session-start

name	session-start
description	Use at the beginning of any new session — before any work begins — to establish effort tier, session goal, and skill routing. Triggers on: "start a session", "let's begin", "new session", "kick off", "where do we start", or whenever a fresh context needs calibration before a build, research, or planning task.

Session Start

The judgment layer entry point. Runs before any work — establishes the three variables every downstream skill needs: effort tier (which model to route to), goal clarity (what success looks like), and skill routing (which skills apply this session).

Announce at start: "Using session-start to calibrate this session."

---

Step 1 — Calibrate with AskUserQuestion

Run one AskUserQuestion call with these three questions together:

Question 1 — Effort tier
header: "Effort level"
question: "What's the effort level for this session?"
options:
  - label: "Quick exploration (Recommended for most sessions)"
    description: "Sonnet 4.6 throughout. Fast, focused, minimal token spend. Good for spikes, research, one-off tasks."
  - label: "Focused sprint"
    description: "Sonnet 4.6 for implementation, Opus 4.8 for plan writing and review. Best for shipping a real feature."
  - label: "Deep build"
    description: "Opus 4.8 throughout — planning, implementation review, loop design. Reserve for architectural decisions or high-stakes launches."
  - label: "Constrained (low tokens this week)"
    description: "Sonnet 4.6 only. Hand heavy implementation off as a written plan file to an external executor when tokens run low."

Question 2 — Domain
header: "Domain"
question: "What are you working on this session?"
options:
  - label: "New product idea"
    description: "Validating demand, deciding what to build, or writing a first spec."
  - label: "Existing product — new feature"
    description: "Designing, speccing, or implementing a specific feature in an existing codebase."
  - label: "Research or analysis"
    description: "Market research, competitor analysis, technical spike, data exploration."
  - label: "Infrastructure / operations"
    description: "Deployment, CI/CD, dependency audit, migrations, devops."

Question 3 — Outcome
header: "Done looks like"
question: "What does 'done' look like at the end of this session?"
options:
  - label: "A decision I can act on"
    description: "A recommendation with rationale I trust enough to move forward."
  - label: "A plan file I can hand off"
    description: "A written spec, PRD, or implementation plan ready for an implementation model or external executor."
  - label: "Working code merged or deployed"
    description: "A concrete, verified, shipped change."
  - label: "A brief or artifact I can share"
    description: "A doc, deck, report, or diagram for a stakeholder."

---

Step 2 — Set the session context

After collecting answers, announce the session context aloud and save it:

## Session Context — [date]

**Effort tier:** [Quick / Focused / Deep / Constrained]
**Model routing:** [Sonnet throughout / Sonnet impl + Opus plan+review / Opus throughout / Sonnet + external-executor fallback]
**Domain:** [new product / existing feature / research / infra]
**Done looks like:** [decision / plan file / merged code / artifact]
**Goal (from user):** [one sentence from their message or AskUserQuestion notes]

Save to ~/.claude/session-context.md (overwrite each session — this is a scratchpad, not memory).

---

Step 3 — Route to the right skill

Based on domain + outcome, suggest the skill chain to use:

Domain	Done looks like	Suggested chain
New product idea	Decision	/market-validation → /build-options
New product idea	Plan file	/market-validation → /build-options → /prd → /tasks
Existing feature	Plan file	/prd → /tasks → implementation
Research	Artifact	/market-validation (or direct research)
Infra	Merged code	/tasks (or inline — no chain needed)
Any multi-repo scope	Plan file	/ecosystem-planning

Announce: "This session I'll route to [skill chain]. Say the skill name to begin, or ask me to start directly."

---

Step 4 — Load memory

Before doing any work, read your Claude Code memory index (~/.claude/projects/<project>/memory/MEMORY.md, if it exists — see Memory). Summarize any items marked PENDING, BLOCKED, or awaiting-operator that are relevant to this session's domain. Surface them as a short list before proceeding.

---

Model routing reference

Tier	Fable planning	Implementation	Review
Quick	Sonnet 4.6	Sonnet 4.6	Sonnet 4.6
Focused	Opus 4.8	Sonnet 4.6	Opus 4.8
Deep	Opus 4.8	Opus 4.8	Opus 4.8
Constrained	Sonnet 4.6	Sonnet 4.6 → external executor	Sonnet 4.6

---

What this skill builds on

Anthropic primitive	Role in this skill
AskUserQuestion	Collects session context before any tokens are spent on work
Memory system (~/.claude/projects/*/memory/)	Surfaces pending items from prior sessions
Model routing (Opus / Sonnet / external executor)	Selected once here; downstream skills inherit the choice
Skills as judgment layer	This skill activates the right chain; it does not do the work itself

See: Agent SDK · Interactive mode / AskUserQuestion · Memory · Models overview