intent-detection

name	intent-detection
description	Routing Ruby/Rails/Grape/Sidekiq/AR/Sequel/Hotwire work to the correct /rb: command BEFORE doing it. Pushy gateway, never blocks. Triggers: "my Rails app", "which command", multi-step task. Do NOT use when user typed /rb:<name>.
user-invocable	false
effort	low

Intent Detection — Workflow Routing

When user describes work WITHOUT specifying a /rb: command, analyze their intent and suggest the appropriate workflow BEFORE starting work.

Routing Table — Visible Skills

Signal phrase	Detected intent	Workflow target
"bug", "error", "crash", "failing", "broken", stack trace	Bug investigation	/rb:investigate
"add", "implement", "build", "create" + multi-step	New feature	/rb:plan
"review", "check" code	Code review (changed files)	/rb:review
"audit code", "audit project", "codebase health"	Project audit (whole codebase)	/rb:audit
"fix" + small/specific scope	Quick fix	handle directly or /rb:quick
"refactor", "clean up", "improve"	Refactoring	/rb:plan (needs scope)
"research", "how to", "what's the best"	Research	/rb:research
"evaluate", "compare", "adopt", "library", "should we use"	Library evaluation	/rb:research --library
"test", "spec", "coverage"	Testing	handle directly or /rb:plan
Describes 1-2 file changes, < 50 lines	Small task	handle directly
"deploy", "release", "production"	Deployment	/rb:verify then deploy
"performance", "slow", "N+1", "memory"	Performance	/rb:perf
"PR review", "review comments", "address feedback", "respond to PR"	PR response	/rb:pr-review
"permission prompts", "stop asking", "settings.json", "allow commands", "permission fatigue"	Permission tuning	/rb:permissions
"that worked", "fixed it", "problem solved"	Knowledge capture	/rb:compound
"enhance plan", "more detail", "deepen"	Plan enhancement	/rb:plan --existing
"triage", "which findings", "prioritize fixes"	Finding triage	/rb:triage
"brainstorm", "explore options", "not sure how", "multiple approaches", "what's the best way to architect"	Design exploration	/rb:brainstorm

Routing Table (DMI roster — slash-only)

Skills below carry disable-model-invocation: true. They never auto-load through the routing prompt; reach them via /rb:<name> slash invocation when the signal matches. Suggest the right command when triggers appear.

Signal	Detected Intent	Suggest
"interview", "brainstorm", "discovery", "explore ideas"	workflow on-ramp; pre-plan discovery	/rb:brainstorm
"adversarial review", "devil's advocate", "stress test"	adversarial-mode review	/rb:challenge
"YARD", "RDoc", "ADR", "write docs"	post-implementation docs	/rb:document
"show examples", "codebase patterns", "representative code"	codebase pattern surface	/rb:examples
"tutorial", "getting started", "what can you do"	onboarding	/rb:intro
"learn from this", "remember this mistake", "correction"	in-flight lesson capture	/rb:learn
"slow", "latency", "memory", "p95", "queue backup"	performance analysis	/rb:perf
"permissions", "allow command", "approval prompts", "settings.json"	permission tuning	/rb:permissions
"PR feedback", "review comments", "reviewer said"	PR review-comment handling	/rb:pr-review
"quick fix", "typo", "one-liner"	trivial-fix path	/rb:quick
"rubydoc", "gem docs", "API reference", "Rails Guides"	cheap doc lookup	/rubydoc-fetcher
"secrets", "leaked", "API key", "credentials", "betterleaks"	pre-push secret scan	/rb:secrets
"do everything", "full lifecycle", "end to end", "hands-off"	full lifecycle orchestration	/rb:full
"audit project", "codebase health", "architecture review"	project-wide audit	/rb:audit
"compression report", "share compression stats"	internal QA (telemetry report)	/rb:compression-report
"scan provenance", "audit research quality", "trust state"	research-trust audit	/rb:provenance-scan
"service boundaries", "split monolith", "fat controller", "package boundary"	service-boundary analysis	/rb:boundaries
"request state", "CurrentAttributes", "session leak", "Redis state"	request-state hygiene	/rb:state-audit
"Tidewave", "live runtime", "introspect models", "running app"	Tidewave runtime introspection	/rb:runtime
"tech debt", "cleanup", "callback sprawl", "stale code", "overgrown service", "decorative abstraction"	tech-debt logging	/rb:techdebt
"Karafka", "Kafka consumer", "event streaming"	Kafka consumer patterns	/rb:karafka
"Hotwire Native", "Turbo Native", "iOS bridge", "Android Hotwire"	native mobile Hotwire	/rb:hotwire-native
"async gem", "fiber concurrency", "Falcon server", "I/O-bound"	fiber concurrency	/rb:async-patterns
"dry-rb", "dry-validation", "dry-types", "dry-monads"	dry-rb gem patterns	/rb:dry-rb-patterns
"Sequel", "Sequel ORM", "Sequel dataset", "Sequel migration"	Sequel ORM patterns	/rb:sequel-patterns
"Kamal deploy", "Docker deploy", "Thruster", "release Rails"	deployment configuration	/rb:deploy
"discovery report", "trigger-table tuning", "injection telemetry"	internal eval-tuning tool — drafts a redacted report from skill-discovery telemetry	/rb:discovery-report

Behavior

1. Read user's first message
2. Match against routing table (use keyword + context signals, not exact match)
3. If lock (single valid route): suggest the command directly — "This looks like [intent]. I'd suggest [command] — want me to run it, or should I just dive in?"
4. If fork (multiple valid routes): present top 2-3 options with one-sentence rationale each
5. If trivial task (typo, single-line fix, config change): skip suggestion, just do it
6. If user already specified a /rb: command: follow it, don't re-suggest
7. NEVER block the user — suggestion only, not mandatory

Routing Modes

Lock (single valid route — act immediately)

High confidence, one right answer. Don't deliberate — suggest directly:

• Stack trace or error message pasted → /rb:investigate
• "Fix CI" / "fix rubocop" → auto-fix pattern
• "Run checks" → /rb:verify
• "Permission prompts annoying" → /rb:permissions
• "That fixed it" → /rb:compound
• "Add [feature] with [multiple components]" → /rb:plan
• "Review my changes" or "check this PR" → /rb:review

Fork (multiple valid routes — don't pick silently)

Genuinely ambiguous. Present top 2-3 options with one-sentence rationale each:

• "Work on billing feature" → /rb:plan or /rb:work depending on whether plan exists
• "Improve the auth flow" → /rb:plan (scope unclear) or /rb:investigate (if bugs mentioned)
• "Clean up this module" → /rb:quick (small) or /rb:plan (large) — ask about scope
• "Fix [thing]" — could be quick or complex, suggest based on scope description
• "Update [thing]" — could be small edit or refactor

Trivial (just do it)

Low confidence or obvious scope. Don't suggest a workflow:

• Single file mentioned, clear change
• "Change X to Y"
• Configuration or dependency updates

Complexity Signals

When a task matches a workflow command, check complexity before suggesting:

Trivial signals (suggest /rb:quick or handle directly):

• Single file mentioned explicitly
• "exclude X from Y", "add X to config", "rename", "change X to Y"
• Problem + solution both stated ("X is wrong, change to Y")
• One-line fix described

Complex signals (suggest /rb:plan or /rb:investigate):

• 3+ modules or files mentioned
• "intermittent", "race condition", "sometimes", "random"
• Stack trace with 5+ frames
• "across", "all", "every" (scope indicators)

Override rule: If user invokes /rb:full but task matches trivial signals: "This looks like a quick fix. Want /rb:quick instead, or stick with the full cycle?"

Iron Laws

1. NEVER block on suggestion — If user starts explaining, just do the work
2. One suggestion max — Don't re-suggest if user ignores first suggestion
3. Commands are shortcuts, not gates — All work can be done without commands

Integration

This skill is consulted at session start. It works alongside:

• SessionStart hook (shows plugin loaded message + injects Iron Laws + Preferences)
• Individual workflow skills (activated by commands)