Skip to content

feat(ai): add model compatibility matrix and skill runtime - #398

Merged
szmyty merged 1 commit into
mainfrom
feat/ai-skill-runtime-396
Sep 11, 2026
Merged

szmyty merged 1 commit into
mainfrom
feat/ai-skill-runtime-396

Conversation

@szmyty

@szmyty szmyty commented Sep 11, 2026

Copy link
Copy Markdown
Contributor

Summary

  • add the versioned renderflow.ai-model-catalog/v1 registry and JSON Schema with model-specific modalities, operations, limits, locality, availability, determinism, licensing, advisory evidence, and provenance requirements
  • add deterministic skill-to-model resolution with explicit rejection/unavailability reasons, local-only guarantees, and separate request-level versus skill-policy remote gates
  • add the versioned renderflow.ai-skill/v1 contract, conservative JSON Schema validator, four redistribution-safe proving skills, and a digest-pinned synthetic visual fixture
  • add an artifact-native AiSkillRuntime that validates inputs, applies pre/post privacy and protected-reference hygiene, sends bounded schema-aware provider requests, validates structured output, and records review-required candidate artifacts with redacted renderflow.ai-execution/v1 evidence
  • add renderflow ai matrix, renderflow ai resolve, and renderflow ai skills list|inspect|validate human/JSON/YAML surfaces
  • document local-first operation, extension points, caching/resume identity, rights boundaries, downstream use by Add optional modality-aware artifact DNA extraction and similarity guidance #387/Add a rights-aware coloring-book publication profile and line-art pipeline #348, and probabilistic rollback/reproduction expectations

Safety and scope

  • no model weights, paid API calls, or protected third-party assets are bundled
  • remote execution remains explicit and opt-in; a local-only request cannot select a remote model
  • catalog availability is reported honestly as unverified until bounded runtime discovery supplies evidence
  • AI outputs are always stored as candidates requiring validation/approval, never silently canonical
  • raw private inputs and prompts are represented by digests in durable evidence

Focused validation

  • PATH="/root/.cargo/bin:$PATH" cargo check --workspace --all-features --locked
  • PATH="/root/.cargo/bin:$PATH" cargo clippy --workspace --lib --bins --all-features --locked -- -D warnings -A clippy::too_many_arguments
  • PATH="/root/.cargo/bin:$PATH" cargo test --package renderflow --lib "ai::" --locked — 110 passed
  • focused catalog, skill, and runtime test filters — 3 + 4 + 2 passed
  • JSON parsing for all three published schemas and all bundled catalog/skill assets
  • CLI smoke tests for matrix output, skill validation, local-only resolution, explicit remote opt-in, and skill-policy remote rejection
  • git diff --check

Deferred for this sprint

The full repository test suite and GitHub Actions were intentionally deferred per the speed-run instructions. No claim is made that the full suite or CI was run.

Closes #396

@szmyty
szmyty merged commit 5efe903 into main Sep 11, 2026
15 of 18 checks passed
@szmyty
szmyty deleted the feat/ai-skill-runtime-396 branch September 11, 2026 20:46
@devactivity-app

Copy link
Copy Markdown

Pull Request Summary by devActivity

Metrics

Cycle Time: 7m

Achievements

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add a model compatibility matrix and spec-driven AI skill runtime

1 participant