persona-cast
One-line doorway that lets a field harness review an artifact (a screen, a spec, a PR, a page) through a pre-built cast of review-shaped personas — web-review cast by default (designer, ux-writer, fe-dev, be-dev, qa, a11y) — dispatched in isolation, merged with a marginal-coverage stop so a lens that finds nothing unique is reported as decorative, not padded. Triggers: "이 화면 리뷰해줘", "페르소나로 봐줘", "디자이너·QA 눈으로 봐줘", "review this screen with the cast", "persona cast", "persona-cast".
chrono-meta · 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 chrono-meta/forge-harness --skill persona-cast技能说明
name: persona-cast description: 'One-line doorway that lets a field harness review an artifact (a screen, a spec, a PR, a page) through a pre-built cast of review-shaped personas — web-review cast by default (designer, ux-writer, fe-dev, be-dev, qa, a11y) — dispatched in isolation, merged with a marginal-coverage stop so a lens that finds nothing unique is reported as decorative, not padded. Triggers: "이 화면 리뷰해줘", "페르소나로 봐줘", "디자이너·QA 눈으로 봐줘", "review this screen with the cast", "persona cast", "persona-cast".' version: 0.1
persona-cast — a pre-built review cast a field harness can use without knowing the container
Why this exists (operator decision 2026-09-12): FH can synthesize personas on demand through
persona_container_schema.md, but a field harness cannot — it would have to fill seven slots, pick a
tiering rule and run a marginal-coverage stop every time. What a field harness can use is the
container's output, pre-baked: a cast of review-shaped lenses plus one doorway. This skill is that
doorway, the same shape as harness-pr-reviewer (a standalone way in to one FH capability).
Role split:
sim-conductorsynthesizes personas for a simulation and owns the parallax shape.persona-roster-expanderturns a named seed into a tiered cast.persona-castdispatches an already-built cast at an artifact. It builds nothing; a missing lens is a request to the two skills above, never inline improvisation.
Casts (shipped in plugins/fh-commons/agents/*-lens.md)
| cast | lenses | for |
|---|---|---|
web-review (default) | designer-lens · ux-writer-lens · fe-dev-lens · be-dev-lens · qa-lens · a11y-lens | a screen, a flow, a web/mobile spec, a page (review mode) or a live app (explore mode) |
A cast is a list, not a promise: every run measures which lenses earned their seat (Step 4).
Steps
Step 0.5 — Trigger probe (routing skill obligation · measured 2026-09-13)
This skill routes — it decides which agents run — so the FH New-Skill bar owes it a trigger probe: does a floor-tier session that was never told about this skill actually reach it from the trigger phrases in the frontmatter? Reading the description is not evidence; a reader who wrote it is the one reader guaranteed to route correctly.
Measured — blind, floor tier (sonnet), isolated install (this skill + the six lenses only), reps=3 per arm:
| arm | utterance | persona-cast reached |
|---|---|---|
| positive | screen.html 이 화면 리뷰해줘 | 3/3 (seat names appeared 13–20× per run — the cast actually ran) |
| negative (control) | README.md 의 오타를 고쳐줘 | 0/3 |
🟥 The first probe fixture was wrong and measured nothing. It ran the trigger with no artifact in the directory, so the session correctly asked which screen — and whether it happened to name the skill inside that question is incidental, not routing. It scored 1/3 and that number is void. Give the probe a real artifact, or you are measuring the question, not the route.
Re-probe whenever the frontmatter trigger phrases change.
Step 1 — Take the artifact and freeze it
Input = one or more paths (screenshot text dump, HTML→text capture, spec markdown, PR diff, page
snapshot). Record sha256 of each file in the run header. A lens reads files; it never receives the
operator's summary of the files (that is the design-aware contamination the container's isolation
invariant forbids).
🟥 The freeze is lossless. Hand the lens the raw capture (the HTML, the full screenshot, the unabridged spec) and, only in addition, any extracted view. Measured on the first run (2026-09-12): an element extract that truncated attributes at 120 chars produced three «unclosed href» Criticals from the qa lens — the page was fine, the instrument was broken, and the lens had read the artifact faithfully. An extracted view is a convenience, never the artifact.
Step 2 — Pick the cast (mechanical)
--cast web-review (default). --lenses a,b,c narrows. --stance friendly|neutral|hostile sets the
delivery layer for all lenses (recorded explicitly — an omitted stance is written as neutral).
Step 3 — Dispatch each lens in isolation
One Agent per lens, subagent_type = the lens file's name (fh-commons:designer-lens …), prompt =
artifact paths + the stance + nothing else. Lenses run in parallel; none sees another's output.
Each returns the parallax block including its Unique-to-this-seat self-tag.
Step 4 — Merge with the marginal-coverage stop (anti-decorative)
First drop every Concern that carries no [location] — the lens files promise this («A concern
without a location is dropped by the dispatcher»), and this step is where it happens. Count what you
drop per lens and print it (no-loc column below); a dropped concern is never silently lost, and a
lens whose concerns are mostly no-loc read the artifact loosely. From here on total counts only
the concerns kept after this drop. Then normalize the rest by
[location] + first noun phrase. For each lens compute
unique = concerns raised by this lens and no other. Report per lens: total / unique / self-tagged unique that were actually unique. A lens with unique = 0 is reported as decorative on this artifact — not deleted from the cast, but named, so the next run (or the cast maintainer) can drop
or merge it. Self-tag accuracy is the lens's own calibration score.
Step 5 — Output
persona-cast · cast=web-review · stance=neutral · artifact=<path> sha256=<12>
lens total no-loc unique self-tag-hit verdict
designer 4 0 2 2/3 earned
ux-writer 3 1 3 3/3 earned
a11y 2 0 0 0/1 decorative on this artifact
...
Critical (merged, deduped, with the seats that raised each) …
Important …
Open questions (routed to the seat that may answer) …
Explore mode (--mode explore) — 살아 있는 앱에서 탐색적 테스팅을 페르소나의 눈으로
Review mod
阅读完整来源 ↗