AgenTeam Add Role

Add a new team member from a natural language description.

Process

1. Auto-Init Guard

Check for .agenteam/config.yaml, .agenteam.team/config.yaml, or legacy agenteam.yaml. If all are missing:

• Create config dir: mkdir -p .agenteam
• Copy the template: cp <plugin-dir>/templates/agenteam.yaml.template .agenteam/config.yaml
• Set the team name to the project directory name
• Generate agents: python3 <runtime>/agenteam_rt.py generate

2. Parse Intent

Extract role details from the user's natural language request. Examples:

• "Add a performance tuning engineer" ->

name: performance_engineer, focus: profiling + optimization

• "I need a security auditor on the team" ->

name: security_auditor, focus: vulnerabilities + auth

• "Add a docs writer that maintains README and API docs" ->

name: docs_writer, focus: documentation, write_scope: docs/**

• "Add a DevOps engineer for CI/CD" ->

name: devops_engineer, focus: pipelines + deployment

Infer as much as possible from the description:

| Field | How to infer | |-------|-------------| | name | Snake_case from the role title | | description | From user's description | | responsibilities | 3-5 items inferred from the role's domain | | participates_in | Match to pipeline stages: research, strategy, design, plan, implement, test, review | | can_write | Yes if the role creates/modifies files; no if it only analyzes | | write_scope | Infer from what the role writes (docs, src, tests, configs) | | model | Omit to inherit platform default. If the user requests a pin, inspect codex debug models and choose only from the live catalog. Use evaluation evidence to justify a strong model for demanding analysis or a smaller model for bounded worker tasks. Model is a personal override — share-config strips it. | | reasoning_effort | high for analysis roles, medium for execution/writing roles | | system_instructions | Generate focused instructions from the role's domain |

3. Team Size Check

Before confirming, count the current roles:

bash
python3 <runtime>/agenteam_rt.py roles list

• 7-12 roles: No warning. This is the productive range.
• 13+ roles: Warn the user:

"Your team will have N] roles. Teams above 12 can increase coordination overhead and make role selection harder. Consider extending an existing role's system_instructions instead. Proceed anyway? (yes / cancel)"

Also check: if the team will have more than 6 roles, note: "Codex defaults to 6 concurrent agent threads. To run more agents in parallel, set agents.max_threads in your Codex config.toml."

4. Confirm with User

Present the inferred role as a summary and ask for confirmation:

Here's your new team member:

  Name: performance_engineer
  Focus: Profiling, bottleneck analysis, optimization
  Stages: review, implement
  Writes to: src/** (optimization patches)
  Model: inherited default (or `gpt-5.4` if you want to pin it)

  System instructions:
  You are the performance engineer on an AgenTeam. Your primary job is
  to identify bottlenecks and optimize critical paths...

Add to team? (yes / adjust)

If the user says "adjust" or requests changes, update the fields and re-confirm. Do not ask field-by-field -- keep it conversational.

5. Write to Config

Read the current .agenteam/config.yaml (or legacy agenteam.yaml) and add the new role under roles:.

Write the full role block including:

• description
• responsibilities
• participates_in
• can_write and write_scope (if applicable)
• model and reasoning_effort
• parallel_safe (true for read-only roles, false for writers unless scoped)
• system_instructions

If the role participates in a pipeline stage, also add it to the appropriate pipeline.stages[].roles list.

6. Regenerate Agents

bash
python3 <runtime>/agenteam_rt.py generate

7. Confirm

Show the user:

• The generated agent file: .codex/agents/<name>.toml
• How to use it immediately:
• Codex App: @ateam ask <name> to <task>
• Codex CLI: $ateam:assign <name> "<task>"
• Reminder: "Edit .agenteam/config.yaml anytime to adjust this role."