Skip to content

v1.2.0

Latest

Choose a tag to compare

@github-actions github-actions released this 21 Jul 17:04
· 23 commits to main since this release
eee5418
  • New manifest: .kimi-plugin/plugin.json — Official Kimi Code plugin for Azure Cosmos DB, bundling the skills (./skills/) and the Azure MCP server (@azure/mcp). Mirrors the Codex interface block (display name, category, default prompts, brand color).

  • Version script: scripts/version.js now bumps .kimi-plugin/plugin.json alongside the other vendor manifests so npm run version stays atomic.

  • Manifest validation: Added scripts/validate-manifests.js (npm run validate:manifests) that checks every vendor manifest is valid JSON/YAML, version-aligned with package.json, and points skills at an existing path. Wired into the release workflow.

  • Docs: README surfaces list, install description, per-agent plugin-directory table, and compatibility line updated to include Kimi Code. README also adds a Kimi Code CLI install section covering direct GitHub install (/plugins install) and the custom marketplace catalog.

  • New catalog: kimi-marketplace.json — Kimi Code custom marketplace catalog (version: "2") listing the azure-cosmosdb plugin with its GitHub source, so users can add it with /plugins marketplace <url> or KIMI_CODE_PLUGIN_MARKETPLACE_URL. The top-level version is the Kimi marketplace schema version and is intentionally excluded from npm run version.

  • Note: Kimi's Official/curated marketplace tabs are maintained by Moonshot and have no self-serve submission; direct GitHub install and the custom marketplace JSON are the supported publishing paths from this repo.

  • New rule: sdk-go-partition-key-metadata.md — Instructs agents to avoid stale azcosmos pins, prefer current Go SDK releases, and create single-path partition keys with explicit metadata that interoperates with other Cosmos DB SDKs used by verifiers.

  • New eval task: sdk-go-partition-key-metadata.yaml — Covers the stale-azcosmos / missing-kind cross-SDK partition-key metadata failure mode.

  • New rule: pattern-langgraph-async-cosmos-routing.md — Wrap Cosmos DB sync calls in asyncio.to_thread for LangGraph routing functions (CRITICAL).

  • New rule: pattern-langgraph-async-cosmos-writes.md — Use asyncio.to_thread for active agent writes in async node functions (HIGH).

  • New rule: pattern-langgraph-agent-name-attribution.md — Tag AI messages with agent name for API response attribution (MEDIUM).

  • Amended: sdk-langchain-mcp-persistent-session.md — Updated for langchain-mcp-adapters >= 0.2.0 API (get_tools() replaces manual session management); kept < 0.2.0 example for backward reference.

  • Amended: pattern-langgraph-multi-agent.md — Added critical guidance to only check NEW messages (slice by existing count) to prevent infinite recursion loops.

  • Amended: pattern-langgraph-agent-routing-cosmosdb.md — Converted routing function to async with asyncio.to_thread, added 5s timeout, explicit exception handling for 404/timeout/credential errors.

  • Expanded README.md with a section describing the testing framework and how evaluations flow back into the skill.

  • Evaluated 5 Java iterations of the multitenant-saas scenario (testing-v2 batch #209).

  • Zero always-fail tests. When startup succeeded (3/5 iterations), pass rate was 100% across API contract, Cosmos infrastructure, and data integrity tests.

  • Identified a 40% Java startup-failure rate caused by Netty/OpenSSL behavior against the local emulator. Classified as an SDK/framework quirk, not a functional skill gap — sdk-emulator-ssl may benefit from a clearer programmatic Java CI bypass in a future pass.

  • No rules created or modified.

  • Added the docs/ agent-kit website with a GitHub-issue survey, Cosmos DB logo, Google Analytics, and metrics (70+ rules).

  • Added an "Integrations" section covering the MCP Server and Claude / Cursor plugins.

  • Converted the website submodule to a regular directory.

  • Added 4 new SDK rules (4.21–4.24).

  • Added a brand-new Full-Text Search category with 6 rules (12.1–12.6) covering the capability flag, fullTextPolicy, fullTextIndexes, BM25 ranking, keyword matching, and hybrid queries.

  • Skill now totals 89 rules across 12 categories.

  • Reverted explicit permissions: blocks that were breaking CI approval gates.

  • Fixed a race where post-create issue edits caused single-child failures in batch runs.

  • Fixed batch aggregation so parent issues resolve to the correct child PRs when iterating results.

  • Added intra-document pre-computed aggregates guidance to model-denormalize-reads.

  • Synced the skill index and added tooling rules covering build / validate scripts.

What's Changed

New Contributors

Full Changelog: v1.1.0...v1.2.0