Table System Migration

Overview

Use this skill to turn a messy estate of one-off HTML tables into a durable table system: a canonical component, compatibility rules for legacy/admin surfaces, a ratcheting linter, and real-browser layout regression tests.

Confidentiality Gate

Before writing or publishing any reusable output, scrub it as if the target repository is public.

• Do not include local absolute paths, private repo names, production domains, customer names, emails, usernames, hostnames, tokens, passwords, keys, analytics IDs, database credentials, or commit hashes.
• Do not paste private source files, proprietary copy, unreleased content, exact defect counts, or internal incident details into the skill or writeup.
• Generalize examples. Prefer "content site", "admin grid", "canonical table class", and "legacy table class" over project-specific names.
• If the user explicitly wants project-specific wording, keep it in the private repo or ask before staging it in a public skills repo.
• Run a secret/string scan on the final diff before staging. Treat suspicious matches as blockers until explained or removed.

Workflow

1. Map The Table Estate

Inspect the repository before editing.

• Identify content directories, templates/components, CSS entry points, generated assets, test commands, and CI gates.
• Inventory table markup patterns: canonical classes, legacy classes, wrapper classes, naked <table> elements, per-page <style> blocks, captions, footnotes, scroll wrappers, numeric cells, total rows, and admin/data-grid tables.
• Separate user-facing content tables from operational/admin grids. They may share CSS selectors but often need different layout behavior.
• Note existing failures or intentional baseline debt. Do not promise a full cleanup if the requested task is only a ratchet or migration of priority surfaces.

2. Define The Canonical System

Specify the smallest table API that covers the real use cases.

• Choose one canonical table class and one canonical wrapper class.
• Define header, body, numeric, label, total-row, caption, source, footnote, methodology/detail, grouped-header, and horizontal-scroll behavior only when each is actually needed.
• Preserve a compatibility layer for legacy/admin table classes if removing them would widen the blast radius.
• Avoid one-off per-post or per-page table classes unless they encode a genuinely distinct component.

3. Migrate In Risk Order

Make narrow, reviewable passes.

• Start with the reported broken table or highest-traffic content.
• Convert repeated old patterns to the canonical wrapper/table pair.
• Keep content semantics intact: headings remain headings, captions remain captions, source notes remain source notes, and data values do not change unless the user asked for content edits.
• Preserve intentional wide-table behavior with a wrapper-level scroll container instead of forcing every cell to wrap.
• For a long supporting table that sits below a summary or callout, collapse it by default inside a native disclosure element and surface a row-count or scope hint in the toggle, so the summary carries the answer and the full table stays available as on-demand detail.
• Avoid unrelated copy edits while touching content files.

3a. Adding The Canonical Class Is Not Enough — Neutralize Competing Overrides

This is the silent failure of a class-only migration. You add the canonical class, the table validates, the ratchet is green — and it still renders off-brand, because a pre-existing per-page rule out-specifies the canonical component and wins.

• The mechanism. The canonical header rule is typically table.<canonical> thead th (specificity 0,1,3). A leftover page rule like table.<legacy> thead th { background: #1a2330 } has the same specificity (0,1,3), so the cascade falls through to source order — and a page <style> block (or any later stylesheet) comes after the linked canonical CSS, so the leftover wins. The header silently keeps its old color. Multi-class (.a.b th), id (#x th), and !important rules beat the canonical component outright.
• A .legacy thead th (0,1,2) without the table. prefix LOSES to the canonical table.<c> thead th (0,1,3) — so it is already overridden and harmless. The dangerous ones are element-prefixed (table.X …), multi-class, id-scoped, or !important. The bundled ratchet's canonical_override check flags exactly these.
• Remediation: remove the competing paint, do not re-add it. Delete the background/color from the offending rule so the canonical styling takes over. Preserve structural rules in the same block — min-width (horizontal scroll), position: sticky, @media card-mode, display:none hidden columns, numeric text-align, special semantic cell classes, ::before/::after content. Removing the whole block is fine when it is pure cosmetic re-implementation; reduce it to the structural declarations when it is not.
• Do not blanket-strip — preserve genuine visualizations. Heatmaps, comparison matrices where header color encodes data, frozen-corner pivots, and JS-rendered/sortable tables are deliberate. Allowlist their class and re-baseline; do not flatten them to the canonical look.
• Verify the canonical style actually wins — by computed style, with the cache bypassed. Stylesheet caching is notoriously sticky: a page can keep rendering the old header until a hard reload. Confirm the fix by reading the computed background-color of a thead th (e.g. it equals the canonical brand value) after a cache-ignoring reload — never trust source-only or a soft refresh.

4. Add A Ratchet

Add a deterministic check that blocks new table debt without requiring a full historical cleanup.

• Start from the bundled table-ratchet.py script and references/table-ratchet-checklist.md when the target repo does not already have a table-debt gate. The checklist lives under references/ because it is implementation guidance; the executable script stays at the skill root so it can be copied or invoked directly.
• Configure the canonical table class, canonical wrapper class, generated-folder excludes, and any narrow legacy/admin table allowlists before creating the baseline.
• Scan content/templates for new unsanctioned table classes, new per-page table CSS, new unwrapped tables, and canonical_override rules (page-local selectors that out-specify and re-paint the canonical header — see step 3a). The canonical_override kind is what catches the "class is present but it still looks wrong" case that the other three checks miss.
• Store a baseline when legacy debt already exists.
• Refresh the baseline only after reviewing current debt and confirming the new counts are intentional:

python3 table-ratchet.py --root <repo> \
  --canonical-table-class <class> \
  --canonical-wrapper-class <wrapper> \
  --write-baseline

• Fail only when a file exceeds the baseline or a new file introduces debt.
• Print actionable remediation text: which file regressed, what count increased, and which canonical classes to use.

5. Add Real-Browser Regression Tests

Use browser-backed tests for layout behavior that DOM-only environments cannot see.

• Mount minimal fixtures for each shared surface: content table, admin grid, wide table, caption/source block, total row, grouped header, and any special numeric alignment convention.
• Assert computed style and actual box geometry where layout matters: widths, min-width behavior, wrapping, sticky columns, overflow, hover backgrounds, and padding after prose resets.
• Assert that the canonical styling actually wins on a migrated table: read the computed thead th background-color and confirm it equals the canonical brand value, not a leftover per-page color (the canonical_override regression, step 3a). A class-presence assertion is not enough — it passes even when the header is silently overridden.
• Build/import the real compiled CSS bundle rather than duplicating CSS in the test.
• Include at least one test that protects admin/data-grid tables from changes made for prose/content tables.

6. Validate Before Calling Done

Run the smallest complete gate set for the touched stack.

• Run the table ratchet.
• If using the bundled artifact, run python3 table-ratchet.py --root <repo> ... with the same class and allowlist flags used to write the baseline.
• Rebuild CSS/assets if tests import generated bundles.
• Run browser/CSS regression tests.
• Run compile/type/lint checks that CI uses for touched code.
• Check git diff --check and confirm the working tree contains only intended files.
• If full tests are known to have unrelated failures, report the exact focused gates that passed and the known reason full tests were not used.

Output Contract

When reporting status, distinguish these clearly:

• Landed changes: canonical CSS/classes, migrated files, tests, linter, docs.
• Verification: exact commands run and pass/fail results, including the table-ratchet command and baseline path if one was added.
• Remaining debt: legacy classes/unwrapped tables that are intentionally baselined.
• Review notes: files staged, files not staged, and any suspicious scrub findings that were removed or intentionally kept.

Example Prompts

• "Use $table-system-migration to migrate this blog's tables to one canonical component and add a no-new-violations check."
• "Use $table-system-migration after this table CSS change; I want proof that admin grids did not collapse."
• "Use $table-system-migration to turn this private table cleanup into a public-safe skill without leaking project details."

Skill Maintenance

When editing the ratchet script, run:

PYTHONDONTWRITEBYTECODE=1 python3 -m unittest tests/test_table_ratchet.py