writing-validate

name	writing-validate
description	Validate canonical writing-plan claims against draft outputs before review; records live gaps in TaskList and returns coverage results.
user-invocable	false
disable-model-invocation	true
allowed-tools	Read, Grep, Glob, Bash, TaskCreate, TaskList, TaskUpdate

Claim Validation

!bun ${CLAUDE_SKILL_DIR}/../../scripts/load-constraints.ts writing-validate

Validate each canonical writing-plan claim against its mapped draft output before prose review. The receipt-selected immutable {planFile, planHash} contains claims, structure, claim-to-section mapping, source plan, section outputs, and review surfaces. TaskList contains live validation gaps and resolution work. Do not create or consume retired writing planning ledgers.

Iron Law

NO REVIEW WITHOUT CANONICAL CLAIM VALIDATION. Authenticate and read the receipt-selected plan; parse its deterministic section index; read each mapped draft; then capture every missing or partial claim as TaskList work before review begins.

A legacy-only layout may be converted into a fresh approved plan, but it cannot enter this validation flow as authority.

Process

1. Authenticate the selected {planFile, planHash} and verify the canonical writing grammar/index. Missing or malformed canonical state fails closed.
2. Load writing constraints, the style selected in ## Writing Intent, and ai-anti-patterns.
3. Read ## Claims, ## Counterarguments, ## Claim → Section Map, ## Section Outputs, and the corresponding drafts/ deliverables.
4. For every claim, assess: mapped output exists, substantive argument, source/evidence support, thesis threading, domain compliance, and AI-pattern compliance.
5. Classify each claim as COVERED, PARTIAL, or MISSING. Create one TaskList item for every PARTIAL or MISSING finding, tied to the current plan hash and exact draft path.
6. Return the coverage result to the caller. If no gaps are open, review may proceed. If gaps remain, present them for the user’s decision: revise through TaskList or explicitly proceed with known gaps.

Return contract

Claim validation result
- Plan: receipt-selected `{planFile, planHash}`
- Coverage: [covered]/[total]
- Covered claims: [CLAIM-NN list]
- Open findings: [TaskList IDs, CLAIM-NN, draft path, exact gap]
- Mechanical checks: [commands and results]
- Review admission: READY | USER_DECISION_REQUIRED

Red flags

About to	Stop because	Do instead
Read retired précis or master-outline files	Canonical PLAN is the approved writing specification	Read the receipt-selected plan and deterministic index.
Use an active-workflow marker for style	Stable configuration belongs in the plan	Read ## Writing Intent.
Write VALIDATION.md, phase summaries, or learning ledgers	They become competing lifecycle authority	Use TaskList and return the result.
Claim full coverage without reading drafts	Mentioning is not arguing	Verify each mapped draft.
Silently pass partial or missing claims	The user must decide whether gaps are acceptable	Create TaskList findings and return them.