# Investigate

> Investigate bugs by spawning parallel agents to explore multiple hypotheses

- Skill: `rilldata/investigate` (Agent Skill)
- Install (CLI): `npx skillmds@latest add rilldata/investigate`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rilldata/investigate/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: rilldata (https://skillmd.com/u/rilldata)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rilldata/investigate

---


Investigate a bug by gathering context, then spawning parallel Task agents to test multiple hypotheses simultaneously.

Input: $ARGUMENTS

## Instructions

### 1. Gather Context

Parse the input for links and fetch context:

- **Slack link** (contains `slack.com/archives`): Fetch the message and thread replies via Slack MCP tools
- **Linear issue** (ID like `ENG-1234` or URL): Fetch via `mcp__linear__get_issue`, including comments and attachments
- **Free text**: Use directly as the problem statement

User-provided hunches should be prioritized when forming hypotheses.

### 2. Investigate in Parallel

Generate 3-5 hypotheses, then use the **Task tool** to spawn one agent per hypothesis **in a single message** (parallel, not sequential). For each agent:

- `subagent_type: "Explore"`, `model: "sonnet"`
- A focused prompt with the hypothesis, what to look for, and what would confirm or refute it

### 3. Synthesize

After all agents complete, report: which hypotheses were confirmed, refuted, or inconclusive, and recommend next steps.