# Praxis Table AI Validation

> Praxis Table AI Validation

- Skill: `codexrodrigues/praxis-table-ai-validation` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx get-skill add codexrodrigues/praxis-table-ai-validation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/codexrodrigues/praxis-table-ai-validation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: codexrodrigues (https://skillmd.com/u/codexrodrigues)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/codexrodrigues/praxis-table-ai-validation

---


# Praxis Table AI Validation

Use this skill for table AI authoring and validation. The table assistant must operate through declared semantic contracts: component edit plans, runtime operations, capability catalogs, filter field catalogs, selected-record context, global action catalogs, resource surfaces, and AI registry ingestion. It must not route primary intent through local keyword matching.

For shared AI runtime, manifest, registry, or semantic intent work, pair this skill with `praxis-ai-assistant-runtime`, `praxis-ai-authoring-manifests`, `praxis-ai-registry-ingestion`, or `praxis-ai-semantic-intent`.

## Source Audit

Inspect the AI and registry surface:

- `projects/praxis-table/AGENTS.md`
- `projects/praxis-table/src/lib/ai/praxis-table-authoring-manifest.ts`
- `projects/praxis-table/src/lib/ai/table-component-edit-plan.ts`
- `projects/praxis-table/src/lib/ai/table-component-ai-capabilities.ts`
- `projects/praxis-table/src/lib/ai/table-ai-capabilities.ts`
- `projects/praxis-table/src/lib/ai/table-authoring-registry.ts`
- `projects/praxis-table/src/lib/ai/table-agentic-authoring-turn-flow.ts`
- `projects/praxis-table/src/lib/ai/table-ai.adapter.ts`
- `projects/praxis-table/src/lib/ai/table-context-pack.ts`
- `projects/praxis-table/src/lib/detail/table-detail-authoring.ts`
- `projects/praxis-table/src/lib/ai/filter-form-dialog-host-context-pack.ts`
- `projects/praxis-table/src/lib/ai/*.spec.ts`
- `projects/praxis-table/test-dev/e2e/table-ai-*.playwright.spec.ts`
- `projects/praxis-table/test-dev/e2e/table-ai-assistant-selection-context.playwright.spec.ts`

Inspect `tools/ai-registry/AGENTS.md` when registry ingestion or component authoring catalog output changes.

## AI Contract Rules

- Component edit plans are for authoring config changes, not for executing runtime operations.
- Runtime operations are for declared immediate actions such as applying filters, exporting data, or opening related surfaces.
- Use `table.export.run` only when the user asks to run an export now. Its available formats must come from the runtime export configuration and normalized supported formats; enabling/configuring export belongs to `export.configure` in a component edit plan, and ambiguous export requests should ask for scope/format clarification instead of fabricating a download operation.
- Filter questions must be grounded in `filterFieldCatalog`, resource metadata, and runtime operations.
- Column visibility, public/private presentation, format, renderer, style, value mapping, computed columns, toolbar, row, bulk, and export authoring must use declared edit plan change kinds and operation ids.
- Treat `visual-only` operations such as `column.conditionalRenderer.add`, `column.conditionalStyle.add`, row conditional renderers, badges, chips, icons, animations, and `effects` as table presentation materialization only. They may explain how a row or cell is displayed when a JSON Logic condition matches, but they must not classify the record, synthesize boolean status patches, create `domain-rules`, infer `backend_validation`, or become row authorization/business policy.
- Capability discovery questions should answer from the catalog and stay consultative until the user explicitly asks to add, apply, or materialize.
- Global actions must be chosen from declared catalogs and payload schemas. Ask concise clarification when required payload cannot be grounded.
- Selected-record context must be sanitized and human-facing; do not expose raw IDs, internal fields, endpoint paths, schema keys, or payload examples unless explicitly requested.
- Do not invent unsupported OR/nested boolean filters when the resource/runtime contract does not materialize them.
- Author contextual detail only through declared operations such as `detail.configure`, `detail.source.configure`, and `detail.presentation.configure`. Retrieve the current manifest before selecting an operation and keep row expansion interaction separate from the shared `behavior.detail` source/presentation contract.
- Source-only operations must preserve presentation and unknown extensions. Validate resource/source fallback completeness and cycles, inline Rich Content documents and class names, and action references against declared row actions.
- A validator name in a manifest is insufficient. The owning `praxis-config-starter` registry must dispatch executable validation with failing and passing contract tests, and its effect compiler must materialize the same canonical path used by the Angular runtime.
- When the table authoring manifest changes, regenerate the ingestion registry into the explicit Config Starter destination, update the snapshot hash contract, and validate both the frontend catalog and backend classpath snapshot before integration.

## Intent Guardrail

Do not add local keyword, regex, alias, or fuzzy matching as the primary intent router. Text matching is allowed only after a canonical intent/operation scope exists, for ranking declared fields/actions/surfaces or asking clarification.

When a required operation is missing, model or extend the canonical AI/tool contract. Do not replace the missing tool with local prompt parsing inside table components.

## Validation

- Manifest changes: `praxis-table-authoring-manifest.spec.ts` and table authoring registry specs.
- Component edit plans: `table-component-edit-plan` specs and AI adapter specs.
- Runtime operations: `table-agentic-authoring-turn-flow.spec.ts`, `table-ai.adapter.spec.ts`, and E2E flows for selected records, filters, export, and consultative follow-up.
- AI registry: `npm run generate:registry:ingestion` and `npm run validate:catalog` when the manifest/catalog surface changes.
- Contextual detail: `table-detail-operations.spec.ts`, detail authoring/editor specs, backend `TableDetailAuthoringSemanticsTest`, effect-compiler and validator-registry tests, `AgenticAuthoringManifestServiceTest`, and `AiRegistrySnapshotContractTest`.
- Browser-real AI: use focused Playwrights such as `table-ai-assistant-component-edit-plan`, `table-ai-assistant-selection-context`, `table-ai-assistant-consultative-follow-up`, and opt-in live matrix only when live AI is intentionally required.
- Managed Agents API previews: keep host HTTP/provider qualification separate from the Angular rendering proof. Feed the server-emitted `pagePreview`, `currentPreview`, real response/filter schemas, and sample rows to `table-agents-api-host-pilot.playwright.spec.ts`; require the existing table input `configPersistenceStrategy: 'volatile'` and assert the compiled columns and filters without another AI call or autonomous write. This browser proof alone is not a full HTTP authoring E2E.
- In that browser proof, classify POST `/filter` and `/stats/*` using the owning Quickstart controller's read-query contract, not the HTTP verb alone; block and record actual mutations. Inspect the filter schema before asserting field availability. The filter-field manager stages checkbox changes until **Apply**: capture the accepted-preview screenshot first, then apply a removal before checking the resulting inline filters.

State whether live AI E2E was skipped. Do not use GitHub Actions as exploratory validation.

## Companion Skills

- Use `praxis-table-data-source-precedence` when the assistant authors or explains local/remote mode, `resourcePath`, `queryContext`, or persisted bindings.
- Use `praxis-table-selection-export-runtime` when the assistant works with selected records, bulk actions, export operations, or selected-record context.
- Use `praxis-table-analytics-stats-runtime` when the assistant works with analytics projections, stats, or dashboard table views.
- Use `praxis-table-renderer-state-diagnostics` when the assistant changes renderers, formatting, empty/loading/error states, or diagnostics.
- Use `praxis-table-runtime-data` for the broader runtime state umbrella.
- Use `praxis-table-filter-actions` for filter/action operation grounding.
- Use `praxis-table-authoring-settings` for Settings Panel and config editor round-trip.
- Use `praxis-table-rule-ai-validation` when AI authoring targets table visual effects, presets, semantic animations, or `tableIntegration.delegate`.
- Use `praxis-table-rule-table-integration` when table-owned renderer placement must coordinate with table-rule-builder effect payloads.
- Use `praxis-core-runtime-contracts` and `praxis-core-resource-runtime` for shared AI contracts, resource capabilities, surfaces, and schema grounding.
- Use `praxis-authoring-editors` for cross-component authoring/persistence patterns.

