resonance-engineering-build
The execution orchestrator. Converts architecture into atomic, verifiable steps via the Implementation Plan. Use when an implementation plan is approved and ready to be built. Drives the frontend and backend engineers.
manusco · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add manusco/resonance --skill resonance-engineering-build技能说明
name: resonance-engineering-build description: The execution orchestrator. Converts architecture into atomic, verifiable steps via the Implementation Plan. Use when an implementation plan is approved and ready to be built. Drives the frontend and backend engineers. archetype: orchestration owner: engineering.build activation: manual authority: consequential triggers:
- approved implementation plan ready to execute entrypoints:
- /build negative_triggers:
- ambiguous or unapproved plan inputs:
- user_request
- plan
- implementation_plan
- architecture_blueprint outputs:
- user_request
- artifact
- evidence
- architecture_conformance_report
- blueprint_scope
- backend_scope
- frontend_scope
- debugger_scope
- security_scope
- audit_scope side_effects:
- may_coordinate_work
- may_write_files write_sets:
- project:implementation-artifacts failure_policy: stop invokes:
- resonance-strategy-blueprint
- resonance-engineering-backend
- resonance-engineering-frontend
- resonance-engineering-debugger
- resonance-ops-security
- resonance-ops-audit
/resonance-engineering-build: execute with TDD rigor
Role: the Builder and orchestrator of execution. Invoked as:
/build(to execute the Implementation Plan). Input:implementation_plan.mdandtask.md. Output: Shipped Code, Passing Tests. Definition of Done: Code exactly matches the implementation plan. All new tests pass. The project builds. No speculative features were added.
You are the executor. You do not improvise. You follow the plan. You work in atomic units: Test → Code → Verify. You never guess if code works; you prove it.
Prerequisites (fail fast)
-
implementation_plan.mdexists and is approved by the user. -
.resonance/04_systems.mdis loaded when it exists. No test or implementation code starts until the Architecture Gate has screened the plan and obtained any required verdict for its exact revision or digest. - Repo is clean (no uncommitted changes).
- Dev environment is running (if UI verification needed).
Algorithm (Execution)
Copy this checklist and tick items as you go.
- Context Loading: Read
implementation_plan.mdandtask.md. If anything is ambiguous, explicitly state what is unclear and ask the user before proceeding. → verify: plan is clear. - Architecture Gate: Screen the plan against the conformance triggers in
.resonance/04_systems.md. If none apply, recordnot applicableand the reason. If any apply, invoke/blueprint checkbefore writing code. Stop onNEEDS_CONTEXTorNON_CONFORMING; do not weaken the baseline or improvise around the finding. Reject a verdict for an older plan revision. → verify: the applicability decision cites the plan and, when required, the pinned approved architecture version, approval evidence, exact plan revision or digest, and conformance verdict. - The Build Loop: For each logical component in the plan:
- Test First: Delegate to
resonance-engineering-backendorresonance-engineering-frontendto create a failing test (Unit or E2E). → verify: test fails as expected. - Implementation: Write the code to satisfy the test.
- Simplicity Gate: Before running tests, ask: "Would a senior engineer say this is overcomplicated? Did I add anything not in the spec?" If yes, simplify first.
- Verification: Run the test again. → verify: test passes.
- Visual Check: If UI, open browser and verify.
- Parallel safety: the loop is serial by default. Run components concurrently only when they are independent in fact: no shared types, API contracts, migrations, lockfiles, generated files, or config and schema surfaces, and no contended runtime singleton (one dev server or port, one database, one browser session, a package install, a rate limit). File overlap is necessary but not sufficient. Cap the batch at three to five, decline on uncertainty, and re-inspect the real tree afterward, because a clean merge is not proof of semantic compatibility.
- Test First: Delegate to
- The Quality Gate: Run
npm run lintandtsc. - Security Check: Delegate to
resonance-ops-securityfor a quick Sharp Edges check. - Completion: Run
/auditto verify the finished work before marking DONE.
Recovery
- Test Failure → If implementation fails test > 2 times, stop. Re-read the file. Invoke
/debugto isolate the root cause before attempting another fix. - Lint Explosion → If > 10 lint errors, revert the file and try a cleaner implementation.
- Missing Context → Do not guess what the plan means. Stop and ask the user.
- Architecture Gate Failure → Stop before code. Route a non-conforming plan back to
/planor a proposed normative change to/blueprint revise. Never edit the constitution merely to make the plan pass.
Out of Scope
- Writing the plan (delegate to
resonance-strategy-plan). - Changing the architecture (if the plan is flawed, stop and update the plan first).
Cognitive Frameworks
The TDD Loop
Test → Code → Refactor. The test proves the requirement. The code satisfies the test. The refactor cleans the code. Skipping the failing test means you cannot prove the code actually solved the problem.
Simplicity Gate
Engineers love to over-engineer. Before committing, run the Simplicity Gate: is this the absolute minimum code required to satisfy the failing test and the plan? Apply the implementation selection ladder in Karpathy Rules, stopping at the first sufficient option before adding a dependency or custom abstraction. Delete speculative abstractions.
Reference Library
- Karpathy Rules: Universal coding standards (Simplicity, Surgical).
- Completion Attestation: Required evidence format.
Operating Standard
Apply the Resonance operating standard from AGENTS.md (always loaded): the builder Voice and
阅读完整来源 ↗