Upgrade Interrogation Pages

Upgrade Interrogation Pages

Invoke as $upgrade-interrogation-pages.

Use this skill when a repository already has generated interrogation/*.html round pages and the user wants to audit or update them against the current local interrogation-page convention. This skill targets generated interrogation HTML artifacts only.

Inputs

• Default target repository: current working directory.
• Optional --repo <path> to target another repository.
• Optional explicit page paths under the target repository's interrogation/ directory.
• Optional --apply, or clear user wording such as "apply the upgrade" or "upgrade these pages", to allow page rewrites.

Mode

• Default mode is audit/dry-run.
• The skill name alone is not apply intent; do not modify files unless the visible user request explicitly asks to apply or upgrade existing pages.
• In audit/dry-run mode, inspect pages and report drift without creating, editing, moving, or deleting any files.

Process

1. Resolve scope:

   • Locate the project-root interrogation/ directory; stop if absent.
   • Enumerate interrogation/*.html round pages.
   • If the user provided page paths, include only those paths after confirming they are inside the project-root interrogation/ directory.

2. Load local standards:

   • Read applicable local convention sources from AGENTS.md, CLAUDE.md, source docs/interrogation-page-convention.md, packaged assets/interrogation-page-convention.md, and legacy sibling INTERROGATION-PAGE.md files found under base/, packs/, .claude/skills/, and .codex/skills/.
   • Treat those local files as read-only standards. Do not fetch external standards.
   • Extract the current required HTML behaviors: the six open-question markers (data-open-input, visible data-recommended-answer, hidden data-agent-recommended-answer, data-agent-confidence with a value in {high, medium, low}, a data-clarify-copy button, and a data-apply-recommended button on every data-open-question block), the ≥1 open input per round rule, the page-metadata attributes (data-visual-tier, data-interrogation-status, data-interrogation-round, data-interrogation-gate), round-file naming ({skill}-r{N}-{branch}.html), the data-answer-sidecar reference, the Brief Me TTS include, dark-mode styling, responsive layout, the embed prohibition, and archive-before-replace behavior.

3. Audit every selected page:

   • Read the whole HTML file before judging it.
   • Identify page-specific substance: title, meta/context, assumptions manifest, open questions, visible recommendations, hidden agent recommendations, confidence badges, apply-recommended controls/scripts, gate controls, coverage summaries, prior context, and any page-specific scripts.
   • Report missing or stale standard features per page.
   • Classify each page as current, upgrade-needed, or blocked-content-loss-risk.

4. Stop on content-loss risk:

   • If a page's structure is too ambiguous to upgrade while preserving substantive content, do not rewrite it.
   • Report exactly what content could not be mapped and what human decision is needed.

4b. Batch-mode handoff (apply mode only):

• Count pages classified as upgrade-needed (excluding blocked-content-loss-risk).
• If count > 2:
  • Generate a phase in tasks/todo.md titled "## Phase: Upgrade Interrogation Pages" with:
    • Execution Profile: serial
    • A Context block listing the standards sources discovered in step 2 and the archive timestamp path (docs/history/archive/YYYY-MM-DD/HHMMSS/)
    • One checklist step per upgrade-needed page: - [ ] Upgrade interrogation/<filename>.html — archive original, rewrite to current standard preserving all content, verify content preservation
    • A final step: - [ ] Run node scripts/audit-interrogation-pages.mjs and verify all archives; commit and push
    • Acceptance criteria matching the audit findings
  • If tasks/todo.md already has unchecked items, append the phase at the bottom with a separator (---) and warn the user about pre-existing items.
  • Output the full audit report.
  • Report: "Generated exec-loop task plan with N steps. Recommended next route: approved task artifact"
  • Stop. Do not modify any HTML files.
• If count <= 2: proceed to step 5 (inline apply).

5. Apply only when explicit:

   • Before replacing each page, archive the original to docs/history/archive/YYYY-MM-DD/HHMMSS/interrogation/<filename>.html.
   • Rewrite the page contextually from the original, preserving page-specific assumptions, questions, gates, coverage context, and prior answers.
   • Preserve the round filename, the r{N} round number, and the data-answer-sidecar path on rewrite.
   • Author the now-required helpers — visible data-recommended-answer, hidden data-agent-recommended-answer, data-agent-confidence, and data-apply-recommended — for each open question from the page's own elicitation context.
   • Preserve an existing visible recommendation as user-facing guidance when useful, and preserve or create a distinct hidden agent payload from the page's recommended default. The hidden agent payload must be answer-shaped and free of user-facing instructions.
   • Ensure each data-apply-recommended button is labeled Apply recommended, fills the nearest data-open-input from the nearest hidden data-agent-recommended-answer, falling back to data-recommended-answer only for transition compatibility, confirms before replacing non-empty input, dispatches input and change events after setting the value, supports textarea and text input controls, and does not use clipboard APIs.
   • Add or modernize missing standard markers without replacing the page with a generic template.
   • Preserve relative links and keep the page self-contained unless the original already used explicit external resources.
   • Highlight upgrade changes where useful with inline notes, a change summary section, or scoped diff styling.

6. Verify after apply:

   • Confirm every changed page has a matching archive copy.
   • Re-open each changed page textually and confirm the preserved title, assumptions, questions, visible recommendations, hidden agent recommendations, apply-recommended controls/scripts, and gates still exist.
   • Run node scripts/audit-interrogation-pages.mjs to confirm the changed pages now conform to the convention.

Audit Report

Include:

• Target repository and mode.
• Standards sources read.
• Pages inspected and pages excluded.
• Per-page status with missing standard features.
• Content preservation risks or blockers.
• In apply mode, archive paths and changed page paths.
• Verification performed.
• Next action: usually node scripts/audit-interrogation-pages.mjs after successful apply, or no mutation needed when all pages are current.
• In batch mode: the generated tasks/todo.md phase with execution steps and the approved-artifact route.

Constraints

• Do not edit CLAUDE.md, AGENTS.md, any SKILL.md, or any bundled INTERROGATION-PAGE.md file.
• Do not modify non-interrogation HTML files.
• Do not rewrite pages in audit/dry-run mode.
• Do not replace page-specific content with a generic template.
• Stop and report when an upgrade is likely to lose assumptions, questions, visible recommendations, hidden agent recommendations, apply-recommended controls/scripts, gates, or coverage context.
• Do not create or modify GitHub Actions workflows.

Default Shipping Contract

• Audit mode: no file mutations; no commit or push.
• Apply mode (inline, ≤2 pages): modified generated interrogation pages and archive copies are normal repo artifacts. Follow the target repository's shipping rules after verification.
• Apply mode (batch, >2 pages): generates tasks/todo.md phase only; does not modify any HTML files. Commit the task plan, then stop; the approved task artifact records the next executable step.
• Default next-step routing: after successful inline apply, Recommended next command: node scripts/audit-interrogation-pages.mjs; after batch-mode handoff, Recommended next route: approved task artifact; otherwise report the audit result and recommend no follow-up skill when no upgrades are needed.