sci-tools

name

sci-tools

description

Browse and search the Harvard ToolUniverse catalog of 2,200+ biomedical tools from the CLI, and create custom scientific skills with validated templates. Two modes: browse (search, filter, drill-down on biomedical tools) and create (generate sci-* skills from natural language descriptions with scientific defaults). Triggers on: "tools for", "find tools", "search tools", "browse tools", "ToolUniverse", "biomedical tools", "what tools exist for", "tool catalog", "refresh tools", "update catalog", "create a research skill", "build a science skill", "new scientific skill", "custom research skill". Does NOT trigger for: general "create a skill" without science context (use meta-skill-creator), data analysis (use sci-data-analysis), literature search (use sci-literature-research).

Tool Ecosystem

Outcome

Browse 2,200+ biomedical tools from Harvard's ToolUniverse catalog with instant local search, or create custom sci-* skills from natural language descriptions. Browse results display as compact tables (per D-03). Custom skills are validated and auto-registered into the ecosystem.

Context Needs

File	Load level	Purpose
research_context/research-profile.md	full	Field context for skill creation personalization
context/learnings.md	## sci-tools section	Previous feedback

Dependencies

Skill	Required?	What it provides	Without it
meta-skill-creator	Required (for create mode)	Full skill creation workflow	Browse mode still works; create mode unavailable

Step 0: Auto-Setup

Run .claude/skills/sci-tools/scripts/setup.sh if first invocation. Downloads the ToolUniverse catalog snapshot to data/tooluniverse-catalog.json if not present.

Step 1: Detect Intent

Parse user request into mode:

• browse -- "tools for", "find tools", "search tools", "browse tools", "ToolUniverse", "biomedical tools", "what tools exist for", "tool catalog", "show categories", "list categories"
• refresh -- "refresh tools", "update catalog", "re-download catalog"
• create -- "create a research skill", "build a science skill", "new scientific skill", "custom research skill", "I need a skill that" (in science context)

If ambiguous, ask which mode.

Step 2: Browse Mode (TOOL-01, TOOL-04)

1. Run catalog search using catalog_ops.search_catalog(query, category=category, limit=20).
2. If user specified a category, use it. If category doesn't match exactly, use fuzzy matching (substring). If no results, show available categories via catalog_ops.list_categories().
3. Display results using catalog_ops.format_results_table(results) -- compact table with Name, Category, Type, Description (per D-03).
4. Show result count and "last refreshed" date from catalog metadata.
5. Ask: "Want details on any of these tools? Name one to drill down."

Step 3: Drill-Down (TOOL-04)

1. Run catalog_ops.get_tool_details(tool_name) to get full schema via tu info --json.
2. Display: full description, parameters (name, type, required), return schema, test examples, usage notes.
3. If MCP tools available (mcp__tooluniverse__get_tool_info), use those as alternative for richer documentation.

Step 4: Refresh Mode (TOOL-01)

1. Run catalog_ops.refresh_catalog().
2. Report: "Catalog refreshed: {total_tools} tools. Last update: {refreshed_at}."
3. If refresh fails (network error, uvx unavailable), report error and note catalog still usable from last snapshot.

Step 5: Create Mode (TOOL-03, TOOL-05)

1. Load research_context/research-profile.md for field context.
2. Get scientific defaults from validate_ops.get_scientific_defaults():
   • Pre-fill category prefix: sci-
   • Pre-fill output path: projects/sci-{skill-name}/
   • Pre-fill context needs: research-profile.md (full)
   • Include reproducibility logging step
3. Read .claude/skills/sci-tools/references/sci-skill-template.md for the template structure.
4. Read .claude/skills/meta-skill-creator/SKILL.md and follow its methodology starting from "Capture Intent", but with these pre-filled values as defaults. The scientist only needs to describe what the skill should do -- scientific conventions are assumed (per D-06, D-07).
5. After scaffold is generated: offer to test (per D-08). Scientist can say "looks good" to skip eval, or provide test prompts to iterate.

Step 6: Validate Created Skill (TOOL-05)

1. Run validate_ops.validate_skill(skill_dir). Report errors and warnings.
2. Run validate_ops.check_trigger_conflicts(skill_dir). If conflicts found, show them with suggestions (per D-11): "Trigger phrase '{phrase}' conflicts with {existing_skill}. Suggested alternative: '{suggestion}'." Non-blocking -- scientist decides.
3. If no errors: skill is ready. Heartbeat reconciliation will auto-register it in CLAUDE.md on next session (per D-10).
4. If errors: show them and offer to fix.

Step 7: Feedback

Ask: "How did this land? Any adjustments?" Log feedback to context/learnings.md under ## sci-tools.

Rules

• Always run setup.sh before first catalog search
• Always show compact table format for browse results (per D-03)
• Always use local catalog for search (per D-01), MCP/CLI for drill-down only (per D-02)
• Never duplicate meta-skill-creator logic -- delegate to it (per D-06)
• Always pre-fill scientific defaults for create mode (per D-07)
• Always run validation after skill creation (per D-09)
• Trigger conflicts are warnings, not blockers (per D-11)
• SkillsMP marketplace is deferred to v2 (per D-05). If asked, explain it's planned for a future version.

---

Self-Update

If the user flags an issue, update the ## Rules section with the correction and today's date.