Agency setup

Great Minds Agency — Setup Wizard

Initialize a new Great Minds project through a 3-phase conversational wizard.

Usage:

• /agency-setup — full setup (5–8 questions)
• /agency-setup quick — quick setup (project name only, all defaults)

---

Phase 1 — Scan (silent, no user input required)

Run these checks silently before asking any questions.

1a. Detect project name

basename "$PWD"

Store the result as the default project name.

1b. Check for existing config

test -f .great-minds.json && cat .great-minds.json || echo "NO_CONFIG"

If a config exists, skip to the Reconfigure section below.

1c. Detect PRD directory

Check these paths in order; use the first that exists:

test -d docs/prds && echo "docs/prds" || test -d prds && echo "prds" || test -d docs && echo "docs" || echo "NOT_FOUND"

1d. List available agent specs

ls ~/.claude/agents/*.md 2>/dev/null | xargs -I{} basename {} .md || echo "NO_AGENTS"

Store the list of agent names (without .md).

---

Phase 2 — Questions

After scanning, summarize what you found in one short paragraph, then ask questions one at a time. Never ask more than one question per message.

If the user passed quick as an argument, or says "quick" at any point: Ask only Question 1, then proceed to Phase 3 with all other fields at defaults:

• Agents: use all detected agent specs if found, otherwise []
• PRD dir: use detected path if found, otherwise ./docs/prds
• Schedule: null (manual only)
• Auto-ship: false
• Telegram: skipped
• Token budget: 50000
• Custom rules: none

In the completion message for quick mode, explicitly note the assumed PRD dir so the user knows.

Question 1 — Project name

"I'll set this up as {detected-name}. Is that the right project name, or would you like a different one?"

If they confirm, use the detected name. If they give a new name, use that.

Question 2 — Active agents

"Which agents should be active on this project? Available: {list-from-scan}

You can say 'all', list specific names, or 'none' to leave agents unconfigured for now."

If the user says 'all': write every name from the scanned list (1d) into agents.active, each normalized (lowercase, strip .md).

Normalize each name (lowercase, strip .md). If the user names an agent not in the detected list, say:

"I don't see {name} in ~/.claude/agents/ — did you mean one of these? {closest-matches}. You can also say 'skip' to leave it out."

Question 3 — PRD directory

If a directory was detected:

"I found a PRD directory at {detected-path}. Is that correct, or would you like a different path?"

If nothing was detected:

"Where should the pipeline look for incoming PRDs? (e.g. docs/prds, prds/, or 'skip' to configure later)"

Do NOT create the directory if it doesn't exist — just record the path.

If the user says 'skip': use "./docs/prds" as the default value for prds.dir.

Question 4 — Pipeline schedule

"How often should the pipeline run? Examples: 'hourly', 'every 30 minutes', 'daily', 'manual only'"

Convert the answer to a cron expression using these rules:

• "hourly" / "every hour" → 0 * * * *
• "every N minutes" (N 1–59) → */N * * * *
• "every N hours" (N 1–23) → 0 */N * * *
• "daily" → 0 0 * * *
• "manual only" / "manual" / "skip" → null (no cron script generated)
• A raw 5-part cron expression → use as-is
• Out-of-range or unrecognized → ask again

Question 5 — Auto-ship

"Should the pipeline auto-ship after QA passes, or pause for your manual approval? (auto / manual — default: manual)"

If skipped or anything other than "auto", use false.

Question 6 — Telegram notifications (optional)

"Telegram notifications? Paste your bot token and chat ID (e.g. 1234567:ABC… -987654321), or say 'skip'."

If skipped, store botToken: "" and chatId: "".

Question 7 — Token budget (optional)

"Token budget per agent per session? (default: 50000 — say 'skip' to use default)"

If skipped, use 50000.

Question 8 — Custom rules (optional)

"Any custom rules or banned patterns to add? (paste them, or say 'none')"

If none, store bannedPatterns: [] and customRules: []. Otherwise, split by newline or comma and store each entry as a string in customRules.

---

Phase 3 — Generate

Once all questions are answered, say:

"Creating project files..."

Then write all files. Use the exact formats below.

File 1: .great-minds.json

Write this JSON to .great-minds.json in the current working directory:

{
  "project": "{project-name}",
  "version": "1.0",
  "createdAt": "{ISO-timestamp}",
  "prds": {
    "dir": "{prd-dir}"
  },
  "agents": {
    "active": ["{agent-1}", "{agent-2}"],
    "tokenBudgetPerAgent": {token-budget}
  },
  "pipeline": {
    "schedule": "{cron-expression-or-null}",
    "autoShip": {true-or-false}
  },
  "notifications": {
    "telegram": {
      "botToken": "{token-or-empty}",
      "chatId": "{id-or-empty}"
    }
  },
  "rules": {
    "bannedPatterns": [],
    "customRules": ["{rule-1}", "{rule-2}"]
  }
}

Use null (not "null") for schedule when manual-only was selected.

File 2: STATUS.md

# Great Minds Agency — Status

## Current State
- **state**: idle
- **project**: {project-name}
- **last updated**: {YYYY-MM-DD}
- **pipeline**: {cron-expression, or "manual only" if null}

## Active Agents

| Agent | Status | Activity |
|-------|--------|----------|
{one row per active agent: | {agent-name} | Active | — |}

(If no agents configured: | — | — | Run /agency-setup to configure agents |)

## PRD Directory
- {prd-dir}

Date format: Use YYYY-MM-DD (date-only, e.g. 2026-04-10) for all {YYYY-MM-DD} tokens. Do NOT use the full ISO timestamp.

File 3: SCOREBOARD.md

# Great Minds Agency — Scoreboard

**Project**: {project-name}
**Session started**: {YYYY-MM-DD}
**Last updated**: {YYYY-MM-DD}

---

## Agency Totals

| Metric | Count |
|--------|-------|
| Commits | 0 |
| PRs merged | 0 |
| QA reports | 0 |
| Board reviews | 0 |

## Agent Scores

_(Updated as work progresses)_

File 4: TASKS.md

# Great Minds Agency — Master Task Board

**Project**: {project-name}
**Managed by**: Phil Jackson (Orchestrator)
**Last updated**: {YYYY-MM-DD}

## Open Tasks

_(No tasks yet — run /agency-plan to generate the first task list)_

## Recently Completed

_(none)_

## Blocked

_(none)_

File 5: crons/pipeline-runner.sh (only if schedule is not null)

Create the crons/ directory if it doesn't exist, then write crons/pipeline-runner.sh:

#!/bin/bash
# Great Minds Agency — Pipeline Runner
# Schedule: {cron-expression}
# Project: {project-name}
# Generated by /agency-setup on {YYYY-MM-DD}
#
# To activate: use /agency-crons or run CronCreate with cron: "{cron-expression}"

set -euo pipefail

PROJECT_DIR="{absolute-path-to-project}"
cd "$PROJECT_DIR"

# Run one pipeline cycle
npx tsx daemon/src/daemon.ts --once

Make it executable with chmod +x crons/pipeline-runner.sh.

If schedule is null (manual only), do NOT create the cron script. Instead add this note to STATUS.md's pipeline line:

- **pipeline**: manual only (run /agency-plan to start a pipeline cycle manually)

---

Completion Message

After all files are written, say:

"Project initialized.

Created:

• .great-minds.json — machine-readable config
• STATUS.md — live state tracker
• SCOREBOARD.md — metrics scoreboard
• TASKS.md — task board {- crons/pipeline-runner.sh — pipeline cron (schedule: {cron}) — include only if applicable}

Next steps:

• Run /agency-crons to activate the pipeline schedule
• Drop a PRD into {prd-dir} to start your first project
• Run /agency-plan to kick off a manual pipeline cycle"

---

Reconfigure (existing config found)

If Phase 1 found an existing .great-minds.json, say:

"I found an existing config for {project-name}. Do you want to reconfigure it (I'll pre-fill your current values), or cancel?"

If they say reconfigure:

• Pre-fill all Phase 2 questions with the existing values as defaults
• User only needs to answer questions where they want to change something
• After Phase 3, overwrite .great-minds.json, STATUS.md, and TASKS.md with fresh content
• Do NOT regenerate SCOREBOARD.md during reconfigure — it tracks cumulative metrics and overwriting would destroy existing data. Leave it unchanged.

If they say cancel: stop and do nothing.