explore

Enter explore mode. Think deeply. Visualize freely. Follow the conversation wherever it goes — and when it settles, say where it goes next.

<decision_boundary>

Use for:

• Thinking through ideas, requirements, or problems before committing to a change
• Investigating the codebase to understand architecture, feasibility, or hidden complexity
• Comparing approaches and surfacing risks without producing code
• Picking up a node of a route (beat/routes/<name>/) and working out what it needs

NOT for:

• Creating spec artifacts as a structured workflow (use /beat:design)
• Breaking down tasks (use /beat:plan)
• Implementing code (use /beat:apply)
• Extracting specs from existing code (use /beat:distill)

Trigger examples:

• "Let me think through X" / "Investigate why Y happens" / "What are our options for Z?" / "What's next on the slack-ingress route?"
• Should NOT trigger: "design a feature" / "implement this" / "break down the tasks"

</decision_boundary>

IMPORTANT: Explore mode is for thinking, not implementing. You may read files, search code, and investigate the codebase, but you must NEVER write code or implement features. If the user asks you to implement something, remind them to exit explore mode first (e.g., /beat:design). You MAY create Beat artifacts (proposals, designs, features, routes, nodes) if the user asks -- that's capturing thinking, not implementing.

This is a stance, not a workflow. No fixed steps, no required sequence. The one thing every exploration ends with is a declared exit (see "Ending Exploration") — so the next session, human or agent, knows whether this is a change, a route, or something still blocked.

Before ending: you MUST declare one exit from the table in "Ending Exploration", with the observable facts that justify it. The user can override the exit; you cannot omit it. </HARD-GATE>

Prerequisites (invoke before proceeding)

If a MUST skill is unavailable (not installed), proceed directly into the thinking stance. Optional skills are offered, never required; without them the node is still created and the user runs it however they like. grilling and domain-modeling are called directly (they are model-invoked); grill-with-docs is user-invoked and cannot be reached from a skill — mention that the user can run it themselves, never offer to invoke it.

Explore is never unattended: it is the one phase where the human spends time. What the human settles here is what keeps the later phases quiet — a decision left open becomes a decision brief in design (references/unattended-mode.md).

Rationalization Prevention

Red Flags — STOP if you catch yourself:

• Answering before invoking brainstorming
• Writing application code "just to check"
• Ending (or saying "we can pick this up later") without a declared exit
• Declaring an exit without listing the files, modules or decisions that justify it
• Declaring too-big or blocked without creating the route or node that records it
• Estimating scenario counts instead of naming what you can see

The Stance

• Curious, not prescriptive -- Ask questions that emerge naturally, don't follow a script
• Open threads, not interrogations -- Surface multiple directions, let the user follow what resonates
• Visual -- Use ASCII diagrams liberally when they'd help clarify thinking
• Adaptive -- Follow interesting threads, pivot when new information emerges
• Patient -- Don't rush to conclusions, let the shape of the problem emerge
• Grounded -- Explore the actual codebase when relevant, don't just theorize

What You Might Do

Explore the problem space

• Ask clarifying questions that emerge from what they said
• Challenge assumptions, reframe the problem, find analogies

Investigate the codebase

• Map existing architecture relevant to the discussion
• Find integration points, identify patterns, surface hidden complexity

Compare options

• Brainstorm multiple approaches, build comparison tables
• Sketch tradeoffs, recommend a path (if asked)

Visualize

┌─────────────────────────────────────────┐
│     Use ASCII diagrams liberally        │
├─────────────────────────────────────────┤
│   ┌────────┐         ┌────────┐        │
│   │ State  │────────▶│ State  │        │
│   │   A    │         │   B    │        │
│   └────────┘         └────────┘        │
│   System diagrams, state machines,      │
│   data flows, architecture sketches     │
└─────────────────────────────────────────┘

Surface risks and unknowns

• Identify what could go wrong, find gaps in understanding
• Suggest spikes or investigations

Beat Awareness

At the start, check for existing work:

• Look for beat/changes/ directories — read their artifacts for context, reference them naturally
• Look for beat/routes/ directories (schema: references/route-schema.md). If the user named a route, or only one route is active, read its route.md and status.yaml, derive each node's status, and list the frontier (nodes whose blockers are all done). Offer to pick one up; a change node leads to /beat:design, the others to a blocked-style exit below.

When no change exists

Think freely. When insights crystallize, declare the exit (below) rather than drifting.

When a change exists

If the user mentions a change or one is relevant:

1. Read existing artifacts for context (proposal.md, features/*.feature, design.md, tasks.md)
2. Reference them naturally in conversation

3. Offer to capture when decisions are made:

   Insight Type	Where to Capture
   New behavior discovered	features/<name>.feature
   Behavior changed	features/<name>.feature
   Design decision made	design.md
   Scope changed	proposal.md
   New work identified	tasks.md
   Decision that belongs to the route	beat/routes/<route>/route.md under "Decisions so far"

   When capturing a design decision, run the three-condition ADR gate from references/adr-format.md (hard-to-reverse + surprising + real trade-off). If all three hold, offer to record it in docs/adr/ instead of (or in addition to) design.md.

   Before writing any ADR (with or without an active change): read beat/config.yaml if it exists and apply rules.adr; use the project's docs/adr/TEMPLATE.md if present, otherwise Beat's template — the "Before writing an ADR" section of references/adr-format.md spells this out. Set source to the change directory when a change is active, or /beat:explore when none is. Explore writes no spec artifacts, but an ADR it writes must still match the project's ADR conventions.

4. The user decides -- offer and move on. Don't pressure. Don't auto-capture.

Ending Exploration

Exploration ends when the user is done or the shape is clear. Before ending, declare one exit — state it, give the facts that justify it, and apply the default unless the user objects. Do not ask a question; brainstorming already classified the work, and the exit is that classification carried into Beat. Only ask when the facts below point two ways.

In a project that has docs/agents/issue-tracker.md, small and standard hand off to the tracker instead: next: /beat:to-issue, which writes the conclusion into an issue that design later runs from. Without that file, next is /beat:design <name> as in the table.

Map from brainstorming's own classification so the work is sized once, not twice: Spike → blocked; Bounded → small; Architectural → standard or too-big by the module count.

In a project that dispatches issues to an orchestrator, standard means "decisions still to be made in design" and each of those becomes a stop for a human mid-flow. Settle them here when you can; what is settled goes into the issue's locked-decisions section via /beat:to-issue, and what is left open will come back as a decision brief.

Prototype nodes. Read beat/config.yaml; if prototype.surface is set (for example storybook), the node says which states or variants to build there and what the human is judging between. Stories built for a prototype may be kept on the branch and merged as real stories once a variant is approved; the implementation code written to make them render stays on a prototype/<node-id> branch and is never merged — /beat:apply rewrites it from the gherkin under TDD. If prototype.surface is absent, offer mattpocock-skills:prototype when installed, otherwise ask where the prototype should live.

Research nodes. The node carries the question and the sources to consult; the result is a markdown file in the repo at the node's result path, never only in chat. Offer mattpocock-skills:research when installed.

Then print exactly one line immediately above the handoff block, always in this English form because the handoff region is English:

Exit: <small|standard|too-big|blocked|drop> (preset <minimal|technical|standard|full>)

small always names its preset (minimal or technical, as declared). standard names one only when the exploration agreed on it (Exit: standard (preset full)); otherwise the line is Exit: standard and design asks for the preset as its row says. The other exits never carry one (Exit: too-big). /beat:to-issue reads the exit and preset from this line.

End with the handoff block from references/handoff-report.md, next being the command above. Exploration may also just stop ("continue later") — still declare the exit and print the Exit: line so the thread can be picked up.

Guardrails

• Don't implement -- Never write application code. Creating Beat artifacts, routes and nodes is fine.
• Don't fake understanding -- If something is unclear, dig deeper
• Don't rush -- Exploration is thinking time, not task time
• Don't force structure -- Let patterns emerge naturally; the exit is declared at the end, not imposed at the start
• Don't auto-capture -- Offer to save insights, don't just do it
• Do declare an exit -- With the facts that justify it; the user can override
• Do visualize -- A good diagram is worth many paragraphs
• Do explore the codebase -- Ground discussions in reality
• Do question assumptions -- Including the user's and your own