Roles and Stances

GDD defines three independent axes that shape how a session works: roles (what kind of work you're doing), stances (how the framework adapts its pace and ceremony), and mentoring (a composable boolean overlay that teaches as it works). These are set per session in the session file (ws session set …), not stored in Thalamus frontmatter.

Roles

Role	Focus
Developer	Writing and shipping code
Designer	Defining behavior (scenarios, specs)
Reviewer	Quality and safety
Scribe	Note-taking and vault work — PARA layout, frontmatter, daily notes, capture-process-organize loop

Roles aren't skill levels. A first-time contributor and a 20-year veteran can both be in the Developer role — they'll just have different stances active. The role describes the session's kind of work for both human and agent; the agent operates within it, bounded by the permission and access systems.

Today only the Scribe role has a dedicated skill file; the other roles are the assumed default and rely on stance and practice skills rather than a single role file.

Stances

Stances modify the ceremony level and pace of a session. They compose freely with any role and with the mentoring overlay.

Quick

Minimal ceremony for short time windows. 15 minutes on your phone between responsibilities? Quick stance suggests appropriately-sized tasks, recovers context fast, and skips questions it can infer.

Session sizing examples: - 15 min: Write one BDD scenario, review one PR comment, fix one small bug - 30 min: Implement step definitions, triage review findings, write a skill stub - 45 min: End-to-end small feature with PR

Stance philosophy: quick = minimize ceremony for small rapid interactions. A deferred future idea: quick could one day relax scoped guardrails behind loud disclaimers ("YOLO-lite") — out of scope for current GDD.

Zen

Full ceremony for deep focus sessions. Saturday morning deep dive? Zen stance leans into thorough brainstorming, comprehensive reviews, auditing accumulated concerns, and completing large chunks of work end-to-end.

Zen stance may proactively suggest housekeeping if observations have accumulated in the Thalamus.

Stance philosophy: zen = shed ceremony for intense focus WITHOUT bypassing any guardrail. Deep work, full integrity.

Flow

Adaptive ceremony for sessions that drift productively across multiple topics. The agent matches the human's rhythm, incorporates tangents naturally, and treats the Thalamus as a live collaboration surface rather than a session-bookend artifact. Often the natural default when no stance is set.

Stance philosophy: flow = allow meandering but nudge back toward the core objective — even one that shifts mid-session. The target shifts; the discipline of tracking it does not.

Mentoring

Mentoring is a boolean overlay, not a stance. It composes with any stance (Quick + Mentoring, Zen + Mentoring, Flow + Mentoring) and with any role.

When mentoring is on, the AI explains decisions, teaches practices in context, and offers more scaffolding. Not tied to seniority; anyone can activate it. First time touching BDD? Turn on mentoring, even after a decade of coding.

The AI's job with mentoring active is to grow the human, not just ship the code — every interaction is an opportunity to transfer understanding.

Tool-practice workflows pair naturally with mentoring. For example, gdd-k8s (the guarded-Kubernetes practice skill) activates whenever a session involves Kubernetes; with mentoring on, it additionally explains each cluster interaction and why the guard shapes it that way, rather than just running safely.