Coding Rules — Headquarters

🧭 Task Routing

When an agent receives a task, it must route automatically based on type. Agents do not ask which agent to use.

Task Type	Route To	How
Code / logic / files	Cursor Agent	`sessions_spawn(runtime="subagent", agentId="cursor", task="...")`
Vision / screenshots / UI	Codex	`sessions_spawn(runtime="acp", agentId="codex", task="...")`
Planning / coordination	Self	Handle directly

Rule: Analyze the task and route automatically. Never ask the user which agent to use.

✅ Post-Coding Verification

MUST After every coding task, agents must complete all three steps:

1. Run locally

→

2. Screenshot

→

3. Verify

1. Run the Project Locally

Start the dev server (npm run dev, yarn dev, etc.)
Wait for it to be ready (check for "ready" output or port binding)

2. Take Screenshots

Use Playwright or browser automation to capture the running app
Save screenshots to the workspace
Post them using the MEDIA: directive on its own line

3. Verify Functionality

Check that the changes work as expected
Note any issues or console errors
Report findings in the response

# Example: Run, wait, screenshot
npm run dev -- --port 3456
# Wait for "ready" signal, then:
npx playwright-core screenshot --browser=chromium \
  --viewport-size=1400,900 --wait-for-timeout=3000 \
  "http://localhost:3456" ./screenshot.png

# Post to chat
# MEDIA:./screenshot.png

🚀 Deployment

When Asked to "Check In" or "Deploy"

Always include Vercel links in the response:

Deployment page: Check VERCEL_DEPLOYMENT_URL from the agent's .env file
Hosted URL: Check VERCEL_HOSTED_URL from the agent's .env file
Include direct links to specific pages that were changed

Response Format

📦 Deployment: <VERCEL_DEPLOYMENT_URL>
🌐 Live site: <VERCEL_HOSTED_URL>
🔗 Pages to test:
- <VERCEL_HOSTED_URL>/page1
- <VERCEL_HOSTED_URL>/page2

Vercel Authentication

All agents share a single Vercel token:

# Load credentials
source ~/.openclaw/workspace/.env.vercel

# Deploy
vercel --token $VERCEL_TOKEN --yes

Token location: ~/.openclaw/workspace/.env.vercel
Generate new tokens: vercel.com/account/tokens

Environment Variables (per agent)

Each agent should have these in their workspace .env file:

Variable	Purpose
`VERCEL_DEPLOYMENT_URL`	Link to Vercel deployments dashboard
`VERCEL_HOSTED_URL`	Link to the live deployed site

The orchestrator provides these values per-project.

📝 Publishing Markdown Outputs

REQUIRED Any task that produces a .md file as output MUST publish it to GitHub.

Path Structure

microspaces/<discord-server-name>/<channel>/<bot>/<task>/<task>_<date>.md

Example

microspaces/superstarsale/marketing/clio/campaign-analysis/campaign-analysis_2026-06-12.md

How to Publish

# Load credentials
source ~/.openclaw/workspace/.env.github

# Publish
~/.openclaw/workspace/scripts/github-publish-md.sh \
  <server> <channel> <bot> <task> <markdown-file>

The script auto-creates the repo microspaces/<server> if it doesn't exist, commits to main, and returns the GitHub URL.

Always include in response: 📄 Output: https://github.com/microspaces/<server>/blob/main/...

🖥️ Shell & Environment

Windows (Current)

Item	Value
Host	DesktopWestport (Windows 11)
Default shell	PowerShell (`pwsh`)
Shell scripts	Git Bash (`C:\Program Files\Git\bin\bash.exe`)
WSL	Not installed — do not use
Node.js	v24.18.0
Python	Not installed — use `node` for scripting
Workspace	`C:\Users\AI\.openclaw\workspace`
Repos	`C:/Users/AI/Repos/`

⚠️ All .sh scripts must run via Git Bash. Do not use WSL — no distros are installed.

Path Conventions

Use C:/Users/AI/... with forward slashes in configs and code
Backslashes (C:\Users\AI\...) for PowerShell native commands
Line endings: Git handles conversion; .sh files stay LF

🔐 Safety Rules

Safe to Do Freely

Read files, explore, organize, learn
Search the web, check calendars
Work within the workspace

Ask First

Sending emails, tweets, public posts
Anything that leaves the machine
Anything you're uncertain about

Destructive Commands

Never run destructive commands without asking. trash > rm — recoverable beats gone forever.

Private Data

Don't exfiltrate private data. Period.
Don't load MEMORY.md in shared contexts (Discord, group chats)
When in doubt, ask before acting externally

🔀 Development Workflow

Universal rules for all coding agents across all repos. Full Workflow Rules →

1. Branch from main

→

2. Develop locally

→

3. Run & screenshot

→

4. Push branch

→

5. Vercel preview

→

6. Open PR

→

7. Post to Discord

→

8. Merge & deploy

Branching

Branch from main as agent/<description> — never commit directly to main
Run locally, take screenshots, save to test-screenshots/ directory
Push branch → Vercel auto-creates preview deployment

Pull Requests

Open PR with template (summary, testing checklist, screenshots, preview URL, Convex note)
Post screenshots + preview link + PR link to Discord
Merge to main = Vercel production deploy

Convex Auto-Deploy

Convex backend deploys via GitHub Action on merge to main
Configured for: crm, almightydollar-app, surroundshow-v2

Safety net: test-screenshots/ is in .gitignore on main. All GitHub repos have AGENTS.md with workflow section + .github/PULL_REQUEST_TEMPLATE.md.

📊 Status Updates

Agents must update their status at key moments:

When	Status
Starting work	In Progress
Finished work	Done
Blocked	Blocked + specify blocker
When asked	Respond with current status

# Update both board and HTML page
source ~/.openclaw/workspace/.env.github

~/.openclaw/workspace/scripts/update-agent-status.sh \
  <agent> <status> <task> [server] [blocked-by] [discord] [priority] [type] [progress] [notes]

~/.openclaw/workspace/scripts/update-status-html.sh \
  <agent> <status> <task> [server] [blocked-by] [discord]