intent

name	intent
description	Run a task intent gate to classify the request and route to the right OmA stage/command.

Run OmA intent gate.

Input: $ARGUMENTS

Protocol:

1. [Pre-check] Detect depth keywords:
   • low, medium, high: High-fidelity mode.
   • skip: Explicitly skip interview.
2. [State Guard] If an unfinished interview is detected via .omg/state/interviews/active.json, prompt the user to $intent-resume or $intent-restart.
3. [Classification] Classify the primary intent into one:
   • team-assemble: Dynamic multi-role team setup.
   • plan: Planning ambiguity or roadmap needs.
   • prd: Scope/requirement ambiguity.
   • exec: Implementation-ready tasks.
   • verify: Validation or testing requests.
   • fix: Defect-only bug fixes.
   • research: Option exploration or analysis.
   • lifecycle: Mode or environment operations.
4. [Constraint Audit] Identify missing constraints or acceptance criteria needed before execution.
5. [Workspace Audit] Detect when multi-root, dirty-lane, or worktree-sensitive setup is needed.
6. [Routing Decision]:
   • If interview keywords are present: Transition to workflow_state: interviewing, invoke interview agent, and provide the First Socratic Question.
   • Otherwise: Detect command-intent mismatch and recommend the standard next OmA command.
7. [Persistence] If filesystem tools are available, write/update .omg/state/intent.md. When interview mode starts, initialize .omg/state/interviews/[slug]/context.json and point .omg/state/interviews/active.json at that session.

Routing hints:

• depth keywords (low, medium, high) detected -> /oma:interview (via interview) [Note: Flags like --deep are deprecated and replaced by keywords]
• canonical automated flow when implementation is expected -> /oma:team-assemble (orchestrates plan -> prd -> taskboard -> exec -> verify -> fix)
• manual staged flow for granular control -> /oma:team-plan -> /oma:team-prd -> /oma:taskboard sync -> /oma:team-exec -> /oma:team-verify -> /oma:team-fix
• multi-root, dirty-lane, or worktree-sensitive implementation -> /oma:workspace (use /oma:workspace audit when lane health or trust is unclear)
• dynamic multi-role team setup -> /oma:team-assemble
• planning ambiguity -> /oma:team-plan or $plan
• scope ambiguity -> /oma:team-prd or $prd
• implementation-ready -> /oma:team-exec or $execute
• discovery/trace-to-interview pass -> $deep-dive
• high-risk/quality-first planning -> $ralplan
• durable multi-goal workflow -> $ultragoal
• technical/API research -> $research
• context bloat/efficiency -> $context-optimize
• pattern extraction/session learning -> $learn
• validation request -> /oma:team-verify
• defect-only fix -> /oma:team-fix
• task drift or completion-proof drift -> /oma:taskboard
• mode/lifecycle operations -> /oma:mode, /oma:launch, /oma:checkpoint, /oma:stop

Response:

• Keep it concise and operator-facing.
• For interview mode: Include Intent Gate, Depth Detected, Confirmed Intent, and First Socratic Question.
• For standard mode: Include Intent Gate, Missing Inputs, Recommended Route, and Ready-to-Run Prompt (with --intent flag).