evaluate

Public Goods Evaluation Council

Orchestrate a multi-agent council to evaluate a public goods project. Three waves of agents gather data, evaluate from different lenses, and synthesize a final report.

Input

$ARGUMENTS is the project name or URL to evaluate.

If not provided, AskUserQuestion: "What project should the council evaluate?" with options for well-known projects or free text.

Store as $PROJECT. Derive $SLUG by lowercasing, replacing spaces/special chars with hyphens, truncating to 40 chars.

Step 1: Parse Input + Discover Agents

Discover the council roster dynamically from the filesystem:

Glob agents/data-*.md → $DATA_AGENTS
Glob agents/eval-*.md → $EVAL_AGENTS
Glob agents/synth-*.md → $SYNTH_AGENTS

For each agent file, read the frontmatter name and description fields.

Create the output directory:

mkdir -p council-out/$SLUG/data council-out/$SLUG/eval

Present the roster to the user:

AskUserQuestion:
  question: "Council assembled. Proceed with evaluation?"
  header: "Council"
  options:
    - label: "Run evaluation"
      description: "Start the 3-wave evaluation"
      preview: |
        Project: $PROJECT

        Wave 1 — Data Gathering ($N agents):
        ├─ [name]: [description]
        ├─ [name]: [description]
        └─ ...

        Wave 2 — Evaluation ($M agents):
        ├─ [name]: [description]
        └─ ...

        Wave 3 — Synthesis ($K agents):
        └─ [name]: [description]

        Output: council-out/$SLUG/REPORT.md
    - label: "Modify roster"
      description: "Add or remove agents before running"

If "Modify roster" → invoke the add-agent skill via the Skill tool, then re-discover.

Step 2: Create Team + All Tasks

TeamCreate(team_name="council-$SLUG", description="Public goods evaluation council for $PROJECT")