handoff-fusion-engineering
Prepare a traceable local Fusion engineering review draft with source identities, artifacts and unresolved assurance gates. Not for sending, publication or release approval.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-11
- Stars
- ★ 20
- Forks
- 14
01 前言 Front Matter
- name
- handoff-fusion-engineering
- description
- Prepare a traceable local Fusion engineering review draft with source identities, artifacts and unresolved assurance gates. Not for sending, publication or release approval.
02 资源文件 1
- 参考文档agents/openai.yaml221 B
03 SKILL.md 正文
Use fusion_handoff_prepare to assemble the user's review evidence, and fusion_handoff_inspect to check an existing draft against current source and stored receipt state. Read the handoff contract when composing structured checks or interpreting freshness failures.
The accepted fields are title, plan_ids, document_ids, requirements, assumptions, reviewers, checks and external_evidence. Only the title is required; missing requirements, sources and checks stay explicit. Select actual producing plans to retain artifact and pending/uncertain job evidence. Do not invent unsupported job_ids, solver-result fields or approval flags.
Discover the active operation schemas and inspect the actual response before choosing check pointers. Automated handoff checks are limited to document.inspect, parameters.list, geometry.measure, geometry.check and cam.inspect. Numeric ranges require a value pointer and an exact observed unit pointer/string; there is no implicit unit conversion. Use identity assertions alongside array positions when the measured parameter or entity matters. Do not select credential fields or infer equality from redaction placeholders; rejected or redaction-altered comparisons cannot establish a result. Fixture values and response shapes do not qualify a live Fusion check.
Link requirements and reviewer IDs explicitly. Use manual procedures for unavailable solver, simulation, drawing, ECAD or engineering review work; they remain not_run. External evidence references and reported hashes are unfetched, unverified metadata. Reviewer assignments do not authenticate reviewers, notify them or record approval.
Known freshness_gaps, malformed coverage, changed source states and missing fingerprints prevent current use of dependent checks. A linked artifact must match its exact producing plan, document, preparation/completion source and implementation contract; an old-source or legacy receipt remains historical. Keep failed, pending and unknown plan outcomes visible. Do not rewrite a stored draft, drop unresolved evidence or refresh a stale assertion into a pass. Collect new observations into a new draft when the task calls for current evidence, keeping the original for comparison.
Return the local draft reference, its evidence scope and the unresolved decisions with their practical impact. A met criterion, screenshot, generated toolpath, setup sheet or provider success is not engineering, machine-collision or regulatory proof. These tools do not send the draft, publish artifacts, update enterprise systems or advance a release. Such actions require their own destination authority and qualified adapter; preparing an already authorized local draft does not require repetitive permission requests.
04 同仓库更多 Skills
recommend-review-reference-architecture
Recommend a conservative shadow-only review reference architecture from valid offline evidence. Use only when the user explicitly asks for a reference-architecture recommendation or uses the bundled synthetic fixture.
normalize-review-runs
Validate and normalize declared local review-run exports into a bounded canonical summary. Use only when the user explicitly asks to normalize review runs or use the bundled synthetic fixture.
benchmark-review-configs
Deterministically score comparable imported review-run lanes with root-cause metrics and uncertainty. Use only when the user explicitly asks to benchmark normalized local review runs or use the bundled synthetic fixture.
audit-review-eval-validity
Audit whether a declared local review evaluation can support fair quality claims. Use only when the user explicitly asks to audit review-eval validity or use the bundled synthetic fixture.
verify-swiftui-parity
Verify one approved SwiftUI feature slice against its behavioral parity contract with deterministic and conditional Apple-platform checks.
port-react-native-slice
Port one explicitly approved React Native or Expo feature slice into a declared SwiftUI target while preserving its parity contract.