# Agent Review

Critically review a Stencila agent and suggest improvements.

- Repository: https://github.com/stencila/stencila (914 stars)
- Category: AI and agents
- License: Apache-2.0
- Last updated: 2026-11-01
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add stencila/stencila --skill agent-review -g -a claude-code -y`
Codex: `npx skills add stencila/stencila --skill agent-review -g -a codex -y`
Cursor: `npx skills add stencila/stencila --skill agent-review -g -a cursor -y`
Gemini CLI: `npx skills add stencila/stencila --skill agent-review -g -a gemini-cli -y`
Claude Code plugin marketplace: `/plugin marketplace add stencila/stencila`

## Description

Critically review a Stencila agent and suggest improvements. Use when asked to review, audit, critique, evaluate, or improve an agent directory or AGENT.md file. Covers frontmatter validation, system instruction quality, configuration correctness, and adherence to the Agent schema.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

## Overview

Review an existing Stencila agent for quality, correctness, and completeness. Produce a structured report with specific, actionable suggestions. The review covers frontmatter fields, system instruction quality, configuration correctness, security posture, and adherence to the agent schema defined in `schema/Agent.yaml`.

## Steps

1. Identify the agent to review from the user's request, accept an agent name, a directory path, or an `AGENT.md` file path
2. Resolve the agent file: if given a name, look for `.stencila/agents/<name>/AGENT.md` walking up from the current directory; also check `~/.config/stencila/agents/<name>/AGENT.md` for user-level agents. If given a path, use it directly
3. Read the full `AGENT.md` file and any supporting files in the agent directory (check for `scripts/`, `references/`, and `assets/` subdirectories)
4. Read `schema/Agent.yaml` to verify the checklist covers current agent schema fields and to identify any unknown frontmatter properties
5. Evaluate the agent against each criterion in the Review Checklist below
6. Produce a structured review report with a summary, per-criterion findings, and a prioritized list of suggestions
7. If the user asks you to apply improvements, make the changes and validate the result with `stencila agents validate <agent-name>`

## Review Checklist

### Frontmatter

- **name**: present, matches directory name, valid kebab-case (`^[a-z0-9]([a-z0-9-]{0,62}[a-z0-9])?$`), follows the `thing-role` naming convention (e.g., `code-reviewer`, `data-analyst`)
- **description**: present, not empty, not a placeholder (`TODO`, `<placeholder>`), recommended to be concise (under ~1,024 characters), specific enough to convey the agent's purpose

### Optional Fields

Check each present field for validity:

The rest: https://github.com/stencila/stencila/blob/HEAD/.stencila/skills/agent-review/SKILL.md
