The Skills capability has two faces:

1. Knowledge skills — Markdown SOPs/runbooks the model can search semantically.
2. Code skills — Safe Python utilities (calculator, unit converter, current time, text analyzer) the model can call directly.

Both live under one ToolRegistry.

Knowledge skills — .claude/skills/<name>/SKILL.md

Format follows the Agent SDK convention: YAML frontmatter + Markdown body.

---
name: code-review
description: Comprehensive code review guidance — process, checklist, feedback conventions. Use when the user asks about code review flow, review criteria, best practices, or how to give/receive review feedback.
keywords: [code review, 代码审查, PR, pull request, review]
---

# 代码审查指南

## 适用场景
…

At startup SkillsService.load():

• Discovers every .claude/skills/<dir>/SKILL.md.
• Parses frontmatter (our hand-rolled parser needs no YAML dep).
• Indexes description + body into a dedicated Chroma collection (skills), so skill_search finds the best skill by semantic similarity.

Code skills

Live in datamind/capabilities/skills/code_skills.py:

Tool	Purpose
calculator	Safe eval against a pinned math namespace: sqrt sin cos tan log exp pow abs floor ceil pi e min max round sum. Rejects __ / import / exec / etc.
unit_convert	Table-driven length / mass / temperature conversions
get_current_time	Local date + weekday
analyze_text	Char / line / paragraph / word counts

Adding a new one is three lines:

async def _uuid4() -> dict:
    import uuid
    return {"uuid": str(uuid.uuid4())}

SPECS.append(ToolSpec(
    name="uuid4",
    description="Generate a random UUID v4.",
    input_schema={"type": "object", "properties": {}},
    handler=_uuid4,
    metadata={"group": "skill.code"},
))

Tools exposed to the agent

Tool	Purpose
skill_search	Semantic search across knowledge skills
skill_get	Return the full Markdown body for a named skill
skill_list	List every registered knowledge skill
calculator, unit_convert, get_current_time, analyze_text	Code skills

Example

python -m datamind.scripts.hello_skills

skill_search 'how should I review a pull request?'
  score=0.513  name=code-review
  score=0.194  name=db-ops-sop

skill_search '慢查询怎么排查'
  score=0.392  name=db-ops-sop
  score=0.330  name=code-review

Writing a new skill is one file with zero code changes:

mkdir -p .claude/skills/incident-response
cat > .claude/skills/incident-response/SKILL.md <<'EOF'
---
name: incident-response
description: Outage runbook — triage, comms, post-mortem. Use when the user asks about incidents or on-call duties.
keywords: [incident, outage, on-call, post-mortem]
---

# Incident Response

## Triage
…
EOF

The next python -m datamind chat will pick it up after calling agent.warmup().