Studio Project Intake

Overview

Turn an ambiguous game-studio request into a bounded task packet before implementation or diagnosis begins.

When to use

Use for new repositories, new sessions, cross-subsystem work, unclear ownership, risky changes, or tasks mentioning Unity, Lua, C++, services, databases, generated data, or external projects.

When NOT to use

Do not repeat full intake when a current task packet already has verified scope and the request is a micro-change inside that scope.

Required inputs and context discovery

Collect goal, user-visible success, workspace and nested Git roots, branch and dirty state per repository, engine/version, runtime topology, exact subsystem, risk tier, owners, dependencies, generated and excluded paths, do-not-touch paths, available commands, optional code-intelligence provider/version/index freshness, and blocked capabilities. For that optional lane, record provider/version, repository/revision/worktree binding, index state, capability, required languages, artifacts, side effects, and blocker.

Safety and risk level

Intake is read-only. Inspect status and structure without starting services, opening editors, importing databases, or modifying project files.

Workflow

1. Identify the requested outcome and the smallest repository boundary that can deliver it. Completion criterion: the goal and excluded scope are one sentence each.
2. Inspect Git status, applicable instructions, project maps, manifests, and relevant subsystem entry points. Completion criterion: current state and ownership risks are captured without mutation.
3. Classify risk as read-only, low, medium, or high and list forbidden actions. Completion criterion: each planned side effect has an approval and rollback requirement.
4. Record engine/version, services, ports, database/schema, generated-source relationships, and test commands only when verified. Completion criterion: snapshots are separated from hypotheses.
5. Select the next workflow skill and, for verified multi-repository work, include a conservative project-profile draft with unknown owners and commands labeled. Completion criterion: the task packet is executable without rediscovery and routes through studio-workspace-routing when applicable.
6. The graph lane is optional. Apply code-intelligence-contract only when graph context or capability is requested or materially useful for scoped intake; provider existence alone never triggers it. Pure intake remains intake, and the optional graph lane may stay absent or BLOCKED without failing intake; intake may still succeed when the graph lane is BLOCKED because intake owns scope discovery, not a graph verdict. Completion criterion: the packet records provider/version, repository/revision/worktree binding, index state, capability, required languages, artifacts, side effects, and blocker while keeping graph extraction separate from source/runtime verification.

Evidence and output contract

Produce task-packet.json with goal, scope, risk, repository snapshot, owners, do-not-touch paths, dependencies, verification commands, expected artifacts, optional graph-lane state, and BLOCKED items.

Handoff contract

If intake cannot resolve a required fact, hand off the exact question, inspected paths, evidence label, and the safest next read-only action.

Pitfalls and anti-rationalization

• Do not infer engine or server versions from folder names alone.
• Do not treat credentials, ports, or DB names in old docs as current facts.
• Do not expand scope because adjacent cleanup looks useful.

Verification checklist

• Goal and excluded scope are explicit.
• Owners and do-not-touch paths are listed.
• Risk and rollback expectations are assigned.
• Snapshot facts cite inspected paths or commands.
• The next workflow skill is named.

References and scripts

Use project-local AGENTS.md, HANDOFF.md, .agents/CONTRACT.md, project maps, and registry/capabilities.yaml when present.