create-product-verification
Create a project-local verification capability for a real product UI when explicit acceptance needs repeatable browser evidence.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- create-product-verification
- description
- Create a project-local verification capability for a real product UI when explicit acceptance needs repeatable browser evidence.
02 资源文件 3
- 参考文档references/execution-plan.schema.json2.1 KB
- 参考文档references/execution-record.md749 B
- 参考文档references/feature-map.md779 B
03 SKILL.md 正文
For visual work, finish Impeccable, obtain independent Astra visual critique,
resolve its findings, then record final delivery evidence from that candidate.
Working screenshots may support critique. Astra retains live Computer Use;
Luna may prepare bounded scripts or media files, never inherit browser control
merely to reduce model cost. See design-quality and evidence-capture.
For recorded UI evidence, use evidence-capture: one session driver, stable
host-tool batches and compact receipts. Luna may prepare scripts and media;
Astra keeps acceptance and visual judgment. Existing valid verification evidence
is reusable while candidate, state and environment match. Honor explicitly
requested exhaustive coverage; ordinary changes use a representative flow plus
distinct risk states.
Create Product Verification
Use this skill only when the operator explicitly requests a product verification capability, asks for real browser acceptance, or is preparing a repository that does not yet have one. It is not a default ceremony for ordinary implementation.
Inspect first
Read the repository instructions and identify the product's current truth. Map the feature surface, how it runs, how an authorized user drives it, how state and side effects can be observed, and how the environment is isolated. Reuse existing scripts, fixtures, test accounts, preview commands, and artifact conventions. Do not invent a second harness when the repository already has a reliable one.
Generate a local capability
Create the repository's canonical verify-<product> skill (or update the
existing equivalent) with a small current-truth Feature Map. Each mapped
feature must name its launch route, role/fixture, discriminating behavior,
deterministic probes, cleanup, and evidence location. Keep private evidence
out of source control and exclude credentials, tokens, PHI, and customer data.
Keep the execution plan and acceptance rubric separate:
execution-plan.jsoncontains only neutral typed steps, exact origin/path and action allowlists, reference-only inputs, side-effect mode, and the host-owned evidence path. Itssha256binds canonical JSON in schema order with normalized step keys and thesha256field omitted. Every navigation target must satisfy the declared origin and path policy. Inputs are opaquefixture.*,host.*,session.*, orvault.*references resolved only by the authorized host; never place values in the plan.acceptance-rubric.jsonremains private to the orchestrator/judge. Never materialize it in the repository, shared workspace, runner prompt, execution plan, or evidence directory.
The generated capability must support before/after side-effect probes and a single mapped end-to-end proof before handoff. A proof is evidence, not a claim: record exact route, identity, viewport, steps, screenshots/video, probe results, runtime errors, and cleanup.
The pure planner never authorizes authorized-writes: data supplied in its
own JSON cannot prove authority. A trusted host must consume and verify an
opaque, unexpired receipt before dispatching any write-mode run. Until the
host exposes that capability, write-mode plans stop without a runner lane.
Computer Use boundary
The Astra computer-use runner executes authorized UI steps and returns a neutral execution record. It may report incomplete execution, observations, screenshots, video, and runtime errors, but it must not decide whether the product passed, failed, was blocked, or was inconclusive. The Astra orchestrator retains the rubric, compares evidence with before/after probes, and owns that semantic judgment.
Do not grant browser authority from this skill or from the pure planner. Use the existing authorized browser/profile only when the operator has authorized it. Stop and report missing credentials, destructive actions, or ambiguous environment routing.
See references/feature-map.md, references/execution-plan.schema.json, and
references/execution-record.md for the minimum portable shapes. Use the
global skill-creator skill for general skill authoring and validation; do not
duplicate its guidance here.
04 同仓库更多 Skills
working-backwards
Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.
flow-implement
Implement a requested software change through verification, corrections, and its authorized delivery endpoint.
development-steward
Run the scheduled, read-only weekly maintenance review for the six explicitly opted-in repositories and prepare one concise private report for Check-in.
grill-with-docs
Run one recoverable interview for functional, visual, or mixed software grills; reuse equivalent answers and offer the shared HTML questionnaire.
global-agent-guardrails
Install, audit, test, or update the Development System's global destructive-command guard for Codex, Claude Code and T3 Code. Use when configuring agent safety hooks or investigating whether dangerous shell operations are blocked.
impeccable
Design or refine frontend UI using project context and a focused design playbook; also handles visual critique and polish.