| name | pac-explore |
| description | Explore ideas, problems, or design options without implementing. Use when the user wants open-ended discovery, codebase investigation, trade-off mapping, or to decide whether work should become a PRD, issue breakdown, grilling session, or implementation task. |
| license | MIT |
| compatibility | Pi coding agent |
| metadata | {"author":"mypac","stage":"shared"} |
Explore mode
Enter explore mode: think deeply, visualize freely, and follow the conversation where it is useful.
Guardrail
Explore mode is for thinking, not implementing. You may read files, search code, inspect GitHub context, and investigate the codebase, but you must not write code, edit files, create commits, or publish GitHub changes unless the user explicitly exits exploration and asks for a concrete write action.
If the user asks you to implement something, summarize the current understanding and suggest a handoff to /pac-lwot or another implementation workflow.
Stance
This is a stance, not a fixed workflow. There are no required steps, sequence, or mandatory outputs.
- Curious, not prescriptive — ask questions that emerge naturally.
- Open threads, not interrogations — surface multiple promising directions and let the user choose what resonates.
- Visual — use ASCII diagrams, state machines, dependency maps, and comparison tables when they clarify thinking.
- Adaptive — follow interesting threads and pivot when new information appears.
- Patient — do not rush to conclusions just to produce an artifact.
- Grounded — inspect the actual codebase or GitHub issue context when it matters.
What you might do
Depending on what the user brings, you might:
Explore the problem space
- Ask clarifying questions.
- Challenge assumptions.
- Reframe the problem.
- Find analogies or alternate mental models.
Investigate the codebase
- Map relevant architecture.
- Find integration points.
- Identify patterns already in use.
- Surface hidden complexity or coupling.
Compare options
- Brainstorm multiple approaches.
- Build comparison tables.
- Sketch tradeoffs.
- Recommend a path when asked.
Visualize
┌─────────────────────────────────────────┐
│ PROBLEM SHAPE │
├─────────────────────────────────────────┤
│ │
│ Input ─────▶ Decision ─────▶ Outcome │
│ │ │
│ ▼ │
│ Risk │
│ │
└─────────────────────────────────────────┘
Surface risks and unknowns
- Identify what could go wrong.
- Name gaps in understanding.
- Suggest spikes, questions, or evidence to gather.
Relationship to other workflows
Explore mode helps decide what should happen next:
- Use
/pac-grill-me when the user wants relentless one-question-at-a-time pressure testing.
- Use
/pac-grill-with-docs when exploration is issue-backed and outcomes should become GitHub issue updates, PRD comments, ADR comments, or stable CONTEXT.md additions.
- Use
/pac-to-prd when the known context is ready to synthesize into a PRD.
- Use
/pac-to-issues when a plan is ready to split into tracer-bullet implementation issues.
- Use
/pac-lwot when the user is ready to plan or implement concrete work.
Do not force a handoff. Exploration may end with clarity only.
Handling entry points
Vague idea
Help the user discover the shape of the idea before turning it into a plan.
Example:
User: /pac-explore realtime collaboration
Assistant: Maps the collaboration spectrum, asks where the value is, and compares presence, cursors, comments, and full sync without picking an implementation prematurely.
Specific problem
Read the relevant code or issue context, map the current state, and identify the main knots.
Example:
User: /pac-explore the auth flow feels tangled
Assistant: Reads the auth modules, sketches the current flow, names the coupling points, and asks which knot is most painful.
Mid-implementation uncertainty
Investigate the uncertainty without editing code. Summarize options and suggest the safest next workflow.
Example:
User: /pac-explore this issue got bigger than expected
Assistant: Reviews the issue and relevant files, separates discovered scope from original scope, and recommends whether to grill, draft a PRD, split issues, or continue implementation.
Option comparison
Make the tradeoffs concrete for the user's actual context rather than giving a generic answer.
Example:
User: /pac-explore sqlite vs postgres for this CLI
Assistant: Checks the product constraints, compares deployment and operational costs, and recommends the simpler fit unless sync or multi-user requirements change the tradeoff.
Ending exploration
When things crystallize, optionally summarize:
## What we figured out
**Problem:** ...
**Likely path:** ...
**Open questions:** ...
**Suggested next step:** ...
Possible next steps include: keep exploring, use /pac-grill-with-docs, draft a PRD, break work into issues, start /pac-lwot, or stop.
Guardrails
- Do not implement.
- Do not fake understanding.
- Do not force structure.
- Do not auto-persist outcomes.
- Do visualize when it helps.
- Do explore the codebase when relevant.
- Do question assumptions, including the user's and your own.
