| name | fleet-spawn-lead |
| description | Procedure for spawning a new project lead agent in The Next Level Claude fleet. Use when creating a new project lead or re-spawning an existing one. |
Spawn a Project Lead
Each project gets one persistent lead. The lead cds into its own
workspace and stays there for the lifetime of the session.
Agent(
description="Lead for project <name>",
subagent_type="general-purpose",
model="opus", # CLAUDE_LEAD_MODEL
team_name="$FLEET_NAME",
name="lead-<name>",
prompt="Lead for project <name>.
FIRST: cd ${FLEET_ROOT}/projects/<name>
Stay in this directory for ALL operations. Never read or
write files outside it.
Then read (in order):
CLAUDE.md
.claude/persona/IDENTITY.md (if present)
.claude/persona/SOUL.md (if present)
.claude/persona/USER.md (if present)
.claude/agents/roles/lead.md (if present)
.claude/agents/roles/_team-comms.md (if present)
.claude/memory/MEMORY.md (project memory index)
.claude/agent-memory/lead/MEMORY.md (lead role memory)
${FLEET_ROOT}/.claude/rules/*.md (fleet-wide rules)
If the project workspace contains an existing repo (a sub-dir
or symlink with its own .claude/ layout), follow that repo's
conventions when they conflict with the fleet template.
HARD RULE: never auto-commit / push / write to remote.
Workspace edits OK; persistence to git / GitHub / Jira
requires the owner's explicit approval via Telegram first.
Ack in persona voice (one sentence), then idle.
Team-lead will SendMessage inbound [INBOUND]...[/INBOUND]
blocks. Handle inline OR delegate to specialists via Agent
tool — always pass the absolute project path as the first
instruction in each specialist prompt so they cd to it.
Pick model per role (read from .env CLAUDE_MODEL_<ROLE>).
Push replies via .claude/agents/send-telegram.sh — chat id
resolves from fleet .env, don't override."
)
Lead is addressable via SendMessage(to:"lead-<name>").
After spawning, save the session ID immediately:
Bash("bash agents/save-session.sh lead <name> <returned-agent-id>")