Project Description Format¶

This guide documents every field in the project description format used by the Agent Teams Module. For the full JSON Schema, see schemas/project-description.schema.json.

Quick Template¶

Copy this template and fill in your values. Only project_goal is required; all other fields improve the quality of the generated team.

{
  "project_name": "MyProject",
  "project_goal": "One to three sentences describing what this project produces.",
  "deliverables": ["primary deliverable type", "secondary deliverable type"],
  "output_format": "Python 3.11 modules",
  "primary_output_dir": "src/",
  "build_output_dir": "dist/",
  "figures_dir": "docs/figures/",
  "authority_sources": [
    {
      "name": "Source specification",
      "path": "docs/spec.md",
      "scope": "what this source is authoritative for",
      "rank": 1
    }
  ],
  "style_reference": null,
  "reference_db_path": null,
  "reference_key_convention": "AuthorYear",
  "tools": [
    {"name": "Python", "version": "3.11", "category": "language"},
    {"name": "FastAPI", "category": "framework"},
    {
      "name": "PostgreSQL",
      "version": "15",
      "category": "database",
      "config_files": ["docker-compose.yml"],
      "needs_specialist_agent": true
    }
  ],
  "style_rules": [
    "Rule one",
    "Rule two"
  ],
  "components": [
    {
      "slug": "module-name",
      "name": "Human-Readable Module Name",
      "number": 1,
      "output_file": "src/module.py",
      "description": "What this component does.",
      "sections": ["Section 1", "Section 2"],
      "sources": ["SourceKey2024"],
      "cross_refs": ["other-component"],
      "quality_criteria": ["Criterion one", "Criterion two"]
    }
  ]
}

Field Reference¶

`project_goal` (required)¶

Type: string
Purpose: The engine uses this as the top-level description of the project. It appears verbatim in the orchestrator and workstream expert files, and drives project type classification.

What the engine does with it: Analyzes keywords to classify the project type (software, research, content, data-pipeline, mixed), then uses the type to select domain agent archetypes.

Good:

"project_goal": "Build a Python FastAPI backend with JWT authentication, task CRUD endpoints, and automated test coverage to 80%."

Too vague:

"project_goal": "Make a web app."

Common mistake: Describing how to build rather than what to produce. Focus on the deliverable.

`project_name`¶

Type: string
Default: Derived from the project directory name if not provided.
Purpose: Appears in every agent file name and header.

`deliverables`¶

Type: array of strings
Purpose: Tells the engine what types of artifacts the project produces. Used to select domain archetypes (e.g., LaTeX deliverables trigger format-converter; citations in deliverables trigger reference-manager).

Examples:

"deliverables": ["Python modules", "OpenAPI documentation", "test suite"]
"deliverables": ["HTML chapter drafts", "LaTeX manuscript", "BibTeX bibliography"]
"deliverables": ["Python ETL modules", "SQL scripts", "PDF reports"]

`output_format`¶

Type: string
Default: Inferred from project type (e.g., "Python modules" for software projects).
Purpose: The final compiled format. Appears in the format-converter agent and in copilot-instructions.

`primary_output_dir`¶

Type: string (path)
Default: Inferred from project type ("src/" for software, "html/" for content).
Purpose: Where authored files live. This is what agents treat as the "source of truth" for primary deliverables.

`build_output_dir`¶

Type: string (path)
Default: "build/"
Purpose: Where compiled/built artifacts are written. Agents do not directly author files here.

`figures_dir`¶

Type: string (path)
Default: "figures/"
Purpose: Directory for diagram source files and rendered figures. Used by visual-designer if included.

`authority_sources`¶

Type: array of objects
Purpose: Files that agents treat as ground truth. The ranking determines which source wins in a conflict.

Each entry:

{
  "name": "Human-readable name",
  "path": "relative/path/to/file",
  "scope": "What this source is authoritative for",
  "rank": 1
}

What the engine does with it: Builds the authority hierarchy in orchestrator, technical-validator, and conflict-auditor agents. Lower rank numbers = higher authority.

Common mistake: Listing sources that don't yet exist. Only list files that are (or will be) present before agents start working.

`style_reference`¶

Type: string (path) or null
Default: null
Purpose: Path to a style guide, voice samples directory, or editorial conventions file. When provided, the style-guardian archetype is included in the generated team.

Effect when null: style-guardian is not generated. The {STYLE_REFERENCE_PATH} placeholder resolves to {MANUAL:STYLE_REFERENCE_PATH} and appears in SETUP-REQUIRED.md.

`reference_db_path`¶

Type: string (path) or null
Default: null
Purpose: Path to a BibTeX .bib file or similar citation database. When provided, the reference-manager archetype is included and this path populates the agent's database reference.

`reference_key_convention`¶

Type: string
Default: "AuthorYear"
Purpose: Citation key format (e.g., "AuthorYear", "IEEE", "Chicago"). Appears in reference-manager agent instructions.

`tools`¶

Type: array of tool objects
Purpose: Languages, frameworks, libraries, and infrastructure tools used in the project.

Each tool:

{
  "name": "ToolName",
  "version": "1.0",
  "category": "database",
  "config_files": ["path/to/config.yaml"],
  "needs_specialist_agent": true
}

Tool categories and their effects:

Category	Default tier	Override with `needs_specialist_agent`
`language`	Passive (listed only)	—
`framework`	Reference file	Set `true` for full agent
`library`	Reference file	Set `true` for full agent
`cli`	Reference file	Set `true` for full agent
`database`	Specialist agent	Set `false` to downgrade
`deployment`	Specialist agent	Set `false` to downgrade
`pipeline`	Specialist agent	Set `false` to downgrade
`compiler`	Specialist agent	Set `false` to downgrade
`other`	Passive	Set `true` for full agent

Auto-detection: If --project points to an existing directory, the engine also scans requirements.txt, pyproject.toml, package.json, Cargo.toml, and go.mod to supplement the tools list.

`style_rules`¶

Type: array of strings
Purpose: Project-specific coding or writing conventions. Written verbatim into governance agents.

Examples:

"style_rules": [
  "All public functions must have docstrings",
  "Type annotations required on all public signatures",
  "No mutable default arguments"
]

`components`¶

Type: array of component objects
Purpose: One component = one workstream expert agent. Each component maps to a distinct deliverable module, chapter, or subsystem.

Each component:

{
  "slug": "auth-module",
  "name": "Authentication Module",
  "number": 1,
  "output_file": "src/auth/",
  "description": "What this component produces.",
  "sections": ["JWT token validation", "Permission decorators"],
  "sources": ["AuthorYear2024"],
  "cross_refs": ["tasks-api"],
  "quality_criteria": ["All endpoints covered by tests", "No hardcoded secrets"]
}

Fields: - slug — machine-readable identifier; becomes the file name {slug}-expert.agent.md - name — human-readable label for the agent - number — ordering hint - output_file — path where this component's output is written - description — appears in the workstream expert's invariant core - sections — sub-sections or modules within this component - sources — citation keys or source names this component is grounded in - cross_refs — slugs of other components this one depends on - quality_criteria — acceptance criteria the expert uses to ACCEPT/REVISE

Markdown Format Alternative¶

If you prefer Markdown input, structure the file with these section headers:

## Project Goal
One to three sentences describing the project.

## Deliverables
- Primary deliverable type
- Secondary deliverable type

## Output Format
Python 3.11 modules

## Primary Output Directory
src/

## Tools
- Python 3.11
- FastAPI (framework)
- PostgreSQL 15 (database, specialist agent required)

## Style Rules
- All public functions must have docstrings

## Components
- auth-module: Authentication Module
- tasks-api: Tasks API

The Markdown parser uses section headings to identify fields. Unrecognized sections are appended to project_goal. JSON format is recommended for complex projects.

Common Mistakes¶

Mistake	Effect	Fix
Omitting `components`	No workstream experts generated	Add one component per major deliverable
Using `needs_specialist_agent: true` on `language` tools	Generates an agent for "Python"	Set only on tools with real operational complexity
Setting `reference_db_path` when no bib file exists	reference-manager agent generated but non-functional	Set to null if the bib file doesn't exist yet
Component slugs with spaces	Invalid file names	Use hyphens: `auth-module`, not `auth module`
Long `project_goal`	Agent files become verbose	Keep to 1–3 sentences; use `components` for detail

Project Description Format¶

Quick Template¶

Field Reference¶

project_goal (required)¶

project_name¶

deliverables¶

output_format¶

primary_output_dir¶

build_output_dir¶

figures_dir¶

authority_sources¶

style_reference¶

reference_db_path¶

reference_key_convention¶

tools¶

style_rules¶

components¶