v0.47.0

[Baziar]

Release Notes

Historical archive: workspai is the current package and command, with
wspai as an optional alias. Older entries intentionally preserve historical
rapidkit commands and .rapidkit paths. Use the CLI README and
Command Reference for current usage.

Latest Release: v0.47.0 (July 21, 2026)

Evidence-Backed Workspace Knowledge Graph

This minor release makes the current Workspace Model directly useful for
proof-backed retrieval across projects, code, packages, APIs, infrastructure,
delivery, documentation, decisions, tests, and ownership. The graph is derived
deterministically from the canonical model and bound to its exact SHA-256
revision.

What's New:

• Queryable workspace knowledge

  • Added bounded graph search, typed entity/evidence/path queries, complete
    portable emission, change/PR overlays, and retrieval payload benchmarks.
  • Added stable identities, proof paths, derivation/trust metadata, quality
    diagnostics, secret filtering, and deterministic query indexes.

• One graph revision for every consumer

  • The Model stage now publishes workspace-model.json and
    workspace-knowledge-graph.json atomically.
  • CLI, unified runner, context, Agent Sync, MCP, IDE capability discovery, CI,
    and workspace contract graph output consume the same model-bound revision.
  • Agents now start with bounded search and proof traversal instead of loading
    the complete graph into every prompt.

• Versioned public contracts

  • Added Knowledge Graph, change overlay, bounded search, and token-efficiency
    schemas.
  • Expanded the published catalog, artifact registry, runtime command surface,
    architecture contract, and semantic conformance gates.

• Clearer user documentation

  • Added a two-minute graph guide, reproducible benchmark methodology,
    Agent/IDE/MCP workflow, and plain-language glossary.
  • Separated deterministic Workspace Intelligence from the optional embedding
    module recommender and removed unqualified AI score/cost claims.

• Security and canonical paths

  • Updated vulnerable brace-expansion and fast-uri dependencies to their
    patched releases; the final online npm audit reports zero vulnerabilities.
  • Canonicalized bridged project metadata into .workspai while retaining
    legacy read compatibility.

Breaking changes: None.

Verification:

• 2,062 CLI tests passed with 8 explicit skips.
• All 11 runtime-conformance stages and required artifacts passed.
• Documentation links, examples, drift guard, README smoke, and live command
  surface parity passed.
• The measured 16-project fixture produced 1,738 entities, 2,244 relations, and
  2,106 portable proofs; fixture statistics are not universal product claims.

Upgrade:

npm install -g workspai@0.47.0
workspai workspace intelligence run --for-agent codex --strict --json
workspai workspace graph search "authentication endpoint" --limit 12 --json

Full Release Notes

---

Previous Release: v0.46.0 (July 18, 2026)

Contract-Backed Unified Workspace Intelligence Runner

This minor release turns the complete Workspace Intelligence lifecycle into a
single governed execution surface. Runtime behavior, ordered stage evidence,
artifacts, contracts, CLI help, CI guidance, and agent grounding now share the
same canonical source of truth.

What's New:

• Unified intelligence execution

  • Added workspace intelligence run for the mandatory sync-through-
    remediation lifecycle.
  • Added deterministic per-stage status, command, timing, artifact, and failure
    evidence for humans, CI systems, IDEs, portals, and agents.

• Semantic runner contract

  • Added the versioned workspai-workspace-intelligence-run-v1 report contract.
  • Linked the report through the runtime registry, artifact registry, published
    contract catalog, and producer-command inventory.
  • Added semantic enforcement for canonical stage ordering and blocking
    outcomes in addition to JSON Schema validation.

• Transactional project lifecycle

  • Added rollback protection for create, adopt, import, register, and mirror
    operations.
  • Hardened metadata path safety, registry consistency, archives, platform
    probes, runtime adapters, and polyglot project generators.
  • Added Windows-safe lifecycle journal durability handling and concurrent
    journal initialization recovery without weakening atomic writes.

• Release-gate integrity

  • Repaired the npm tree used by CycloneDX SBOM generation so Nunjucks, Tsup,
    ESLint, chokidar, and minimatch resolve without invalid nodes.
  • Preserved the 80% aggregate coverage target and made test, lint, and audit
    collection fail closed.
  • Expanded the suite to 2,042 passing tests, including targeted contract,
    Workspace Intelligence, Doctor, recovery, archive, runtime adapter, and
    platform-boundary coverage.
  • Verified 81.66% statement, 71.27% branch, 91.95% function, and 82.38% line
    coverage; the official metrics gate reports 82%.
  • Made test:coverage publish a machine-readable Vitest result for metrics,
    removing the duplicate suite execution and fragile console parsing that
    could report a passing run as invalid.
  • Hardened Linux, macOS, and Windows parity for project-local Python venvs,
    pip metadata, Doctor PowerShell remediation guards, Go adapter Makefile
    routing, npm runner paths and environment keys, and filesystem mode tests.

• Documentation and runtime alignment

  • Aligned CLI documentation, help, contracts, artifact catalogs, CI examples,
    and agent-grounding material around the mandatory lifecycle.
  • Removed accidental Python-version pins from config examples. Python 3.10
    remains the supported minimum, with automatic newer-version detection and
    optional explicit pins for reproducible CI.

Breaking changes: None.

Verification:

• Monorepo checks and CLI build/test gates.
• Generated/shared contract drift and documentation validation.
• Unified runner contract and CLI-chain suites.
• Lifecycle transaction, frontend execution, Python-engine state, registry,
  mirror, archive, runtime adapter, and package-publish suites.
• 2,042 tests passed, 8 tests were explicitly skipped, ESLint reported zero
  errors or warnings, and npm audit reported zero vulnerabilities.

Upgrade:

npm install -g workspai@0.46.0
workspai workspace intelligence run --for-agent codex --json

Full Release Notes

---

This discussion was created from the release v0.47.0.