Motivation

• The Exa research templates had drifted from the current Exa API/SDK semantics and could send incorrect defaults or unsafe request data (unencoded IDs and unset query params).

Description

• Changed the research template default model from exa-research to exa-research-fast to match the current Exa client default in packages/tracecat-registry/tracecat_registry/templates/tools/exa/research.yml.
• Updated get_research to reflect single-fetch semantics (removed wording about automatic polling) and URL-encode the research_id in the path using FN.url_encode in packages/tracecat-registry/tracecat_registry/templates/tools/exa/get_research.yml.
• Modified list_research to dynamically build the params object via a core.script.run_python step so optional limit/cursor values are omitted when unset in packages/tracecat-registry/tracecat_registry/templates/tools/exa/list_research.yml.

Testing

• Parsed all Exa template YAML files successfully using a small script (python -c "import yaml; ...") which returned ok.
• Ran the targeted registry test command uv run pytest tests/registry -k exa -q, which could not complete in this environment because PostgreSQL is not available on localhost:5432, causing test setup to fail before template assertions.
• Verified the updated YAML files load and that template changes align with the current exa-py client behavior inspected from the official SDK sources (used for API semantics reference).

Codex Task

Summary by cubic

Aligns Exa research templates with current API behavior to prevent incorrect defaults and unsafe requests. Updates the default model, request encoding, and pagination params to match the latest SDK.

• Bug Fixes
  • Set the research template default model to exa-research-fast to match the Exa client.
  • get_research: clarify single-fetch behavior and URL-encode research_id in the path using FN.url_encode.
  • list_research: build params dynamically so limit/cursor are omitted when unset.

^{Written for commit dc53db4. Summary will update on new commits.}