NCCR Data Request Skill — Installation

A skill that helps a researcher draft and validate an NCCR Data Request: the narrative sections, Research Areas, collaborator rules, and data element selection.

It is a single portable file, SKILL.md, with YAML frontmatter (name, description). Most agent frameworks that support skills read that format.

What it pairs with

The skill supplies judgment (how to write a good request, the rules, the prerequisites). The MCP server in ../../mcp/ supplies facts (form fields, character limits, research areas, available data elements, draft validation).

Use both together for the full experience. The skill is still useful on its own.

Relevant MCP tools: list_request_fields, list_research_areas, validate_request_draft, list_data_elements.

Install

Kiro

Copy the skill folder into either location:

# workspace-level (this project only)
mkdir -p .kiro/skills
cp -R skills/nccr-data-request .kiro/skills/

# or user-level (available in every workspace)
mkdir -p ~/.kiro/skills
cp -R skills/nccr-data-request ~/.kiro/skills/

Kiro picks it up on the next session. Ask it something like “help me write an NCCR data request” and it will activate.

Claude Code

mkdir -p ~/.claude/skills
cp -R skills/nccr-data-request ~/.claude/skills/

(Use .claude/skills/ inside a project for project-scoped installation.)

Claude Desktop / claude.ai

Upload or add SKILL.md through the app’s Skills interface (naming and location vary by version — see Anthropic’s current Skills documentation).

Strands Harness / other agent frameworks

Strands Harness can load pre-existing skills; point it at this folder. For frameworks without a formal skill mechanism, include the contents of SKILL.md in the system prompt or supply it as context.

No skill support at all

Paste the contents of SKILL.md into the conversation before you start. It is written to work that way too.

Verify it’s working

Ask your assistant:

“I need to request NCCR data for a study of late cardiac effects in AYA CNS tumor survivors.”

You should see it:

raise prerequisites (IRB, SEER Research Plus accounts, CCDI mappings approval PDF),
warn that four fields are publicly displayed before drafting them,
report character counts against the 1500-character limits,
explain that external collaborators must file their own referencing request.

If it skips the public-display warning, the skill probably isn’t loaded.

Keeping it current

The form’s fields and limits are also described structurally in ../../request-form.ttl. If the live form changes, update both that file and SKILL.md.

“Recommended” means two different things, and only one of them is metadata:

Baseline defaults the form selects and requires whenever a given source is in the request, regardless of the science. Recorded in request-form.ttl and reported by list_data_elements. Confirmed for CTC (6 elements, always present because CTC is in every request) and CCDI Mappings (2 elements, when that source is requested). Other sources report none until their defaults are confirmed.
Aim-specific recommendations that NCCR staff raise when they review a submitted request, flagging variables the research aims imply but the request left out. That is per-request judgement, so it isn’t modelled as metadata. SKILL.md instructs the assistant to run the same coverage check while the draft is still editable.