| name | team-builder-packaging |
| description | Use when generating or auditing a multi-role agent team package with orchestrator, PM Soul, Memory Curator, Policy Gate, workers, eval, QA, handoffs, and runtime adapters. |
Team Builder Packaging
Procedure
-
Start with the orchestrator/HQ.
-
Run contracts/builder-interview-research-gate.md before writing the roster:
ask an 8-12 question first batch, research official sources, similar agent
repositories or comparables, academic/professional theory, and plugin docs,
compare tool/plugin choices, and write the domain-expert synthesis plus
prompt-performance contract.
-
Add PM Soul or project owner.
-
Add Memory Curator and Memory Ticket handoff.
-
Add Policy Gate, eval judge, and QA/evidence gate.
-
Add workers only for real domain ownership proved by interview or research.
-
Add docs/builder-interview.md, docs/research-sources.md,
docs/tool-selection.md, docs/domain-expert-synthesis.md,
docs/prompt-performance-contract.md, and
.agentlas/capability-eval-plan.json unless explicitly creating a minimal
private scaffold.
-
Encode handoff and return contracts.
-
Emit one orchestrator/HQ global command in .agentlas/global-commands.json
and runtime command files. Do not expose worker commands unless requested.
9b. Declare the execution graph in manifest.json. This is what the Hub runtime
reads to build the team; a package that omits it is published, charged for as
a team, and can never be called:
{
"entrypoints": { "orchestrator": "agents/00-orchestrator/agent.md" },
"roster": ["agents/10-<role>/agent.md", "agents/20-<role>/agent.md"]
}
Write both keys explicitly. The runtime also accepts older spellings
(entrypoint, orchestrator, entry; workers, members, team) and can
derive the roster from agents/<name>/agent.md, but relying on that leaves
the team's shape implicit and it drifts. Note that entry in agentlas.json
is the PACKAGE entrypoint, not the team manager — never reuse it for that.
-
Emit runtime adapters and package verification.
-
Run scripts/verify-team-package.sh <package-root> before reporting
completed. If it fails, do not hand off a result; correct the package by
adding an orchestrator/HQ plus company-blueprint topology or by collapsing
it to a valid single-agent package.
Output
Return team_topology, nodes, edges, memory_architecture, gates,
runtime_adapters, global_commands, and verification.