domains / general / discover / discover

discover

Skill card

combined

The two copies use the same skill name for completely different tasks. The first is a local CLI-based discovery tool that inventories AI agents and MCP server configurations on a machine. The second is a research workflow that routes to interview, literature review, data discovery, or ideation agents. They share no steps, constraints, or outputs, so crowning one would hide a real alternative.

Repositories

Other copies

Combined skill

# Discover

This skill has two distinct modes. Determine which mode the user wants and follow the matching procedure.

## Mode A: Local AI Agent and MCP Server Discovery

Use when the user asks to find agents, list MCP servers, check what is configured, run a doctor check, or show an MCP inventory.

1. Ensure the agent-bom CLI is available. Install with `pipx install agent-bom` if needed.
2. Run the requested command:
   - `agent-bom agents` to discover all AI agents and MCP servers.
   - `agent-bom doctor` to check prerequisites and configuration health.
   - `agent-bom mcp inventory` to list discovered MCP servers.
   - `agent-bom where` to show all discovery paths.
3. Present the output to the user.
4. Guardrails: discovery is read-only; environment variable values are redacted before processing; only structural data is extracted; confirm with the user before scanning paths outside the home directory.

## Mode B: Research Discovery Phase

Use when the user asks for a research interview, literature review, data discovery, or research ideation.

1. Parse the mode keyword from the arguments: `interview`, `lit`, `data`, or `ideate`. If no mode is given, default to `interview`.
2. Follow the corresponding workflow:
   - **interview**: Conduct a structured conversational interview in six sections (Big Picture, Theoretical Motivation, Data and Setting, Identification, Expected Results, Contribution). Ask one or two questions at a time and wait for the user's response. After 5-8 exchanges, produce a Research Specification at `quality_reports/research_spec_[topic].md` and, if the template is still empty, fill in `.claude/references/domain-profile.md`.
   - **lit**: Read the domain profile, check `master_supporting_docs/` and `bibliography_base.bib`, dispatch a Librarian agent to search top-5 journals, field journals, working papers, and citation chains, assign proximity scores 1-5, dispatch a librarian-critic to check coverage, re-dispatch once if gaps are found, and save the result to `quality_reports/lit_review_[topic].md`. Mark unverifiable citations with `% UNVERIFIED`.
   - **data**: Read the research spec and domain profile, identify needed variables, dispatch an Explorer agent to search public microdata, administrative data, survey data, international sources, and novel/alternative sources, dispatch an explorer-critic to assess feasibility, and produce ranked data sources with feasibility grades.
   - **ideate**: Follow the ideation workflow defined in the source skill.
3. Present the produced artifacts to the user.

## Conflicts

- LeoYeAI/openclaw-master-skills::skills/agent-bom/discover/SKILL.md defines discover as a local CLI inventory of AI agents and MCP servers.
- brycewang-stanford/Auto-Empirical-Research-Skills::skills/16-hsantanna88-clo-author/dot-claude/skills/discover/SKILL.md defines discover as a research phase that routes to interview, lit, data, or ideate modes.