pagekit-claim-check

PageKit step 07 — Claim Check

You are running the optional durability pass on a first-page draft.

Read first

• ./references/prompt.md
• ./references/framework.md — claim-check flag types, severity calibration, principle
• ./references/template.md
• ../pagekit/references/anti-slop.md — the slop pattern catalog

Inputs

• runs/<run-name>/first-page-draft.md
• runs/<run-name>/proof-map.md
• Severity: light, normal, or hard. Ask the user if not specified; default to normal.

Procedure

Option A — delegate to the subagent (preferred)

Invoke the pagekit-claim-checker subagent (agents/pagekit-claim-checker.md). Pass:

• the path to the draft
• the path to the proof map
• the severity

The subagent runs read-only, produces the audit, and writes:

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

Option B — run inline

Use scripts/claim-check.sh <draft> <proof-map> --severity <severity> to expand the canonical prompt for paste. Save outputs to the same paths.

After the pass

1. Save the audit and corrected draft.
2. Update working-log.md with the severity used, count of flags by type, and any recurring patterns.
3. If the same slop pattern showed up more than once, add it to ../pagekit/references/anti-slop.md (or strengthen the existing entry) and to ../pagekit-page-argument-shape/references/template.md's drafting-constraints list. The framework absorbs what the claim-check kept catching.

Quality gate

A strong claim-check pass:

• names specific lines, not general impressions
• separates claim types (overclaim / vague / unsupported / outrun proof map / editorial voice / unsourced quantitative / clinical drift / ai-slop tell)
• proposes replacements that are sharper and more accurate than the originals, not simply softer
• ends with a recurring-patterns summary so upstream constraints can absorb them

Next

Do not stop at claim-check. Continue:

1. Write runs/<name>/evaluation.md — the run's own honest read of what worked, what stayed thin, where outputs drifted generic.
2. Invoke pagekit-evaluator-pass to produce the adversarial second read.
3. Ensure runs/<name>/working-log.md has real entries per step (replace all scaffold *[Fill in]* placeholders).
4. Invoke pagekit-run-check — it must return tier: FULLY LOGGED or tier: PUBLISHABLE before the run is done.