Verification
Prove the happy path
- Can read/write a GitHub Actions workflow
- Can explain trigger trust boundaries
- Can enforce deterministic install
- Can build once/promote same artifact
Phase 33 · Lesson 16
Mastery Capstone Lab
Learning contract
What you should be able to do after this
- Explain Mastery Capstone Lab without relying on framework buzzwords.
- Connect the lesson to the phase outcome: Build a trustworthy GitHub Actions pipeline that verifies pull requests, creates immutable release artifacts, uses least privilege, caches safely, and controls deployment concurrency.
- Reproduce the relevant behavior and verify it with evidence.
- Diagnose at least one representative failure instead of guessing.
Figure 33.16 — TaskFlow boundary map
pull request
deterministic install
lint/typecheck/test/build
trusted main
immutable artifact
environment gate
deployment
Instructor workflow: predict first → implement the smallest slice → run it → inspect evidence → break one assumption → diagnose → explain it back. Do not copy the whole code block before understanding the first failing boundary.
Mastery Capstone Lab
This is the transfer exercise for Phase 33: CI/CD with GitHub Actions. The detailed lessons are now closed-book reference material. Your task is to prove that the phase outcome can be reproduced from requirements, repository evidence, and debugging—not by copying a finished implementation.
Target artifact .github/workflows/ci.yml plus release/deploy workflows that build once, tag by commit SHA, and gate promotion.
Acceptance criteria - Write the behavior and constraints before implementation. - Identify the owning layer and every boundary crossed. - Implement the smallest complete vertical slice. - Verify the happy path from outside the implementation. - Demonstrate at least one expected failure path. - Run the phase-relevant quality checks. - Review the diff for accidental duplication, leaked responsibilities, missing validation/authorization, and hidden environment assumptions.
Evidence packet Save the command/request used, expected result, actual result, one failure reproduction, the fix, and a short explanation of why the final responsibility belongs where it does. This packet is more valuable than a screenshot of a green page because it demonstrates repeatable engineering reasoning.
Stretch Change one requirement after the first implementation. Measure which files must change. If a tiny domain change causes unrelated layers to move, use that evidence to discuss coupling before adding another abstraction.
Executable lesson studio · ci
Now turn the explanation into engineering skill.
A guided lesson proves you can follow. A capstone proves you can transfer the model to a requirement with fewer cues.
By the end, you should be able to
- Build a trustworthy GitHub Actions pipeline that verifies pull requests, creates immutable release artifacts, uses least privilege, caches safely, and controls deployment concurrency.
- Produce a reproducible evidence packet for success and failure.
- Defend the architecture and identify one improvement without overengineering.
Figure 33.16B — Mental model
pull request
deterministic install
lint/typecheck/test/build
trusted main
immutable artifact
environment gate
deployment
Code-along sequence
One coherent change at a time
- 01
Create CI triggered by pull_request with read-only contents permission and explicit concurrency cancellation.
- 02
Set up Node and pnpm, then install with --frozen-lockfile so manifests cannot drift from the lockfile.
- 03
Run lint/typecheck/tests/build as separate clear steps or jobs with understandable failure ownership.
- 04
Add dependency/package cache only after correctness works; key it from lockfile inputs.
- 05
Create trusted-main release workflow that tags an image/artifact with the commit SHA rather than latest only.
- 06
Add environment protection/permissions for deployment and ensure stale production runs are not silently canceled mid-release.
- 07
Run the phase mastery checklist without looking at the provided implementation.
- 08
Perform a short architecture review and document one trade-off you consciously accepted.
txtCapstone evidence packet
Requirement -> implementation diff -> verification -> controlled failure -> diagnosis -> final proofDo not copy this as a finished answer. Predict what each boundary does, type the smallest useful slice, run it, then compare the result with your prediction.
Break it on purpose
Build debugging memory
- Using pull_request_target with untrusted checkout/secrets
- Broad write permissions for all jobs
SymptomInvestigation path
You can only continue by copying the reference
Return to the mental model and acceptance criteria. Rebuild one smaller slice from first principles before reopening the source.
Everything seems to work but proof is weak
Choose an external observer: HTTP client, browser, database query, test runner, health endpoint, or deployment log, and capture deterministic evidence.
Closed-reference challenge
Can you transfer the idea?
Extend the phase artifact with one requirement not used in the guided example while preserving the same ownership rules and verification strategy.
Explain-back
Answer these without reopening the solution
- What did you build without a transcript?
- Which boundary caused the most reasoning effort?
- Which failure gave the strongest evidence?
- What shortcut did you deliberately avoid?
- What would you automate before repeating this in a team?
Lesson studio
Turn the reading into skill
Do it
Add dependency/package cache only after correctness works; key it from lockfile inputs.
Use the smallest change that demonstrates this lesson. Stop after it works and capture evidence.Break it safely
CI green because important workspace checks never ran
Predict the symptom before causing the failure. Then locate the earliest boundary where reality diverges from expectation.Explain it back
Why does this responsibility live here, what input does it trust, what output proves success, and what would go wrong if another layer owned it?
If your explanation is vague, repeat the boundary trace before moving on.Exit ticket
Before marking complete
Close the reference. Reproduce the core idea from memory or explain the exact request/data flow aloud. Then check your diff or notes for accidental shortcuts that bypass validation, authorization, error handling, or verification earned in earlier phases.