v0.5.2

agent-runtime-kit 0.5.2 refreshes the supported SDK baselines and fixes Antigravity permission and Vertex Express authentication behavior. Existing public APIs remain compatible.

Tested SDK versions

Package	0.5.1 baseline	0.5.2 baseline
Claude Agent SDK	0.2.148	0.2.157
Codex SDK	0.147.0	0.154.0
Coupled Codex CLI	0.147.0	0.154.0
Google Antigravity SDK	0.1.15	0.1.17

The Codex dependency cap is now <0.155; the SDK continues to select its exact CLI version. Provider packages remain optional extras, and Python 3.10+ remains supported.

Changed

• Updated tested Claude Agent SDK to 0.2.157, Google Antigravity SDK to
  0.1.17, and the Codex SDK and its exactly coupled CLI to 0.154.0.
• SDK evolution supports explicit model, reasoning effort and analysis executable
  options, preserves older locked baselines in implementation history, and
  constructs real Claude options and Codex wire parameters in candidate probes.
• Upgrade skill invocations now include necessary runtime adaptations,
  verification, and a regular pull request. Explicit report-only and local-only
  requests retain their narrower scope.

Fixed

• Antigravity permissive deny-lists preserve all unrelated tools when the SDK's
  default tool set changes, including the ask_question change in 0.1.17.
• Codex release evidence uses the current official documentation endpoints.
• Antigravity preserves explicit vertex=True with an API key for Vertex
  Express authentication, rejects mixed regional credentials, and refuses older
  SDKs or configurations that would silently drop the requested endpoint.
• Release-note collection recognizes versions present only in a GitHub
  discussion title and rejects unrelated or longer version-prefix matches.
• PyPI publication validates and uploads Core Metadata 2.5 distributions with
  matching Twine support.

Validation and scope

The upgrade passed 553 local tests with 90.36% package coverage, minimum-dependency checks, Ruff and strict mypy. Live Codex checks covered structured output, independent task sessions, session resume, process reuse and cleanup. Claude and Antigravity validation covered local configurations and adapter contracts; their live services were not exercised.

See the SDK compatibility review for upstream impact decisions and evidence limits.

Included changes: #58, #59, #60.

Full changelog: v0.5.1...v0.5.2

v0.5.1

agent-runtime-kit 0.5.1 upgrades all supported vendor SDK baselines and makes the SDK evolution workflow evidence-complete, no-cooloff, and publication-safe.

Added

• SDK evolution reports now include recent-release implementation evidence:
  shipped-file fingerprints, Python-definition diffs, opaque-runtime limits,
  and implementation trends grounded in observed upstream artifacts.
• Exact baseline/candidate behavior probes now exercise adapter-owned semantic
  contracts in reusable, credential-scrubbed disposable environments.
• Evolution artifacts now preserve independent upstream discovery, current and
  prospective resolver results, exact candidate classifications, implementation
  diffs, and stronger current-state provenance.

Changed

• Updated tested vendor runtimes from Claude Agent SDK 0.2.106 to 0.2.148,
  OpenAI Codex SDK 0.1.0b3 to 0.147.0, its coupled CLI binary 0.137.0a4
  to 0.147.0, and Google Antigravity SDK 0.1.4 to 0.1.15.
• Raised the validated Codex SDK range to <0.148; the standalone Codex CLI
  0.149.0 remains an inspected artifact rather than an install candidate
  because Codex SDK 0.147.0 requires CLI 0.147.0 exactly.
• SDK freshness checks now bypass repository and ambient UV release cooloffs so
  upstream discovery, prospective resolution, and applied locks describe the
  same current package state.
• Direction-of-travel analysis now reports trends in actual upstream
  implementations instead of emitting dependency upgrade or hold advice.
• The repository-owned upgrade skill is the canonical operator workflow for
  report-first discovery, gated implementation, and separately authorized PR
  publication.

Fixed

• Candidate discovery no longer treats the current dependency bounds as proof
  that no newer upstream release exists; excluded releases receive an
  independent prospective resolver check.
• Missing optional SDKs, failed imports, skipped probes, malformed evidence, and
  mismatched baseline/candidate versions can no longer produce a false-green
  compatibility result or bypass implementation gates.
• Dependency application now updates project constraints, uv.lock, and the
  compatibility manifest atomically, verifies exact inspected versions, and
  restores all three artifacts if resolution or validation fails.
• Antigravity now maps short public session IDs deterministically to
  provider-safe UUID conversation IDs while preserving the caller-visible
  session identifier.
• Codex CLI binary snapshots use the correct import surface, and publication is
  skipped for empty, unapplied, rolled-back, or unverified changes.
• Candidate installation and release-note collection now retry bounded transient
  failures and report progress instead of appearing stalled.

Upgrade PR: #57

Full changelog: v0.5.0...v0.5.1

v0.5.0

What's Changed

• fix: enforce AgentKit result and lifecycle invariants by @ebarti in #41
• fix: verify required vendor SDK options by @ebarti in #42
• feat: enforce strict structured output primitives by @ebarti in #43
• fix: validate structured output across providers by @ebarti in #44
• feat: enforce runtime domain invariants by @ebarti in #45
• feat: add task compatibility preflight by @ebarti in #46
• feat: add bounded runtime readiness probes by @ebarti in #47
• fix: delegate default model selection to providers by @ebarti in #48
• feat: add task deadlines and cancellation by @ebarti in #49
• release: prepare v0.5.0 by @ebarti in #50

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

• feat: AgentKit — FastAPI-style ergonomics layer (0.4.0) by @ebarti in #39

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• Add SDK evolution upgrade commands by @ebarti in #16
• Add Codex SDK evolution auth helper by @ebarti in #17
• Add reusable SDK evolution upgrade script by @ebarti in #18
• Clarify provider install docs and package metadata by @ebarti in #15
• Fix stack 04: consistent finish_reason for structured/empty/truncated results by @ebarti in #28
• Add vendor SDK process reuse by @ebarti in #23
• [codex] Harden adapter layer: dedup helpers, reuse isolation, coverage gate by @ebarti in #24
• Fix stack 01: ignore secrets/reports, untrack committed run artifacts by @ebarti in #25
• Fix stack 02: evict reused client on cancellation + aclose/run-lock race by @ebarti in #26
• Fix stack 03: redact camelCase secrets, bound event sanitizer by @ebarti in #27
• Fix stack 05: monotonic permission ladder, honor READ_ONLY, Antigravity posture by @ebarti in #29
• Fix stack 06: per-turn Codex usage + session_id fallback parity by @ebarti in #30
• Fix stack 07: tool-audit and tool-event parity by @ebarti in #31
• Fix stack 08: uniform vendor-drift tolerance + off-loop Antigravity auth by @ebarti in #32
• Fix stack 09 (breaking): open runtime-kind + async lifecycle in protocol by @ebarti in #33
• Fix stack 10: typed model/reasoning_effort fields + FinishReason enum by @ebarti in #34
• Fix stack 11: freeze mapping fields; drop dead sdk_turn_limit by @ebarti in #35
• Fix stack 12: bound pre-1.0 SDKs, test real resolution, harden publish by @ebarti in #36
• Fix stack 13: correct doc drift + add API stability policy by @ebarti in #37
• Fix stack 14: example agent — opt-in/scrubbed candidate install, --draft-pr fix by @ebarti in #38

Full Changelog: v0.2.0...v0.3.0

agent-runtime-kit 0.2.0

Added

• Local SDK evolution agent example with runtime-backed direction analysis, architecture decision, reviewer stages, reports, and optional draft PR creation.
• Release-note collection, candidate API snapshots/diffs, behavior probes, current-state promotion manifests, and deterministic implementation gates for SDK update runs.
• Documentation for running the SDK evolution agent locally with supported vendor credentials and reviewing generated evidence.

Changed

• Updated resolver-selected lock entries for Claude and Antigravity after adapter-contract validation: claude-agent-sdk 0.2.106 and google-antigravity 0.1.4.
• Current-state manifests now record repo-relative checked-in artifact paths instead of machine-local absolute paths.

v0.1.1

Hardening release driven by a full-codebase review, with every adapter change verified against the real vendor SDKs.

Fixed

• Codex approval-mode mapping was inverted: STRICT/CAUTIOUS now map to deny_all (never escalate beyond the sandbox); DEFAULT/PERMISSIVE map to auto_review. Previously strict callers could have escalations auto-approved and permissive callers had them denied.
• Antigravity MCP stdio configuration no longer crashes (required name now passed).
• Codex honors TurnResult.status/error — failed and interrupted turns are no longer reported as success.
• Claude reports finish_reason="max_turns" on max-turns truncation.
• Strict mypy passes identically with and without vendor extras.

Changed (behavior)

• Antigravity DEFAULT posture is safer: nondestructive toolset instead of all tools + allow-all. STRICT/read-only → read-only toolset; PERMISSIVE keeps all tools.
• Previously silently-ignored task fields now raise UnsupportedTaskInputError (codex: tool filters/budget/network; antigravity: budget/network/allow+deny combo; claude: network).
• Antigravity session data moved from /tmp to $XDG_CACHE_HOME/agent-runtime-kit (0700), overridable via data_dir=.
• Missing SDKs/credentials raise typed AgentRuntimeUnavailableError.
• codex extra now requires openai-codex>=0.1.0b3 so it installs on glibc Linux.

Added

• Real incremental streaming events for Claude; tool audits for Codex parsed from TurnResult.items.
• Dropped vendor kwargs surfaced in AgentResult.metadata["dropped_options"].
• CI matrix (Python 3.10–3.13 × core/all-extras); publishing now gates on tests.
• Real-SDK contract tests; sdist allowlist (no internal artifacts shipped).

Full details in CHANGELOG.md.

v0.1.0

First release of agent-runtime-kit.

Highlights:

• Dependency-free core package with typed AgentTask, AgentResult, runtime protocol, registry, fake runtime, and availability diagnostics.
• Normalized task/output/tool/vendor event helpers with redaction and truncation defaults.
• Optional Claude, Codex, and Antigravity runtime adapters with lazy imports and explicit capability handling.
• Fake SDK harness and credential-free adapter tests.
• Same-task multi-runtime example, capability matrix, live smoke test guide, and Mestre migration notes.

Verification:

• uv run pytest: 22 passed, 3 live provider smokes skipped by design.
• uv run ruff check .: passed.
• uv run mypy: passed.
• uv run python -m build: wheel and sdist built.

PyPI publish notes:

• Package name: agent-runtime-kit.
• Version: 0.1.0.
• PyPI name check returned FREE immediately before this release, but recheck once more right before publishing.