nApply the lifecycle and distribution practices from across this path to the library as a whole, not just each skill individually:\n\n**Version the library**, per[Skill Lifecycle and Versioning](/courses/production-agent-skills-engineering/skill-lifecycle-versioning)— a single version for the collection, separate from (but referencing) each pack’s own version.**Write a root-level README**— per the free course’s[Publishing Skills](/courses/agent-skills-mastery/publishing-skills)— listing all eight packs, which ones you composed and how, and an example prompt that reliably triggers each.**Commit it somewhere durable**— a personal skills folder you’ll keep using, or a project/organization repository if any of these packs are genuinely useful to people beyond you.**If you’re ready for it**, take the CI pipeline shape from[CI/CD for Skills](/courses/production-agent-skills-engineering/skill-cicd)and apply it to this library — lint and validate every pack, run each one’s trigger tests, and gate publishing on all of them passing.\n\n## Deliverable Checklist\n\n- At least three skills from this course composed into one working capability\n- Explicit, written routing logic — not left implicit\n- Tested for single-skill triggers, multi-skill triggers, and near-misses across the composed set\n- The full eight-pack library versioned and documented with a root README\n- Committed somewhere you’ll actually keep using it\n\n## Where This Leaves You\n\nYou’ve now completed the full **Agent Skills Engineering Path**: the format (** Agent Skills Mastery**), the discipline (** Production Agent Skills Engineering**), and the practice (** this course**) — sixteen individual skills, composed capabilities, and a real, working library you can keep extending. That’s not a small thing to have built. The next skill you need for your own work is a considerably smaller lift than the first one was, because you now have a whole library’s worth of working examples, a full pattern catalog, a quality checklist, and a tested process to build it with.", "url": "https://wpnews.pro/news/capstone-compose-your-skill-library", "canonical_source": "https://superml.org/tutorials/skill-library-capstone", "published_at": "2026-07-26 00:00:00+00:00", "updated_at": "2026-08-01 05:00:32.241184+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools"], "entities": ["Agentic AI", "Terraform", "Kubernetes", "AWS IAM", "Python", "SQL", "GitHub"], "alternates": {"html": "https://wpnews.pro/news/capstone-compose-your-skill-library", "markdown": "https://wpnews.pro/news/capstone-compose-your-skill-library.md", "text": "https://wpnews.pro/news/capstone-compose-your-skill-library.txt", "jsonld": "https://wpnews.pro/news/capstone-compose-your-skill-library.jsonld"}}