pagekit

PageKit Orchestrator Skill

You are running the full PageKit method end-to-end. The method, the canonical prompts, the scripts and the rules all live in this plugin. Use them.

Read first

1. ./references/method.md — the method at a glance: steps, artifacts, order.
2. ./references/anti-slop.md — hard no-go patterns for the draft.
3. ./references/run-logging.md — what counts as fully-logged.
4. AGENTS.md and CLAUDE.md at the repo root — agent contract and operational rules.

Procedure

For any object the user gives you:

1. Confirm scope

Ask the user (briefly):

• Run name (slug for the folder under runs/).
• Is the object real or a training fiction? If fictional, what concrete facts should serve as ground truth?
• Severity for the eventual claim-check pass: light, normal, or hard.

2. Scaffold

Run scripts/new-run.sh <run-name> (or invoke the pagekit-new-run skill). This creates runs/<run-name>/ with the full fully-logged layout.

3. Source briefs

Fill runs/<run-name>/sources/:

• 01-source-capture.md — what raw material exists, what is missing.
• 02-product-brief.md — concrete product facts. No [Product name] placeholders. For trust-heavy objects, also create 03-wedge-definition-brief.md, 04-mechanism-brief.md, 05-proof-brief.md, 06-comparison-brief.md.

4. Run steps 01-06 in order

For each step:

1. Invoke the per-step skill: pagekit-signal-doc, pagekit-message-spine, pagekit-first-page-decision, pagekit-page-argument-shape, pagekit-proof-map, pagekit-first-page-draft.
2. The per-step skill loads its canonical prompt from its own references/prompt.md, substitutes the run's inputs, prompts the model.
3. Save the prompt actually used (with substitutions) to runs/<run-name>/prompts/NN-*.md.
4. Save the raw output to runs/<run-name>/outputs/NN-*-output.md.
5. Distill the output into the canonical artifact in the run root (signal-doc.md, message-spine.md, etc.).
6. Update runs/<run-name>/working-log.md with what changed and any decisions made.

Quality gate at every step: if the artifact does not meet the quality bar in the corresponding references/framework.md for that skill, stop and fix upstream source material before advancing. Do not paper over weakness with downstream effort.

Step 03 (first-page decision) is special. Do not default to homepage. List candidates considered and rejected. The object decides the first page.

5. Pre-claim-check slop check

Run scripts/slop-check.sh runs/<run-name>/first-page-draft.md. Resolve every flagged pattern by rewriting the line. Re-run until clean.

6. Claim check (step 07)

Invoke the pagekit-claim-check skill at the chosen severity. Internally it delegates to the pagekit-claim-checker subagent (read-only, hard severity by default). Save:

• runs/<run-name>/claim-check.md — the audit (flagged lines, types, rewrites)
• runs/<run-name>/first-page-draft-corrected.md — the corrected draft, if any line was rewritten

7. Evaluation and evaluator pass

Write runs/<run-name>/evaluation.md (honest, names what stayed thin) and runs/<run-name>/evaluator-pass.md (adversarial second read; lists repo-improvement candidates if the run exposed any).

8. Validate

Run scripts/run-check.sh runs/<run-name> (or invoke pagekit-run-check). It must classify the run as FULLY LOGGED before you declare the run done. If it does not, fix the missing items.

9. Hand off

Tell the user:

• The run is at runs/<run-name>/.
• The first page chosen at step 03 (and which candidates were rejected).
• The slop-check and claim-check results.
• Any repo-improvement items the evaluator pass surfaced.

Hard rules (do not violate)

• Do not declare the run done until scripts/run-check.sh runs/<name> returns tier: FULLY LOGGED (or PUBLISHABLE). This is the only completion signal. Step 06 producing a nice-looking draft is NOT completion. If run-check reports ARTIFACT-ONLY or SUMMARY LOGGED, the run is incomplete and you must keep working.
• Do not default to homepage. Step 03 must consider alternatives.
• Do not invent proof. Use *[verification flag: ...]* for unverified product-specific claims.
• Do not skip slop-check before the claim-check step. The script catches the obvious cases; the claim-check catches the rest.
• Do not skip the evaluator pass at the fully-logged tier.
• Do not edit the canonical prompts in any skill's references/prompt.md to fit a single run. They are shared.
• Do not delete or leave blank the scaffolded placeholder files (claim-check.md, evaluation.md, evaluator-pass.md, working-log.md). They are structural; fill them in or commit them as-is until filled.

Are you done? Self-check

Before handing the run back to the user, walk this checklist. Every item must be YES. If any is NO, you are not done — go finish that item.

• runs/<name>/signal-doc.md is filled with real content (not the scaffold placeholder)
• runs/<name>/message-spine.md is filled
• runs/<name>/first-page-decision.md is filled AND lists candidates considered with a paragraph case FOR each
• runs/<name>/page-argument-shape.md is filled
• runs/<name>/proof-map.md is filled
• runs/<name>/first-page-draft.md is filled
• runs/<name>/outputs/NN-*-output.md exists for every step 01 through 07
• runs/<name>/claim-check.md is filled (not the *[Filled in by step 07]* placeholder)
• runs/<name>/first-page-draft-corrected.md exists if claim-check flagged any line
• runs/<name>/working-log.md has real entries per step (not the scaffold *[Fill in]* placeholders)
• runs/<name>/evaluation.md is filled
• runs/<name>/evaluator-pass.md is filled (use the pagekit-evaluator-pass skill/subagent)
• bash scripts/slop-check.sh runs/<name>/first-page-draft.md exits 0 clean
• bash scripts/run-check.sh runs/<name> returns tier: FULLY LOGGED or tier: PUBLISHABLE

Only after every box is checked do you report completion to the user.

When to ask the user

• First-page decision is borderline (homepage vs. non-homepage close call): show both candidates and ask.
• Claim-check at hard severity is rewriting more than 25% of the draft: pause and check whether the proof map needs strengthening rather than the draft.
• Source briefs are too thin to support a meaningful run: stop and ask for more material.

When to delegate

Use the pagekit-claim-checker subagent (agents/pagekit-claim-checker.md) for the claim-check step. It is read-only and has a tighter focus, which keeps the main session uncluttered.

For tooling steps, invoke the corresponding skill rather than calling the script directly so the rationale is logged in the conversation.

Related

• Per-step skills: pagekit-signal-doc, pagekit-message-spine, pagekit-first-page-decision, pagekit-page-argument-shape, pagekit-proof-map, pagekit-first-page-draft, pagekit-claim-check
• Tooling: pagekit-new-run, pagekit-run-check, pagekit-slop-check, pagekit-evaluator-pass
• Subagents: pagekit-claim-checker (step 07), pagekit-evaluator-pass (adversarial read)