Brief format
The Markdown brief that the Plan Assistant splits into units automatically — front-matter config, the PROMPT section, UNITS sections, and placeholders.
A brief is a Markdown file the Assistant splits into a campaign automatically. It’s the easiest way to define a bulk job.
You don’t have to write it from scratch: download the starter template (localized to the
IDE language), reuse a brief saved in My templates, or let an agent write it for you
from a plain-words description — all from step 1 of the Assistant. Note the structural
keywords (# PROMPT, # UNITS, front-matter keys) are never translated: the parser expects
them as-is.
Structure
---
label: 50 Astro sites
provider: claude
agent: # optional sub-agent (→ {agent})
skill: # optional skill (→ {skill})
output_dir: sites
output_name: {unit}
done: dir # file | dir
max_sessions: 4
---
# PROMPT
Build the site described below in sites/{unit}/. Stop when it compiles.
{brief}
# UNITS
## Restaurant landing
One-pager, menu + reservation, warm palette…
## SaaS B2B landing
Hero + pricing + FAQ…
The three parts
- Front-matter (between
---): the campaign config. Keys:label,provider,agent,skill,output_dir,output_name,done(file|dir),max_sessions. All optional except where noted; sensible defaults apply. # PROMPT— the prompt template applied to every unit. Use the placeholders below.# UNITS— each## Titleunderneath is one unit (one loop, one agent). Its body becomes{brief}for that unit.
Placeholders
Available in # PROMPT:
| Placeholder | Replaced with |
|---|---|
{unit} | the unit id — the ## Title turned into a slug |
{brief} | the body of the unit’s ## section |
{output} | the resolved output path for the unit |
{agent} | the front-matter agent: value |
{skill} | the front-matter skill: value |
{unit_path} | source file path (for glob sources) |
How the slug works
The ## Title becomes the unit id via slugification (lowercase, alphanumerics, dashes), which
is used in {unit} and therefore in the output name:
## Restaurant landing→restaurant-landing## ACME Corp (FR)→acme-corp-fr
Reliability detail
Plan writes each unit’s full prompt to a cache file and launches the agent with a short instruction — “read this file and execute it” — in the language of the IDE. This sidesteps command-line length limits and submission quirks for large prompts. You don’t have to think about it; it’s why long, multi-line prompts simply work.
Editing later
Editing the brief file and clicking Retry failed re-reads it, so the campaign picks up your
changes. Adding new ## sections later extends the campaign — units are re-enumerated from
the brief on demand.