菜单
Evaluate the IP constellation — dispatches baseline checks, then 9 lenses in parallel (Millman, Bierut, Appleton, Shaw, Peers, Victore, The Walk, The Walls). Coordinator skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Context-aware story extraction. Loads the maker's full context — identity, expertise, voice, published work — then reads the narrative architecture to understand the story being told and what's missing. Interviews the maker to fill gaps with real moments. Modular: works with any project.
Cross-page coherence. Do the three tiers form a coherent argument when read as a sequence? Read-only diagnostic.
Baseline checks — mechanical health of the constellation. Links, images, cross-refs, attribution, voice violations, SEO, tooltips. Read-only diagnostic.
Where lenses agree and disagree. Maps consensus vs. contradiction across audit results. Read-only diagnostic.
Copy verification checklist — 13-item pass/fail check on draft or published copy. Voice rules, grounding, Shaw check, steward voice. Read-only diagnostic.
Would someone send this? Per-page forwardability — is the work remarkable enough to share? Read-only diagnostic.
| name | audit |
| description | Evaluate the IP constellation — dispatches baseline checks, then 9 lenses in parallel (Millman, Bierut, Appleton, Shaw, Peers, Victore, The Walk, The Walls). Coordinator skill. |
| user_invocable | true |
Evaluate Peter's IP constellation (hub site + spoke repos) in layers: visual inspection, mechanical baseline checks, then nine lenses dispatched in parallel. Reports only — never auto-fixes.
/audit run # Full audit: visual + baseline + all 9 lenses
/audit run baseline # Baseline checks only (dispatches /baseline)
/audit run [lens] # Single lens (millman, bierut, appleton, shaw, peers, victore, the-walk, the-walls)
/audit check [page-or-repo] # All checks scoped to one target
/audit rubric # Show/create rubric file
/audit report # Show last saved report
_governance/, _infrastructure/, _output/ WITHOUT published: false_blog/ WITHOUT published: falsebio/index.md, contact.md, colophon.md, thinking.md, vocabulary.md_data/index.json + _layouts/systemworks.html~/homelab/projects/active/ for repos containing a CNAME file_data/navigation.jsonNORTHSTAR.md — master vision ("the site is a world, not a résumé"). The audit evaluates against this intent.docs/ARCHITECTURE.md — intended IA structure (three tiers, user journeys). Baseline checks mechanical health against this.docs/aetherwright-identity-decisions.md — persona layer (who Peter is).audit/screenshots/[today's date]/ — if full-pass passed a SCREENSHOT_DIR, use that pathdocs/visual-reference-index.md → active reference set.audit/rubric.md if it exists/baseline → mechanical health → pass/fail per check/millman → personhood → STRONG/HOLDS/WEAK/BROKEN + direction/bierut → design thinking → STRONG/HOLDS/WEAK/BROKEN + direction/appleton → living knowledge → STRONG/HOLDS/WEAK/BROKEN + direction/shaw → world feel → STRONG/HOLDS/WEAK/BROKEN + direction/peers → peer benchmark → STRONG/HOLDS/WEAK/BROKEN + direction/victore → bravery → Brave/Safe/Performed/Sanitized + direction/the-walk → example positioning → ONE WORLD/HOLDS/SPLIT/WRONG ROOM + direction/the-walls → ambient frame → COHERENT/HOLDS/SPLIT/NOISE + direction/copy-verify → voice spot-check → pass/fail per item/knowledge → ideation grounding → source-attributed search resultsM1-M5: Real person, vulnerability, voice, life-arc, courage B1-B5: Form/content, problem-solving, meaning, content-driven, intellectual A1-A5: Connections, growth, epistemic disclosure, contextual, evolving S1-S5: World immersion, total environment, stranger orientation PB1-PB5: Structure, visual evidence, public output, social proof, distribution Victore: Fierce self vs performing W1-W5: Room per page, sidebar entry, secondary identity leading, governance examples, one world F1-F5: Persistent frame identity, marginalia signal, frame/content agreement, stranger/initiate gradient, ambient density distribution
/audit runVisual inspection is required. The audit cannot evaluate design lenses by reading markdown source alone.
0.0 Start Dev Server
localhost:4000 is reachablebundle exec jekyll serve --host 0.0.0.0 &0.1 Screenshots
Take screenshots of every published page at 3 viewports (375px, 768px, 1440px) using Playwright or equivalent. Save to .audit/screenshots/YYYY-MM-DD/. Capture all pages, not a sample.
0.2 Visual Assessment Read every screenshot as an image. Evaluate: typography hierarchy, whitespace, visual weight, layout coherence across viewports, visual-content alignment. Note ALL user-facing elements (images, codex strings, glyphs, altitude markers, faculty dots). Compare against previous screenshots if available.
0.3 First Visitor Walkthrough
Dispatch /baseline atomic skill. This runs all mechanical checks:
Baseline results inform lens evaluations (e.g., Appleton A5 needs tooltip results, Bierut B4 needs vocabulary visibility).
Use /knowledge skill to verify copy grounding:
After baseline completes, dispatch all 9 lenses simultaneously using the Agent tool:
/millman — Is this person real? (M1-M5)/bierut — Does the design think? (B1-B5)/appleton — Is the knowledge alive? (A1-A5)/shaw — Does this feel like a world? (S1-S5)/peers — Peer benchmark (PB1-PB5)/victore — Is Peter being fiercely himself?/the-walk — Front door to tattoo chair (W1-W5)/the-walls — What room does the frame build? (F1-F5)/copy-verify — Spot-check voice on a sample of published pages (optional)Each lens reads the same published material and the rubric independently. Each returns its own verdict: STRONG / HOLDS / WEAK / BROKEN (except Victore which returns Brave / Safe / Performed / Sanitized).
Screenshot forwarding: If a SCREENSHOT_DIR was provided (from full-pass), include screenshot paths in the /shaw and /the-walls agent prompts. Shaw evaluates content and visual world-feel. The Walls evaluates whether ambient frame elements (sidebar, glyphs, codex, typography) build the right room — screenshots are essential for this. Example paths to include: SCREENSHOT_DIR/laptop/index.png, SCREENSHOT_DIR/laptop/encore.png, SCREENSHOT_DIR/mobile/index.png.
After all lenses return:
.audit/reports/YYYY-MM-DD.md# Constellation Audit — [DATE]
## Baseline
[NUMBER] checks passed. [NUMBER] failed.
[List each failure with file path and specific issue]
## Millman — [STRONG/HOLDS/WEAK/BROKEN]
[Key findings from /millman — M1-M5 verdicts with evidence]
## Bierut — [STRONG/HOLDS/WEAK/BROKEN]
[Key findings from /bierut — B1-B5]
## Appleton — [STRONG/HOLDS/WEAK/BROKEN]
[Key findings from /appleton — A1-A5]
## Shaw — [STRONG/HOLDS/WEAK/BROKEN]
[Key findings from /shaw — S1-S5]
## Peer Benchmark — [STRONG/HOLDS/WEAK/BROKEN]
[Key findings from /peers — PB1-PB5]
## Victore — [Brave/Safe/Performed/Sanitized]
[Findings from /victore]
## The Walk — [ONE WORLD/HOLDS/SPLIT/WRONG ROOM]
[Key findings from /the-walk — W1-W5: room per page, sidebar entry, secondary identities, governance examples, one world]
## The Walls — [COHERENT/HOLDS/SPLIT/NOISE]
[Key findings from /the-walls — F1-F5: persistent frame, marginalia signal, frame/content agreement, stranger/initiate gradient, ambient density]
## Limitations
- Visual evaluation is Claude's reading as a multimodal viewer, not expert design critique
- First-visitor walkthrough is a proxy, not user testing
- Voice checking catches pattern-matchable violations only
- These are Claude's interpretations of the frameworks, not the original authors' evaluations
- Peer benchmark is based on publicly visible content, not private work or reputation
/audit run [lens]Dispatches a single lens directly. No baseline, no synthesis — just the lens output.
/audit check [target]Same checks, scoped to one page or spoke repo. Target can be:
_output/aiden-jae.mdSavepoint.Protocol/infrastructure/encore/audit rubricRead and display .audit/rubric.md. If it doesn't exist, create it with starter content. Peter's rubric annotations override default evaluation criteria.
/audit reportRead the most recent file from .audit/reports/. If no reports exist, say so.