generate-agent

Generate Agent Definition

Create an agent markdown file from the council plan and research findings. Follows the exact structure of existing agents in the agents/ directory.

Input

$ARGUMENTS is the agent name (e.g., data-audits, eval-governance, synth-debate).

Process

Step 1: Read inputs

The calling skill passes the agent's config (purpose, dimensions, sources) via the prompt.
Read research/$ARGUMENTS.md  → domain expertise

Step 2: Read a template

Determine the agent type from prefix and read the template:

data-* → Read skills/generate-agent/templates/data.md
eval-* → Read skills/generate-agent/templates/eval.md
synth-* → Read skills/generate-agent/templates/synth.md

Templates use UPPER_CASE placeholders (NAME, DESCRIPTION, DIMENSION_1, etc.) that the generator fills in from the plan and research.

For additional reference, you can also read an existing agent of the same type (if any exist) to see a fully realized example. But always use the template as the structural skeleton.

The templates already encode the full structure. For reference, here are the minimal fallbacks if templates are somehow missing:

Data agent minimal template:

---
name: $ARGUMENTS
description: [from COUNCIL-PLAN.md]
tools: Read, Write, WebSearch, WebFetch, SendMessage, TaskUpdate, TaskList
---

# Data Gatherer: [Human Name]

You are a data-gathering agent on a public goods evaluation council. [role description].

## Input

You receive `$PROJECT` (a project name or URL) and `$OUTPUT_DIR` (where to write your findings).

## Process

1. **TaskUpdate**: claim your task (status="in_progress")
2. **Search**: [specific search steps from research]
3. **Fetch**: [specific fetch steps]
4. **Write output**: Write structured markdown to `$OUTPUT_DIR/[name].md`
5. **TaskUpdate**: complete task (status="completed")
6. **SendMessage**: send 2-line summary to team lead

## Output Format

[structured markdown template with tables]

If data cannot be found, note this explicitly — missing data is valuable signal.