Common Ground

Common Ground

Arguments: $ARGUMENTS

Purpose

Claude often operates on assumptions about project context, technology choices, coding standards, and user preferences. This command surfaces those assumptions for explicit user validation, preventing misaligned work based on incorrect premises.

Argument Parsing

Parse arguments to determine mode:

Reference Guide

Load detailed guidance based on context:

Project Identification

Before any operation, determine the project identity:

1. Try git remote:

   git remote get-url origin 2>/dev/null
   

   If found, use the URL as project identifier (e.g., github.com/user/repo)

2. Fallback to path: Use the current working directory absolute path

Store this as {project_id} for file operations.

Default Mode: Surface & Adjust

When no flags provided, execute the two-phase interactive flow.

Phase 1: Surface & Select

1. Analyze current context to identify assumptions:

   • Scan configuration files (tsconfig.json, package.json, .eslintrc, etc.)
   • Review recent conversation context
   • Check existing ground file for tracked assumptions

2. Classify each assumption by type and proposed tier:

   • See references/assumption-classification.md for classification rules

3. Present to user via AskUserQuestion:

   Present assumptions grouped by category. Use multiSelect to let user choose which to track.

   Question format:

   I've identified assumptions about this project. Which should I track?
   

   Options (up to 4 categories, user selects via multiSelect):

   • Architecture & Tech Stack assumptions
   • Coding Standards assumptions
   • Testing & Quality assumptions
   • Other/Uncertain items

   User can select multiple categories or use "Other" to specify individual items.

4. Handle uncertain items: For any [uncertain] assumptions, ask direct clarifying questions.

Phase 2: Adjust Tiers

1. Show selected assumptions with proposed tiers

2. Present tier adjustment options via AskUserQuestion:

   Question format:

   Review the confidence tiers. Any adjustments needed?
   

   Options:

   • Accept all proposed tiers
   • Promote some to higher confidence
   • Demote some to lower confidence
   • Add new assumptions

3. Process adjustments:

   • Promotions: OPEN -> WORKING -> ESTABLISHED
   • Demotions: ESTABLISHED -> WORKING -> OPEN
   • New additions: User specifies via "Other" with format: assumption text [tier] [type]

4. Write ground file:

   • Save to ~/.claude/common-ground/{project_id}/COMMON-GROUND.md
   • Update ground.index.json for machine-readable access
   • See references/file-management.md for file formats

Output

## Common Ground Complete

**Project:** {project_name}
**Tracked Assumptions:** {count}

### Summary
- ESTABLISHED: {count} (high confidence)
- WORKING: {count} (medium confidence)
- OPEN: {count} (needs validation)

**Ground file saved to:** ~/.claude/common-ground/{project_id}/COMMON-GROUND.md

Run `/common-ground --list` to view all assumptions.
Run `/common-ground --check` for quick validation.

--list Mode

Read-only display of all tracked assumptions.

1. Load ground file from ~/.claude/common-ground/{project_id}/COMMON-GROUND.md

2. Display assumptions grouped by tier:

## Common Ground: All Assumptions

**Project:** {project_name}
**Last Updated:** {timestamp}

### ESTABLISHED ({count})
1. {title} - {assumption} [{type}]
2. ...

### WORKING ({count})
1. {title} - {assumption} [{type}]
2. ...

### OPEN ({count})
1. {title} - {assumption} [{type}]
2. ...

---
(Read-only view. Run `/common-ground` to modify.)

3. Handle missing file: If no ground file exists:

   No ground file found for this project.
   Run `/common-ground` to surface and track assumptions.
   

--check Mode

Quick validation of existing assumptions.

1. Load ground file from ~/.claude/common-ground/{project_id}/COMMON-GROUND.md

2. Present summary via AskUserQuestion:

   Question format:

   Quick check: Are these assumptions still valid?
   

   Options:

   • All still valid
   • Some need updates
   • Need full review

3. Handle responses:

   • All valid: Update last_validated timestamp, confirm
   • Some need updates: Ask which ones, then enter Phase 2 of default flow
   • Need full review: Run full default flow

4. Handle missing file: If no ground file exists, redirect to default flow:

   No existing assumptions to check. Starting fresh...
   

   Then execute default mode.

--graph Mode

Generate a mermaid diagram showing Claude's reasoning structure—not just assumptions, but the decision tree that led to the current approach.

Purpose

Make the shape of Claude's reasoning visible:

• Decision points and branches considered
• Paths taken vs alternatives
• Where uncertainty lives in the reasoning chain

Flow

1. Run standard common-ground flow (if no existing ground file, execute default mode first)

2. Analyze reasoning structure behind confirmed assumptions:

   • What decision points led to these assumptions?
   • What alternatives were considered at each branch?
   • What confidence level exists at each node?

3. Generate mermaid diagram following conventions in references/reasoning-graph.md

4. Output files:

   • Update COMMON-GROUND.md with embedded ## Reasoning Graph section
   • Optionally create standalone REASONING.mermaid in project root

Output Format

The reasoning graph is embedded in COMMON-GROUND.md:

## Reasoning Graph

```mermaid
flowchart TD
    ROOT[Task: {task_description}] --> D1{Decision Point?}
    D1 -->|"weight: 0.8 [inferred]"| P1[Chosen Path]
    D1 -->|"weight: 0.2 [alternative]"| P2[Alternative]
    ...

### Conversational Interaction

Since mermaid is text-based, graph manipulation happens conversationally:

| User Says | Claude Action |
|-----------|---------------|
| "Expand the {branch} branch" | Regenerate graph with that path elaborated |
| "Why not {alternative}?" | Explain reasoning, potentially adjust weights |
| "I actually want {alternative}" | Update graph with that branch as chosen |
| "What's downstream of {node}?" | Expand that subtree |

After each significant interaction, regenerate the graph showing updated state.

### Example Output

--graph Complete

Project: {project_name} Reasoning Nodes: {count} Decision Points: {count}

Graph Summary

• Root: {task_description}
• Major Decisions: {count}
• Open Questions: {count} nodes with uncertain status

Graph embedded in: ~/.claude/common-ground/{project_id}/COMMON-GROUND.md

Run /common-ground --list to view assumptions. Run /common-ground --graph to regenerate after changes.

See `references/reasoning-graph.md` for detailed mermaid conventions and node styling.

---

## Constraints

### MUST DO
- Always identify project before file operations
- Use AskUserQuestion for all interactive selections
- Preserve assumption type (audit trail) - users cannot change type
- Write both human-readable (COMMON-GROUND.md) and machine-readable (ground.index.json) files
- Include timestamps for tracking staleness
- When using --graph, show decision points that led to assumptions
- Preserve alternative branches in graph (grayed out) for context

### MUST NOT DO
- Assume context without surfacing assumptions
- Allow type changes (stated/inferred/assumed/uncertain)
- Proceed without user confirmation on tier changes
- Overwrite ground file without preserving history
- Generate graphs without first having confirmed assumptions
- Remove alternative branches from graph (preserve for exploration)