Vocabulary

Vocabulary

Process

Step 1: Gather terms

• Scan the current conversation for domain-relevant nouns, verbs, and concepts
• If $ARGUMENTS specifies a topic, focus extraction on that area
• Read .mpx/CONTEXT.md § Domain Language. If CONTEXT.md doesn't exist, report error and stop.
• Also scan: epic issues, README, key source files for domain terms

Step 2: Identify problems

• Flag ambiguities (same word used for different things)
• Flag synonyms (different words for the same concept)
• Flag vague terms (imprecise language that could cause confusion)

Step 3: Propose vocabulary

• Be opinionated — when multiple words exist for the same concept, pick ONE canonical term
• For each term: canonical name, one-sentence definition
• Group terms into natural clusters (by subdomain, lifecycle, or actor)
• Show relationships between terms with cardinality (e.g., "A User has many Sessions")

Step 4: Confirm with user

Present each candidate term with its full proposed entry text (**Term** — One-sentence definition.):

• New terms — show the exact text that would be written
• Updated definitions — show old → new
• Flagged ambiguities or conflicts
• Ask user to approve, edit, or reject each term

Only write confirmed terms.

Step 5: Write

Update .mpx/CONTEXT.md ## Domain Language section using definition-list format:

## Domain Language

**Workspace** — Top-level container: one GitHub repo + one project folder + one window.
**Issue** — Atomic work unit. One GitHub issue, one worktree, one branch, one color.

_Avoid_: "task" for Issue, "project" for Workspace.

## Relationships

- A **Workspace** has many **Issues** (1:N)
- An **Issue** has many **Sessions** (1:N)

## Flagged Ambiguities

- "workspace" was previously used for both the app container and VS Code workspace — resolved: **Workspace** is the Grovekeeper container only.

If updating: merge new terms into existing structure, update changed definitions, preserve terms that haven't changed.

Step 6: Summary

• Output inline: number of terms added, updated, and unchanged; reconcile these counts with the final file
• List any unresolved ambiguities for future discussion

Rules

• Only include domain terms — skip generic programming concepts (function, class, API, database)
• Keep definitions to ONE sentence maximum
• Flag conflicts explicitly — never silently resolve ambiguity
• Show relationships with bold term names and cardinality
• When re-running: read existing file, incorporate new terms, update definitions, re-flag ambiguities