Dispatch Team Lead

Dispatch Team Lead

Use this skill when you need to dispatch an autonomous Team Lead agent to handle a complex task. Team Leads are senior engineers who own execution end-to-end — you give them direction and context, they figure out the how.

Prerequisites

Before dispatching, ensure the following are available:

• ahpx must be installed — this is the Agent Host Protocol CLI for dispatching agents to AHP servers:

  npm install -g @tylerl0706/ahpx
  

  See ahpx documentation for setup details.

• At least one AHP server must be running and discoverable. Configure servers with ahpx server add. Run ahpx server status to check available servers.

When to Dispatch a Team Lead

Dispatch a Team Lead when:

• The task requires multiple steps — implementation, testing, review, and shipping
• The task needs autonomous decision-making — architecture choices, testing strategy, refactoring decisions
• You want quality enforcement built in — Team Leads carry the development culture (quality over everything, test diversity, self-review, branch policy)
• The task should run independently — you want to hand it off and get a structured summary when it's done

Don't dispatch a Team Lead for:

• Simple one-file edits or quick fixes
• Questions that only need research (use an explore agent instead)
• Tasks where you need real-time control over every step

How to Dispatch

The dispatch pattern uses ahpx prompt to create a named session on an AHP server with the Team Lead culture injected into the prompt:

ahpx prompt -n <session-name> --cwd <project-dir> "<culture-prompt + task>"

Parameters:

• -n <session-name> — a descriptive session name for observability (e.g., fix-auth-bug, add-user-api)
• --cwd <project-dir> — the project directory to work in. REQUIRED when targeting remote servers — the agent needs to know where to operate.
• -s, --server <server-name> — target a specific AHP server in your multi-host fleet. Omit to let ahpx auto-select.
• "<prompt>" — the combined culture + task prompt

Building the Dispatch Prompt

The Team Lead culture lives in references/team-lead-culture.md. Read the FULL contents of references/team-lead-culture.md. Never truncate or summarize it — the complete culture must be injected into every dispatch prompt. This is what transforms a generic agent into a Team Lead.

Step 1: Read the FULL contents of references/team-lead-culture.md.

Step 2: Combine the culture with your task:

<full contents of references/team-lead-culture.md>

---

George (the CTO) has given you this direction:

<your task description>

Step 3: Dispatch with the combined prompt:

ahpx prompt -n <session-name> --cwd <project-dir> "<combined prompt>"

Using the dispatch script

For reliable culture injection, use the bundled dispatch script:

./scripts/dispatch.sh <session-name> <project-dir> "<task>"

This automatically reads the Team Lead culture, verifies ahpx is installed, and combines everything into the dispatch prompt.

Examples

Dispatch to the default server:

ahpx prompt -n add-webhook-support --cwd /path/to/my-api \
  "<full culture> ... George (the CTO) has given you this direction:
Add webhook support for order events. Create POST /webhooks endpoint..."

Dispatch to a specific server in your fleet:

ahpx prompt -n fix-auth-bug -s dev-server-2 --cwd /path/to/my-api \
  "<full culture> ... George (the CTO) has given you this direction:
Fix the authentication bug where JWT tokens aren't being refreshed..."

Note: The <full culture> in the examples represents the FULL contents of references/team-lead-culture.md. Never truncate it.

The server doesn't matter — the culture does. Any AHP server will produce a Team Lead when given the culture prompt.

Remote Filesystem Discovery

When dispatching to remote AHP servers, you may not know the exact filesystem paths. Use ahpx browse to discover project locations:

# Browse the filesystem on a specific server
ahpx browse -s <server-name>

# Browse a specific path
ahpx browse -s <server-name> /home/projects/

This is especially useful when managing a multi-host fleet where projects live on different machines.

Gotchas

• Never truncate the Team Lead culture. Agents will try to summarize it to save tokens — include ALL of references/team-lead-culture.md in the dispatch prompt. Truncated culture produces agents that skip quality gates.
• --cwd is REQUIRED for remote servers. Without it, the agent won't know where to operate. Always specify the project directory when dispatching to remote AHP servers.
• Verify ahpx is installed before first dispatch. Run ahpx --version. If it fails, install with npm install -g @tylerl0706/ahpx. A missing ahpx produces cryptic "command not found" errors.
• Don't dispatch for trivial tasks. Single-file edits, quick lookups, simple questions — handle these yourself. Team Leads will over-engineer simple tasks because the culture tells them to invest in expertise and create PRs.
• Session names must be unique. If you reuse a session name, ahpx may resume an old session instead of starting fresh. Use descriptive, task-specific names like fix-auth-bug or add-dark-mode.

Session Management

ahpx provides session management for observability into dispatched agents:

# List all sessions
ahpx session list

# Check the details of a specific session
ahpx session show -n <session-name>

# View the history of what happened in a session
ahpx session history -n <session-name>

Use descriptive session names (fix-auth-bug, add-webhook-support, refactor-payments) so you can easily identify what each dispatched Team Lead is working on.

Writing Good Dispatch Prompts

Team Leads are autonomous — they don't need step-by-step instructions. Give them what and why, not how.

✅ Good prompts

"Add rate limiting to the API. We're getting hammered by a single client making
thousands of requests per minute. Use a sliding window approach, return 429 with
Retry-After headers, and make the limits configurable per-route."

"Fix the flaky test in auth.test.ts. It passes locally but fails in CI about 30%
of the time. Likely a timing issue with the token expiry mock."

❌ Bad prompts

"Open auth.ts, go to line 45, change the timeout from 30 to 60."

→ Too prescriptive. Just tell the Team Lead what's wrong and let them decide the fix.

"Do stuff with the API."

→ Too vague. Team Leads need clear context about what and why.

What You Get Back

When a Team Lead finishes, it provides a structured summary:

## Agent Summary
**Status:** completed (or failed)
**What was done:** Brief description
**What worked:** What went well
**What failed:** Any issues encountered
**Follow-up recommendations:** Next steps or remaining debt

Plus a detailed report covering implementation, testing, tradeoffs, and any technical debt.