Praxis AI Registry Ingestion
tools/ai-registry is the canonical generator and governance boundary for AI-facing Angular artifacts. Registry output describes existing Praxis contracts for grounding; it does not invent component semantics, route intent, or replace source manifests, metadata, backend schemas, capabilities, actions, surfaces, or runtime code.
Provenance Pipeline
component source, metadata, README/docs, authoring manifests, capability/context packs
-> extract-component-docs.js + generate:ai-metadata
-> dist/praxis-component-registry.json
-> generate-registry-ingestion.ts
-> dist/praxis-component-registry-ingestion.json (canonical aggregate corpus)
-> catalog + authoring-contract validation
-> optional RAG/provider/AI-ready/package/upload projections
The ingestion corpus with components[].chunks is canonical for aggregate analysis. praxis-component-registry-rag.json, provider projections, AI-ready docs, reports, npm package assets, and backend uploads are derived read models. Never hand-edit dist/**, package assets, projections, or reports to repair a source/contract problem.
Read tools/ai-registry/AGENTS.md, its README, workspace AGENTS, and the affected library's AGENTS before editing. Inspect the actual owner source: public API, component metadata, docs manifest, authoring manifest, capabilities/context pack, focused specs, and consumer docs.
generate-registry-ingestion.ts projects authoringManifest, authoringManifestProfiles, capabilities, component context, AI concepts, and authoring-only manifest entries into the canonical corpus. This projection is evidence of declared Angular capability; it is not proof that the backend can execute every target resolver, validator, or effect handler.
When backend registry persistence, bootstrap, classpath snapshot, template search/upsert, RAG document identity, or official upload behavior is in scope, also inspect praxis-config-starter/AGENTS.md, RegistryIngestionController, RegistryIngestionService, RegistryIngestionRequest, AiRegistry, AiRegistryRepository, AiRegistryBootstrapService, AiRegistryStatusService, AiRegistryTemplateController, AiRegistryTemplateService, and src/main/resources/ai-registry/registry-snapshot.json. The backend owns persisted registry identity, release-scoped RAG projection, snapshot bootstrap, template records, status/health, revision/ETag behavior, and rejection of invalid authoring manifests or presentation affordance catalogs.
Decide The Smallest Canonical Operation
| Change |
Owner and minimum action |
| component docs extraction or AI metadata |
source docs/metadata first, then npm run generate:registry:ingestion |
| ingestion generator/schema/chunks/governance |
generate:registry:ingestion; validate catalog and authoring contracts |
| catalog validation only |
npm run validate:catalog |
| completed authoring contract acceptance |
npm run validate:authoring-contracts |
| AI planning/RAG/templates |
npm run validate:ai |
| package-scoped asset after ingestion and library builds |
npm run package:ai-assets, then package/tarball validation if release scope requires it |
| provider projection |
regenerate from the canonical ingestion corpus for the explicit release; never upload secrets/external IDs into it |
| backend upload |
use tools/ai-registry/run-all.sh or npm run ai:sync-backend only when a real backend state change is required |
npm run generate:registry:ingestion is the standard local pipeline: extract component docs, generate AI metadata, generate ingestion corpus, synchronize the Config classpath snapshot and its contract constants, validate catalog governance, and validate authoring contracts. It is the correct response to a source/manifest/capability/context-pack change; it is not required merely to edit a derived report or inspect a package asset.
Isolated Worktree Wiring
The snapshot synchronizer and authoring acceptance validator use the existing
PRAXIS_CONFIG_STARTER_ROOT override, otherwise a sibling Config checkout. Before
running the pipeline in an isolated Angular worktree, point this override to the
explicitly assigned, clean Config worktree. Inspect both owners' instructions;
never direct generated writes into another worker's dirty primary checkout or
create an ad hoc sibling link. The sync writes both registry-snapshot.json and
AiRegistrySnapshotContractTest constants, so reserve both files for one writer.
If generation succeeded but sibling discovery failed, preserve the generated
corpus and its provenance. Resume the failed sync/validation steps with the same
override instead of regenerating timestamps or weakening the validator. Review
the complete semantic snapshot delta against current Angular source: a stale
snapshot may include earlier canonical changes beyond the triggering patch.
Prove corpus equality, manifest preservation and the focused snapshot contract;
source synchronization does not establish backend upload or a public release.
Canonical Content Rules
- Keep component docs extractable from supported literal metadata/factories. If extraction cannot represent a semantic feature, improve the source metadata or official extractor, not a consumer-side registry patch.
- A manifest, capability, target, validator, effect, profile, context pack, and docs chunk must point to the source owner and remain mutually consistent. Do not make a generated registry the primary definition of an operation.
- Treat
npm run validate:authoring-contracts as the Angular acceptance gate for structural and semantic completeness: target kind/resolver, ambiguity policy, validators, effects, affected paths, destructive confirmation, submission impact, remote-binding evidence, and profile metadata. It does not replace backend execution tests for AgenticAuthoringTargetResolverRegistry, AgenticAuthoringValidatorRegistry, or AgenticAuthoringEffectCompilerRegistry.
- When a manifest introduces a new resolver id, validator id, effect kind, or
compile-domain-patch handler, classify it as suportado-parcialmente until the backend registry implements it and focused config-starter tests prove the executable path. Do not upload or package the corpus as executable proof just because the generated Angular artifact contains the declaration.
- Registry projections must preserve the traceable authoring-manifest fields required for semantic decisions and backend validation: component/profile id,
manifestVersion, stable operationId, target.kind, target.resolver, target.ambiguityPolicy, target.required, validator ids, effect kind/handler/handler contract, affectedPaths, submissionImpact, and presentation-affordance sourceRef, defaultTargetKind, and options. Do not collapse these into prose chunks, display labels, quick-reply text, or RAG-only summaries.
- For family components such as
praxis-dynamic-fields, authoringManifestProfiles on child
entries are a generated applicability projection. generate-registry-ingestion.ts matches control
profiles from the family manifest to component docs through meta.id, selector, tags, or metadata
input type. A projected child profile is useful evidence that the corpus preserved applicability,
but it is not a source definition of the profile, not proof that a capability alias is a component
identity, and not a substitute for runtime/editorial/catalog/source metadata evidence.
- Registry grounding is evidence for LLM semantic resolution. Do not encode primary intent as keyword, regex, alias, fuzzy match, display-label parsing, or a registry-only command dialect.
- Quick replies/actions must retain structured canonical action, semantic decision, target, context, risk, and presentation. Labels are not authority.
- If quick replies, semantic decisions, or component edit plans cite registry/RAG/provider projections, they must reference projected manifest operations and targets by stable ids. A projection that loses those ids is derived-artifact drift to repair at the source, not permission for backend, UI, or agents to fall back to text matching.
- Treat missing fields, components, manifests, operations, ports, presets, or chunks in generated registry artifacts as projection
loss or source/extractor drift until proven otherwise. Absence from
component-docs.json, the ingestion corpus, RAG/provider
projections, package assets, or backend snapshots is not proof that the component, capability, authoring manifest, or backend
support does not exist. Re-audit the owner source, provider registration, extractor support, generated artifact lineage, and
backend snapshot before declaring a platform contract gap.
- Preserve the runtime observation contract chunk and its
untrusted_frontend_observation boundary. Observations require backend grounding into canonical context before they influence an answer; registry text cannot grant read/tool/apply authority.
- Catalog governance must reject duplicate IDs/selectors, invalid paths, incomplete target/validator/effect evidence, unconfirmed destructive work, ambiguous targets without policy, inconsistent control profiles, and lost trust boundaries. Repair the source owner or validator, never weaken the gate to pass a local artifact.
- If a dynamic-fields child entry is missing
authoringManifestProfiles, or has a profile that is
not declared by the family manifest, first audit component docs extraction and the source
appliesTo rules. Treat the mismatch as source/extractor/projection drift until proven otherwise;
do not patch the generated corpus, add a registry-only profile, or promote a capability alias into
componentIds just to satisfy the gate.
- Use aggregate corpus for cross-package/release analysis and
@praxisui/<package>/ai/component-registry.json only for source-less package consumers. Package assets remain filtered projections of their owning package.
- Broad backend registry status, reconciliation, startup, and candidate scans must use compact repository projections. Do not hydrate embeddings, full chunk content, raw registry payloads, or large manifests for every row merely to compute identity, readiness, counts, hashes, or candidate summaries; hydrate the exact canonical record only when publication or selected-evidence materialization requires it.
Before New Contracts Or Scripts
Inventory the existing metadata, x-ui, schemas, actions, surfaces, capabilities, option sources, manifests, registry chunks, diagnostics, context packs, docs, and backend API catalog. Classify the need as ja-suportado-so-ux, ja-suportado-mal-nomeado-ou-mal-materializado, suportado-parcialmente, or lacuna-real-de-contrato.
Only a real contract gap justifies a schema, chunk, generator, endpoint, or script change. Map source owner, generated outputs, Angular/package consumers, backend ingestion, security/origin requirements, docs/playgrounds, compatibility, and focused validation first. Do not add ad hoc upload, extraction, or sync scripts outside this tool owner.
Remote And Release Boundaries
- Local generation and validation come first. GitHub Actions are release/final gates, not discovery tooling.
- Backend uploads use the documented environment and official scripts:
BACKEND_URL, CONFIG_ORIGIN, APP_SECURITY_READ_OPEN, and APP_SECURITY_WRITE_OPEN. Do not invent endpoints, origins, ports, or upload ordering.
run-all.sh --with-api-catalog is the canonical combined registry/templates/API-catalog flow. Do not additionally invoke individual upload scripts in the same operation.
run-all.sh generates AI-ready docs, runs npm run generate:registry:ingestion, optionally generates RAG, then uploads API catalog, registry, and templates through the official scripts. If BACKEND_URL is absent and upload is not explicitly skipped, it must fail instead of silently targeting a guessed host.
upload-registry.ts posts the canonical dist/praxis-component-registry-ingestion.json to /api/praxis/config/ai-registry/component-definitions with Origin from CONFIG_ORIGIN, ORIGIN, or the documented http://localhost:4003 fallback. Do not replace this with ad hoc curl snippets unless the task is only diagnosing the official script.
package:ai-assets needs the generated ingestion corpus and built dist/<package> folders. It writes the scoped asset plus package export; validate local tarballs when package publication is in scope.
Evidence And Companion Skills
Run the smallest compatible command and report it exactly. For changes to tooling itself, add its focused Node self-test; for a source component, add that component's build/spec/authoring proof. State whether docs, recipes, manifests, landing, package assets, RAG/provider projections, backend sync, and release validation were affected.
For Angular authoring manifest or ingestion changes that may become executable through config-starter, pair the Angular gate with the backend manifest gate:
npm run generate:registry:ingestion
mvn "-Dtest=AgenticAuthoringManifestServiceTest,AgenticAuthoringManifestControllerTest,AgenticAuthoringManifestContractValidatorTest,AgenticAuthoringTargetResolverRegistryTest,AgenticAuthoringValidatorRegistryTest,AgenticAuthoringEffectCompilerRegistryTest" test
Use this paired proof when adding or renaming operation ids, target resolvers, validators, effect handlers, presentation affordance catalogs, or manifest profiles consumed by /api/praxis/config/ai/authoring/manifests/{componentId}/**.
For backend registry ingestion, snapshot, template, bootstrap, or official upload contract changes, run the focused config-starter gate:
mvn "-Dtest=RegistryIngestionServiceTest,RegistryIngestionServiceIdentityTest,AiRegistrySnapshotContractTest,AiRegistryBootstrapServiceTest,AiRegistryTemplateServiceTest,AiRegistryTemplateControllerTest,AiRegistryRevisionPolicyTest" test
When src/main/resources/ai-registry/registry-snapshot.json is updated from the Angular ingestion corpus, prove both sides: npm run generate:registry:ingestion in praxis-ui-angular, then the backend snapshot gate above. Do not update the classpath snapshot hash, version, generated timestamp, chunk counts, or unsupported-validator expectations without regenerating from the canonical Angular corpus and validating the backend snapshot contract.
Use praxis-ai-authoring-manifests for executable authoring contracts; praxis-angular-docs-playgrounds for public documentation; praxis-angular-public-api-governance for exports; praxis-angular-validation-gates for local proof; and the functional component skill for the changed source. Use praxis-config-agentic-authoring-streaming or praxis-ai-turn-orchestration-transport when registry chunks derive from governed turns, tools, observations, quick replies, or backend transport.
1---2name: praxis-ai-registry-ingestion3description: "Use when implementing, auditing, generating, packaging, or synchronizing Praxis AI registry artifacts: component docs extraction, AI metadata, ingestion corpus/chunks, catalog governance, authoring-contract acceptance, AI-ready docs, RAG/provider projections, package `ai/component-registry.json` assets, or official backend uploads."4metadata:5 source: https://skillmd.com/skills/codexrodrigues/praxis-ai-registry-ingestion6---78# Praxis AI Registry Ingestion910`tools/ai-registry` is the canonical generator and governance boundary for AI-facing Angular artifacts. Registry output describes existing Praxis contracts for grounding; it does not invent component semantics, route intent, or replace source manifests, metadata, backend schemas, capabilities, actions, surfaces, or runtime code.1112## Provenance Pipeline1314```text15component source, metadata, README/docs, authoring manifests, capability/context packs16 -> extract-component-docs.js + generate:ai-metadata17 -> dist/praxis-component-registry.json18 -> generate-registry-ingestion.ts19 -> dist/praxis-component-registry-ingestion.json (canonical aggregate corpus)20 -> catalog + authoring-contract validation21 -> optional RAG/provider/AI-ready/package/upload projections22```2324The ingestion corpus with `components[].chunks` is canonical for aggregate analysis. `praxis-component-registry-rag.json`, provider projections, AI-ready docs, reports, npm package assets, and backend uploads are derived read models. Never hand-edit `dist/**`, package assets, projections, or reports to repair a source/contract problem.2526Read `tools/ai-registry/AGENTS.md`, its README, workspace AGENTS, and the affected library's AGENTS before editing. Inspect the actual owner source: public API, component metadata, docs manifest, authoring manifest, capabilities/context pack, focused specs, and consumer docs.2728`generate-registry-ingestion.ts` projects `authoringManifest`, `authoringManifestProfiles`, capabilities, component context, AI concepts, and authoring-only manifest entries into the canonical corpus. This projection is evidence of declared Angular capability; it is not proof that the backend can execute every target resolver, validator, or effect handler.2930When backend registry persistence, bootstrap, classpath snapshot, template search/upsert, RAG document identity, or official upload behavior is in scope, also inspect `praxis-config-starter/AGENTS.md`, `RegistryIngestionController`, `RegistryIngestionService`, `RegistryIngestionRequest`, `AiRegistry`, `AiRegistryRepository`, `AiRegistryBootstrapService`, `AiRegistryStatusService`, `AiRegistryTemplateController`, `AiRegistryTemplateService`, and `src/main/resources/ai-registry/registry-snapshot.json`. The backend owns persisted registry identity, release-scoped RAG projection, snapshot bootstrap, template records, status/health, revision/ETag behavior, and rejection of invalid authoring manifests or presentation affordance catalogs.3132## Decide The Smallest Canonical Operation3334| Change | Owner and minimum action |35| --- | --- |36| component docs extraction or AI metadata | source docs/metadata first, then `npm run generate:registry:ingestion` |37| ingestion generator/schema/chunks/governance | `generate:registry:ingestion`; validate catalog and authoring contracts |38| catalog validation only | `npm run validate:catalog` |39| completed authoring contract acceptance | `npm run validate:authoring-contracts` |40| AI planning/RAG/templates | `npm run validate:ai` |41| package-scoped asset after ingestion and library builds | `npm run package:ai-assets`, then package/tarball validation if release scope requires it |42| provider projection | regenerate from the canonical ingestion corpus for the explicit release; never upload secrets/external IDs into it |43| backend upload | use `tools/ai-registry/run-all.sh` or `npm run ai:sync-backend` only when a real backend state change is required |4445`npm run generate:registry:ingestion` is the standard local pipeline: extract component docs, generate AI metadata, generate ingestion corpus, synchronize the Config classpath snapshot and its contract constants, validate catalog governance, and validate authoring contracts. It is the correct response to a source/manifest/capability/context-pack change; it is not required merely to edit a derived report or inspect a package asset.4647## Isolated Worktree Wiring4849The snapshot synchronizer and authoring acceptance validator use the existing50`PRAXIS_CONFIG_STARTER_ROOT` override, otherwise a sibling Config checkout. Before51running the pipeline in an isolated Angular worktree, point this override to the52explicitly assigned, clean Config worktree. Inspect both owners' instructions;53never direct generated writes into another worker's dirty primary checkout or54create an ad hoc sibling link. The sync writes both `registry-snapshot.json` and55`AiRegistrySnapshotContractTest` constants, so reserve both files for one writer.5657If generation succeeded but sibling discovery failed, preserve the generated58corpus and its provenance. Resume the failed sync/validation steps with the same59override instead of regenerating timestamps or weakening the validator. Review60the complete semantic snapshot delta against current Angular source: a stale61snapshot may include earlier canonical changes beyond the triggering patch.62Prove corpus equality, manifest preservation and the focused snapshot contract;63source synchronization does not establish backend upload or a public release.6465## Canonical Content Rules6667- Keep component docs extractable from supported literal metadata/factories. If extraction cannot represent a semantic feature, improve the source metadata or official extractor, not a consumer-side registry patch.68- A manifest, capability, target, validator, effect, profile, context pack, and docs chunk must point to the source owner and remain mutually consistent. Do not make a generated registry the primary definition of an operation.69- Treat `npm run validate:authoring-contracts` as the Angular acceptance gate for structural and semantic completeness: target kind/resolver, ambiguity policy, validators, effects, affected paths, destructive confirmation, submission impact, remote-binding evidence, and profile metadata. It does not replace backend execution tests for `AgenticAuthoringTargetResolverRegistry`, `AgenticAuthoringValidatorRegistry`, or `AgenticAuthoringEffectCompilerRegistry`.70- When a manifest introduces a new resolver id, validator id, effect kind, or `compile-domain-patch` handler, classify it as `suportado-parcialmente` until the backend registry implements it and focused config-starter tests prove the executable path. Do not upload or package the corpus as executable proof just because the generated Angular artifact contains the declaration.71- Registry projections must preserve the traceable authoring-manifest fields required for semantic decisions and backend validation: component/profile id, `manifestVersion`, stable `operationId`, `target.kind`, `target.resolver`, `target.ambiguityPolicy`, `target.required`, validator ids, effect `kind`/handler/handler contract, `affectedPaths`, `submissionImpact`, and presentation-affordance `sourceRef`, `defaultTargetKind`, and options. Do not collapse these into prose chunks, display labels, quick-reply text, or RAG-only summaries.72- For family components such as `praxis-dynamic-fields`, `authoringManifestProfiles` on child73 entries are a generated applicability projection. `generate-registry-ingestion.ts` matches control74 profiles from the family manifest to component docs through `meta.id`, selector, tags, or metadata75 input type. A projected child profile is useful evidence that the corpus preserved applicability,76 but it is not a source definition of the profile, not proof that a capability alias is a component77 identity, and not a substitute for runtime/editorial/catalog/source metadata evidence.78- Registry grounding is evidence for LLM semantic resolution. Do not encode primary intent as keyword, regex, alias, fuzzy match, display-label parsing, or a registry-only command dialect.79- Quick replies/actions must retain structured canonical action, semantic decision, target, context, risk, and presentation. Labels are not authority.80- If quick replies, semantic decisions, or component edit plans cite registry/RAG/provider projections, they must reference projected manifest operations and targets by stable ids. A projection that loses those ids is derived-artifact drift to repair at the source, not permission for backend, UI, or agents to fall back to text matching.81- Treat missing fields, components, manifests, operations, ports, presets, or chunks in generated registry artifacts as projection82 loss or source/extractor drift until proven otherwise. Absence from `component-docs.json`, the ingestion corpus, RAG/provider83 projections, package assets, or backend snapshots is not proof that the component, capability, authoring manifest, or backend84 support does not exist. Re-audit the owner source, provider registration, extractor support, generated artifact lineage, and85 backend snapshot before declaring a platform contract gap.86- Preserve the runtime observation contract chunk and its `untrusted_frontend_observation` boundary. Observations require backend grounding into canonical context before they influence an answer; registry text cannot grant read/tool/apply authority.87- Catalog governance must reject duplicate IDs/selectors, invalid paths, incomplete target/validator/effect evidence, unconfirmed destructive work, ambiguous targets without policy, inconsistent control profiles, and lost trust boundaries. Repair the source owner or validator, never weaken the gate to pass a local artifact.88- If a dynamic-fields child entry is missing `authoringManifestProfiles`, or has a profile that is89 not declared by the family manifest, first audit component docs extraction and the source90 `appliesTo` rules. Treat the mismatch as source/extractor/projection drift until proven otherwise;91 do not patch the generated corpus, add a registry-only profile, or promote a capability alias into92 `componentIds` just to satisfy the gate.93- Use aggregate corpus for cross-package/release analysis and `@praxisui/<package>/ai/component-registry.json` only for source-less package consumers. Package assets remain filtered projections of their owning package.94- Broad backend registry status, reconciliation, startup, and candidate scans must use compact repository projections. Do not hydrate embeddings, full chunk content, raw registry payloads, or large manifests for every row merely to compute identity, readiness, counts, hashes, or candidate summaries; hydrate the exact canonical record only when publication or selected-evidence materialization requires it.9596## Before New Contracts Or Scripts9798Inventory the existing metadata, `x-ui`, schemas, actions, surfaces, capabilities, option sources, manifests, registry chunks, diagnostics, context packs, docs, and backend API catalog. Classify the need as `ja-suportado-so-ux`, `ja-suportado-mal-nomeado-ou-mal-materializado`, `suportado-parcialmente`, or `lacuna-real-de-contrato`.99100Only a real contract gap justifies a schema, chunk, generator, endpoint, or script change. Map source owner, generated outputs, Angular/package consumers, backend ingestion, security/origin requirements, docs/playgrounds, compatibility, and focused validation first. Do not add ad hoc upload, extraction, or sync scripts outside this tool owner.101102## Remote And Release Boundaries103104- Local generation and validation come first. GitHub Actions are release/final gates, not discovery tooling.105- Backend uploads use the documented environment and official scripts: `BACKEND_URL`, `CONFIG_ORIGIN`, `APP_SECURITY_READ_OPEN`, and `APP_SECURITY_WRITE_OPEN`. Do not invent endpoints, origins, ports, or upload ordering.106- `run-all.sh --with-api-catalog` is the canonical combined registry/templates/API-catalog flow. Do not additionally invoke individual upload scripts in the same operation.107- `run-all.sh` generates AI-ready docs, runs `npm run generate:registry:ingestion`, optionally generates RAG, then uploads API catalog, registry, and templates through the official scripts. If `BACKEND_URL` is absent and upload is not explicitly skipped, it must fail instead of silently targeting a guessed host.108- `upload-registry.ts` posts the canonical `dist/praxis-component-registry-ingestion.json` to `/api/praxis/config/ai-registry/component-definitions` with `Origin` from `CONFIG_ORIGIN`, `ORIGIN`, or the documented `http://localhost:4003` fallback. Do not replace this with ad hoc curl snippets unless the task is only diagnosing the official script.109- `package:ai-assets` needs the generated ingestion corpus and built `dist/<package>` folders. It writes the scoped asset plus package export; validate local tarballs when package publication is in scope.110111## Evidence And Companion Skills112113Run the smallest compatible command and report it exactly. For changes to tooling itself, add its focused Node self-test; for a source component, add that component's build/spec/authoring proof. State whether docs, recipes, manifests, landing, package assets, RAG/provider projections, backend sync, and release validation were affected.114115For Angular authoring manifest or ingestion changes that may become executable through config-starter, pair the Angular gate with the backend manifest gate:116117```sh118npm run generate:registry:ingestion119mvn "-Dtest=AgenticAuthoringManifestServiceTest,AgenticAuthoringManifestControllerTest,AgenticAuthoringManifestContractValidatorTest,AgenticAuthoringTargetResolverRegistryTest,AgenticAuthoringValidatorRegistryTest,AgenticAuthoringEffectCompilerRegistryTest" test120```121122Use this paired proof when adding or renaming operation ids, target resolvers, validators, effect handlers, presentation affordance catalogs, or manifest profiles consumed by `/api/praxis/config/ai/authoring/manifests/{componentId}/**`.123124For backend registry ingestion, snapshot, template, bootstrap, or official upload contract changes, run the focused config-starter gate:125126```sh127mvn "-Dtest=RegistryIngestionServiceTest,RegistryIngestionServiceIdentityTest,AiRegistrySnapshotContractTest,AiRegistryBootstrapServiceTest,AiRegistryTemplateServiceTest,AiRegistryTemplateControllerTest,AiRegistryRevisionPolicyTest" test128```129130When `src/main/resources/ai-registry/registry-snapshot.json` is updated from the Angular ingestion corpus, prove both sides: `npm run generate:registry:ingestion` in `praxis-ui-angular`, then the backend snapshot gate above. Do not update the classpath snapshot hash, version, generated timestamp, chunk counts, or unsupported-validator expectations without regenerating from the canonical Angular corpus and validating the backend snapshot contract.131132Use `praxis-ai-authoring-manifests` for executable authoring contracts; `praxis-angular-docs-playgrounds` for public documentation; `praxis-angular-public-api-governance` for exports; `praxis-angular-validation-gates` for local proof; and the functional component skill for the changed source. Use `praxis-config-agentic-authoring-streaming` or `praxis-ai-turn-orchestration-transport` when registry chunks derive from governed turns, tools, observations, quick replies, or backend transport.