The Glue Discovery Search operation was renamed to SearchAssets (POST
/search-assets) after launch; also corrects the --filter-clause JSON to
PascalCase and the documented request/response field casing to match the API,
in finding-data-lake-assets and exploring-data-catalog (both the plugin and
specialized-skills copies).
Adds the amazon-opensearch-service skill to the aws-data-analytics
plugin as a copied skill, matching the format of the plugin's other
seven skills. Covers migration (Solr/Elasticsearch), provisioning,
vector/semantic/hybrid search, and log + trace analytics.
- Copies the skill from skills/specialized-skills/analytics-skills/
amazon-opensearch-service into the plugin (real files, no symlinks)
- Expands description and keywords across the Claude, Cursor, and Codex
manifests; adds an OpenSearch defaultPrompt to the Codex manifest
- Bumps aws-data-analytics to v1.1.0
- Updates the plugin README
Signed-off-by: Andre Kurait <andrekurait@gmail.com>
Co-authored-by: Roman Shevchuk <51510790+rshevchuk-git@users.noreply.github.com>
Adds three read-only system-table skills (querying-aws-s3,
querying-aws-cloudwatch, querying-aws-sagemaker-catalog) for SQL over S3
Metadata/Storage Lens, CloudWatch Logs S3 Tables, and SageMaker Catalog asset
metadata, under skills/specialized-skills/system-table-skills. Also wires Glue
Discovery Search/GetAsset/ListIterableForms as a Step 2 'Consult Catalog
Context' lookup in finding-data-lake-assets and exploring-data-catalog
(advisory, gated on CLI availability + user opt-in), updated in both the plugin
and specialized-skills locations.
Cursor's plugin system (github.com/cursor/plugins) reads a root
.cursor-plugin/marketplace.json that indexes plugins, each with its
own .cursor-plugin/plugin.json — the same multi-plugin layout this
repo already uses for Claude Code (.claude-plugin) and Codex
(.codex-plugin). Add Cursor as a third supported agent:
- .cursor-plugin/marketplace.json indexing all three plugins, and a
.cursor-plugin/plugin.json in each of aws-core, aws-agents, and
aws-data-analytics. Descriptions and keywords enumerate the
bundled-skill domains (matching the enriched Claude/Codex
manifests); skills point at ./skills/ and mcpServers reuses the
existing ./.mcp.json, so no config is duplicated. OpenSearch
keywords are included on aws-core (Bedrock Knowledge Bases support
opensearch_serverless / opensearch_managed_cluster vector stores)
but not on aws-agents or aws-data-analytics, neither of which
bundles a skill that operates OpenSearch.
- README: add a Cursor quick-start section (add the repo as a team
marketplace via Settings -> Plugins -> Import from Repo) and list
Cursor alongside Claude Code, Codex, and Kiro.
- tools/validate.py: validate the Cursor marketplace index and each
.cursor-plugin/plugin.json, mirroring the existing Claude/Codex
checks (name present and matching the plugin directory).
Manifests validate against Cursor's published marketplace.schema.json
and plugin.schema.json (ajv). python3 tools/validate.py, markdownlint,
and gitleaks all pass.
Signed-off-by: Andre Kurait <akurait@amazon.com>
Each plugin's bundled skills cover far more than its plugin.json
metadata advertised. The Claude Code spec lists the description and
the keywords array as the discovery surfaces for the /plugin picker
and marketplace search, but the manifests listed only a handful of
keywords each, several Codex longDescription fields lagged Claude,
and the .claude-plugin/marketplace.json index still carried the old
one-line descriptions with a bare ["aws"] keyword list for aws-core
and aws-agents.
Surgically extend each plugin's description to enumerate the
bundled-skill domains while preserving the original phrasing, and
broaden the keywords arrays to cover the services and triggers the
underlying skills actually exercise:
- aws-core: IAM, Amazon Bedrock with Knowledge Bases / RAG /
Guardrails, Amplify (Cognito, AppSync), the SDK families
(boto3, JS v3, Swift), CDK/CloudFormation/SAM, Lambda with
SnapStart / Powertools / durable functions, API Gateway,
Step Functions, ECS/Fargate/ECR, App Runner, the messaging and
streaming services (SQS, SNS, EventBridge, Kinesis, MSK, Flink,
Amazon MQ), observability (CloudWatch, X-Ray, CloudTrail, ADOT),
and cost optimization (budgets, reserved instances, right-sizing).
- aws-agents: Strands/LangGraph, Gateway and MCP tool integration,
multi-agent and A2A orchestration, Cedar policies, evaluation and
observability, and production hardening (inbound auth via JWT and
SigV4, OAuth/OpenAPI outbound, code interpreter, browser tool,
rate limiting).
- aws-data-analytics: the JDBC engines, Redshift, Snowflake,
BigQuery, RDS/Aurora, DynamoDB, Glue Data Catalog, federated
Athena, and S3 Vectors semantic search.
Add OpenSearch keywords (opensearch, opensearch-serverless,
amazon-opensearch-service) and a brief longDescription mention to
aws-core, where Bedrock Knowledge Bases support opensearch_serverless
and opensearch_managed_cluster vector stores. The aws-agents and
aws-data-analytics plugins are left without OpenSearch metadata —
neither bundles a skill that operates OpenSearch.
Mirror every change across the .claude-plugin and .codex-plugin
manifests, and sync the same descriptions and keywords into
.claude-plugin/marketplace.json so all discovery surfaces agree.
python3 tools/validate.py and gitleaks both pass; no markdown
files changed.
Signed-off-by: Andre Kurait <akurait@amazon.com>
Deletes the find-aws-skills skill and all references to it:
- Removes skills/find-aws-skills/
- Removes plugins/aws-core/skills/find-aws-skills/
- Removes the skill row from plugins/aws-core/README.md
- Removes the now-empty Skills section from plugins/aws-agents/README.md
- Removes the now-empty Skills section from plugins/aws-data-analytics/README.md
Rohit Srikanta