Skip to content

Latest commit

 

History

47 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mistral Agents

Mistral Le Chat agent prompts and Mistral Vibe skills for WordPress development.

This repository holds two kinds of reusable AI instructions:

  • Le Chat agents — system prompts you paste into a Mistral Le Chat agent to get consistent output for a specific role, such as building a WordPress page section or reviewing its HTML/CSS.
  • Vibe skillsAgent Skills for Mistral Vibe that run inside your project, read its files, run git and linters, and report — such as /code-review.

Imagewize created these for its own WordPress work — the Elayne block theme, the Nynaeve Sage theme and the Aludra block plugin — but they are written to adapt to other projects. Use as-is, extend, or fork.


📂 Repository Structure

mistral-agents/
├── AGENTS.md                # Instructions for coding agents working in this repo
├── .vibe/
│   └── config.toml          # Vibe project settings for this repo
├── agents/                  # Le Chat agent system prompts
│   ├── FRONTEND-DEV.md      # Frontend Developer (WordPress-focused)
│   ├── REVIEWER.md          # Frontend Code Reviewer
│   └── (add your own!)
├── prompts/                 # Example Le Chat agent runs (prompt → output → feedback)
│   ├── frontend-dev/
│   │   └── elayne-hero-block/
│   │       ├── PROMPT.md        # Input prompt used
│   │       ├── OUTPUT.md        # Raw agent output
│   │       ├── FEEDBACK.md      # Review notes and issues found
│   │       └── canvas.html      # Generated HTML canvas
│   └── reviewer/
│       └── elayne-hero-block/
│           ├── PROMPT.md        # HTML input submitted for review
│           ├── OUTPUT.md        # Structured review report (FAIL)
│           └── FEEDBACK.md      # Review notes
├── skills/                  # Mistral Vibe skills
│   └── code-review/
│       └── SKILL.md         # /code-review for WordPress plugins, themes and sites
├── README.md
├── CHANGELOG.md
└── LICENSE.md

💬 Le Chat Agents

Quick Start

  1. Copy any .md file from the agents/ directory (e.g., agents/FRONTEND-DEV.md)
  2. Paste into Le Chat → Agents → New Agent → System Prompt
  3. Set model to mistral-large-latest or devstral
  4. Start chatting!

Tip: All prompts follow a consistent structure — you can easily modify them for your tech stack.

Available Agents

Agent Role Focus Best For
FRONTEND-DEV.md Senior Frontend Developer WordPress themes, HTML/CSS, block patterns WordPress devs, theme authors
REVIEWER.md Frontend Code Reviewer HTML/CSS output analysis, FSE compliance Frontend devs, designers

Planned Agents:

  • BACKEND-DEV.md - PHP/WordPress & general backend development
  • DEVOPS.md - Deployment, infrastructure, and CI/CD
  • CONTENT-STRATEGY.md - Content planning and SEO
  • UI-UX.md - User interface and experience design
  • QA-TESTING.md - Quality assurance and testing strategies
  • ACCESSIBILITY.md - WCAG compliance and a11y best practices

WordPress Context

The current agents are tailored for these Imagewize projects:

Project Type Tech Stack
Elayne FSE Block Theme Gutenberg, theme.json, patterns
Nynaeve Sage 11 Hybrid Blade, ACF Composer, React blocks
imagewize.com Site Enterprise WordPress

They include full page builds with component extraction, developer handoff notes with WordPress implementation mappings, and project-specific conventions (FSE block compatibility, ACF fields, etc.).

Rules for the Frontend Agents

  1. Production-ready output only - No placeholders, descriptions, or "replace with X" comments
  2. Mobile-first - All output must be responsive by default
  3. No centered overlay heroes - This is the most common failure pattern to avoid
  4. Realistic content - Use relevant dummy content, never lorem ipsum or placeholder text
  5. Consistent formatting - Follow the repository's style guidelines

Example Agent Runs

The prompts/ directory contains real agent runs — input prompt, raw output, canvas, and review feedback — so you can see how each agent performs in practice and what to watch out for.

Run Agent Status
frontend-dev/elayne-hero-block FRONTEND-DEV Reviewed — see FEEDBACK.md
reviewer/elayne-hero-block REVIEWER FAIL — CSS variable namespace + dash syntax errors

Each run folder contains:

  • PROMPT.md — the exact input sent to the agent
  • OUTPUT.md — raw agent response
  • canvas.html — generated HTML/CSS output (frontend runs)
  • FEEDBACK.md — review notes: what worked, what failed, fixes needed

🛠 Mistral Vibe Skills

Skills run inside your project with Vibe's tools. They are invoked as slash commands.

Skill Command What it does
code-review /code-review [target] Reviews a branch diff, a GitHub PR, or a block, pattern, template or path in place, in any WordPress plugin, theme or site. Detects the project type, minimum PHP/WordPress and lint commands from the repository, and applies the project's own rules from .agents/code-review.md.
/code-review                 # this branch against the default base branch
/code-review 51              # GitHub PR #51
/code-review hero            # audit the block or pattern named "hero"
/code-review --help          # usage and valid targets

Install

Per project — shared with everyone who works on it: copy skills/code-review/ to the project's .agents/skills/code-review/.

For all your projects — add this repository's skills/ directory to ~/.vibe/config.toml:

skill_paths = ["~/code/mistral-agents/skills"]

Vibe uses the first skill it finds with a given name, and searches skill_paths before a project's .vibe/skills/ and .agents/skills/. If your config sets enabled_skills, add code-review to it.

Project rules

The skill's generic checks are the same everywhere; what catches most regressions is the project's own rules. Add them to .agents/code-review.md in the project — facts, named audit targets, files to skip, lint commands, and rules with the failure each one prevents. The file format is documented at the end of SKILL.md. For a plugin or theme, keep .agents/ and .vibe/ out of release zips (.distignore, .gitattributes export-ignore).

Fewer approval prompts

Approvals come from the [tools.bash] allowlist in Vibe's config, not from a skill's allowed-tools. Put the commands the skill runs in the project's .vibe/config.toml so everyone on the project gets the same approvals. A project allowlist replaces the one in ~/.vibe/config.toml, so include the read-only basics too — the Setup section of SKILL.md has a starting list.

Project instructions: AGENTS.md, not a system prompt

Vibe loads AGENTS.md files on top of its built-in system prompt; that is the place for project rules. system_prompt_id in config.toml replaces the built-in prompt — tool use, planning and editing instructions included — and is rarely what you want. See Mistral Vibe: AGENTS.md vs a custom system prompt.


📝 Adding Your Own

A Le Chat agent

  1. Create: New .md file in the agents/ directory (e.g., agents/BACKEND-DEV.md)
  2. Name it well: Use role-based naming like ROLE-SPECIALIZATION.md
  3. Structure your prompt:
    • Role description and scope
    • Critical rules (do's and don'ts)
    • Output format expectations
    • Examples of expected output
    • Project-agnostic instructions where possible

Naming Convention:

{ROLE}-{SPECIALIZATION}.md    e.g., BACKEND-PHP.md
{ROLE}.md                    e.g., DEVOPS.md
{TEAM}-{ROLE}.md             e.g., FRONTEND-DEV.md

Quality Checklist:

  • Clear, specific instructions (no "you should probably")
  • Examples of expected output
  • Do's and don'ts explicitly stated
  • Project-agnostic where possible (add variants for specific stacks)
  • Follows existing formatting in other prompts

A Vibe skill

  1. Create skills/<name>/SKILL.mdname is lowercase letters, digits and hyphens.
  2. Frontmatter: name and description are required; add user-invocable: true for a slash command and a metadata.version so copies in projects can be compared.
  3. Keep it generic: project-specific rules belong in the project, not in the skill.
  4. Prescribe allowlist-friendly commands: plain single commands, no shell loops or one-liners; read files with read_file.

🔗 Resources


🤝 Contributing

  1. Fork this repository
  2. Create a new branch for your agent prompt or skill
  3. Add your file following the conventions above (see also AGENTS.md)
  4. Update the README and CHANGELOG
  5. Submit a pull request

We welcome contributions for any domain - not just WordPress!


Created and maintained by Imagewize.

About

Mistral Le Chat agent prompts and Mistral Vibe skills for WordPress development

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages