/ork:auto

Auto Intent-classified router, the front door to OrchestKit and the DEFAULT entry point for any goal-shaped request. Classifies a plain-English goal and routes it to the right specialist skill. Routing is never overhead, so use it even when the target skill seems obvious; skip only when already executing inside another skill (no recursion).

/ork:auto — Intent Router

The front door to OrchestKit. You describe a goal in plain English; the router classifies it and hands off to the right specialist. One entry point, many execution paths.

Why this exists: OrchestKit has 105 skills, but usage telemetry shows users fire only the handful they can name by memory (10 distinct skills across thousands of sessions). The dominant cause of "dead" skills is no front door — not low quality. This router turns "you must know the exact /ork:<name>" into "describe what you want."

Core principle: routing is a deterministic workflow, not an autonomous agent (Anthropic, Building Effective Agents). Classify → confirm → hand off. The router never does the work itself — it picks who does.

When to use

By default, for any goal-shaped request. An unambiguous goal is a 1-step route: auto classifies, confirms in one line, and hands off — no extra hops, so there is no "too obvious for auto".

Design note (2026-07-12): this table previously said "Go direct when you already know the skill / the request maps unambiguously to one". That inverted instruction made the front door structurally unreachable — a competent model always believes it knows the target, so the router recorded near-zero invocations across thousands of sessions (the exact dead-skill problem the "Why this exists" note above describes). Routers must be framed as the default path, not an escape hatch for confusion.

Intent categories → OrchestKit skill

Full per-category parameter extraction + edge cases: references/routing-rules.md.

Model weight (orthogonal second dimension)

Intent picks who does the work. Weight picks how expensive that worker should be. A route is \{intent\} @ \{weight\}. Weight never changes the intent and never replaces it. The taxonomy above and the 7 disambiguation rules are untouched by it.

Tiers are the ones already declared in src/agents/*.md frontmatter (haiku 7 · sonnet 10 · opus 6 · inherit 13). No parallel taxonomy.

Resolution is asymmetric: ANY heavy signal ⇒ Heavy; Light requires ALL light signals; everything else is Standard. Under-powering a security review yields a confident wrong answer nobody catches; over-powering a rename only wastes money.

Weight is per leg, not per route. A PR review can be a Heavy security leg plus a Light lint leg. Full signal table, per-intent defaults, and the honest limits of this lever: references/routing-rules.md.

This is the selector, not the cap. src/hooks/src/pretool/task/team-size-gate.ts is an ex-post, per-session counter keyed on ORK_TEAM_OPUS_MAX (default 8). Its default posture is advisory (outputWarning); with ORK_TEAM_SIZE_HARD=1 it escalates to outputDeny and refuses the spawn outright. Either way it reads the model read-only: it can refuse a premium spawn, but it cannot choose a cheaper one for you. Routing is what chooses. The two compose, cap as backstop and routing as selector; never duplicate the cap's counting here.

The flow

  CLASSIFY  ->  CONFIRM  ->  HAND OFF
     |            |             |
  intent       show the     invoke the
  + weight     route        target skill;
  out loud     + nod        follow ITS phases

1. Classify (reason out loud first)

State your reasoning before committing to a route — this triggers chain-of-thought and is the single biggest accuracy lever (Anthropic, Writing Effective Tools for Agents). Example: "'get latency under 200ms' names a metric + a direction → optimize, not fix."

Apply the disambiguation rules (most specific wins; explicit verb beats inferred intent). The load-bearing one: explicit verb wins — "Fix the slow query" → fix, not optimize. For the full ordered ruleset (all 7, including the truly-ambiguous fallback), references/routing-rules.md is canonical.

Then classify weight in the same pass, naming the signal that decided it: "touches auth and models an attacker → Heavy." Intent first, weight second; a weight call never rewrites the intent you just committed to.

2. Confirm (low ceremony)

Show the chosen route in one line and get a nod before handing off:

Goal:   "{original goal}"
Intent: {category}
Weight: {Light|Standard|Heavy} ({tier}), decided by: {signal}
Route:  {/ork:skill or loop} {extracted args}
        [run] · [adjust] · [cancel]

For low-risk single-pass routes (verify, review), an inline "routing you to /ork:verify — ok?" is enough. Never hand off without a nod.

Premium spend is never silent. Routing down (Light/Standard) needs no approval, because spending less is not a decision the user has to make. Routing up to Heavy is premium spend and gets its own line the user must accept:

⚠️  Heavy route: {N} opus-tier leg(s), triggered by: {heavy signal}
    [approve premium] · [run Standard instead] · [cancel]

If they decline, run Standard and say plainly which check is weakened. Never upgrade mid-handoff or inside a spawned agent the user did not see.

3. Hand off

Invoke the target skill with the extracted parameters and follow that skill's own phases and guardrails — do not override them. The router's job ends at the handoff; the specialist owns execution and its own report.

A hand-off is a Skill-tool invocation, not a recommendation. The failure mode that motivated M170 (#3127): telemetry traced 43 router hand-offs and found ZERO reached an executor skill (/ork:implement, /ork:cover, /ork:fix-issue, /ork:review-pr) — the route was named in chat, then the work happened inline in the main loop, so the executors' specialist wiring (implement → backend-system-architect, cover → test-generator) never activated. Therefore:

• Once the user nods, the SAME turn must contain the Skill-tool call for the routed skill. Never end the routing turn with only a description of what will run.
• Doing the routed work inline "because it's faster" is a routing failure, not a shortcut — the executor's parallel specialists and guardrails are the point of routing.
• If the routed skill genuinely cannot run (missing prerequisite, wrong repo state), say exactly that and stop; do not silently absorb the work into the main loop.

Fallback + honest gaps

• Fallback category. If no category clears a confident threshold, ask exactly ONE clarifying question rather than guessing. A rising fallback rate is the leading indicator that the taxonomy needs work — surface it, don't bury it.
• optimize has no dedicated skill (yet). OrchestKit's metric-driven optimization runs as a /goal loop using the loop recipe library (/ork:prd-to-goal → references/recipe-library.md). Route optimize there and say so plainly — don't pretend a /ork:experiment skill exists.
• improve-skill routes to the evolution gate. Self-optimizing a SKILL.md goes through the champion/challenger holdout-promotion gate (/ork:assess evals + evolution-engine), not a one-shot edit. It requires a benchmark + holdout set first.

Stacked invocation (CC 2.1.199+)

/skill-a /skill-b <goal> loads all leading skills (up to 5) into context at once; the trailing args belong to the whole stack. So /ork:auto /ork:brainstorm <goal> pre-loads the specialist alongside the router — useful when the user already knows part of the route. The router still owns classification and handoff; a pre-loaded specialist does not bypass the confirm step.

Guardrails

• No recursion. /ork:auto must not route to itself, directly or via a spawned agent.
• No bypass. Routing does not skip the target skill's guardrails, readonly enforcement, or confirmation steps.
• Classification quality is the whole job. A misroute that fails silently is worse than a fallback question. When two categories are equally plausible, ask — don't gamble.
• No silent upgrade. A Heavy (opus/fable) leg is premium spend and requires an explicit nod on its own line. Downgrades stay silent.
• The cap is not the router's to move. Never read, set, or suggest raising ORK_TEAM_OPUS_MAX. That is the user's budget.
• A gate denial is not a re-route trigger. If team-size-gate denies a spawn, do NOT relabel a Heavy leg as Standard to slip under the cap. Report the denial and let the user decide.
• Weight is a hint, not an override. The router cannot change an agent's declared model:. It selects skills and agents, and sets the caller's model that the 13 inherit agents adopt. It never overrides a target skill's own agent selection.

Validation

Routing accuracy is gateable, not vibes. routing-benchmark.json holds 50 labeled goal → category pairs (easy + genuinely ambiguous). Validate after any change to the category table or disambiguation rules:

# isolated classification check via the bare-eval harness
/ork:bare-eval   # grade router output against routing-benchmark.json

Target ≥95% category accuracy; track the fallback rate as a degradation alarm as the skill library grows.

References

• references/routing-rules.md — per-category parameter extraction, edge cases, disambiguation
• routing-benchmark.json — 50 labeled goal→category pairs for accuracy validation

Quality Bar

Done means all of these hold:

• Classification reasoning is stated out loud BEFORE a route is committed, naming the chosen intent category and the signal words that triggered it.
• Weight is classified in the same pass as intent, naming the signal that decided it, and resolves asymmetrically (ANY heavy signal ⇒ Heavy; Light needs ALL light signals; else Standard).
• The confirm block names one of the taxonomy's intent categories, its target skill or /goal loop, and the extracted args — on one line.
• Every Heavy (opus/fable) leg is surfaced for explicit approval before handoff; no premium spend is silent, and declining it runs Standard with the weakened check named.
• Adversarial, security, safety, architecture, and cross-cutting work is never routed below opus tier; mechanical single-file IO-bound work is not routed above haiku tier.
• The router neither counts spawns nor touches ORK_TEAM_OPUS_MAX, and never downgrades a Heavy leg to evade a team-size-gate denial.
• No target skill is invoked without an explicit nod (or -y); handoff never precedes confirmation.
• When two categories are equally plausible, exactly ONE clarifying question is asked — the fallback is never silently guessed.
• optimize routes to a /goal loop and improve-skill to the evolution gate; neither claims a dedicated skill that does not exist.
• The router does none of the target work itself and never routes to /ork:auto (no recursion).

Related skills

• /ork:help — static categorized directory (browse, don't route)
• /ork:prd-to-goal — decompose a spec into a /goal line (the optimize route's engine)
• /ork:fix-issue · /ork:cover · /ork:brainstorm · /ork:implement · /ork:review-pr · /ork:verify — the route targets
• /ork:assess — champion/challenger holdout gate (the improve-skill route)

References (1)

Routing Rules

Routing Rules

Per-category parameter extraction + edge cases. Read during the Classify step to configure the target skill correctly.

fix → /ork:fix-issue

• Extract: bug description (full goal), target files if named, ticket/issue #N, quoted error message.
• Invoke: /ork:fix-issue \{description or #N\}
• Edges: "fix the tests" is fix (repair broken tests), not cover (add new tests). "fix performance" is ambiguous → ask: debug a specific issue, or optimize a metric?

diagnose → /ork:fix-issue (investigation-first)

• A "why…" question is a gentler entry than a fix command. Frame the plan as observe → hypothesize → propose, then offer to apply the fix.
• A "why…" question is ALWAYS diagnose, even when it names a failure ("why isn't the build green", "why does the API return 500", "why can't users log in"). The question form is what makes it diagnose — without one, a statement of breakage or a repair imperative is fix ("there's a regression in checkout", "resolve the 500 errors on /api/users").
• Invoke: /ork:fix-issue \{question\} with an investigation framing. ("I'll investigate first, then propose a fix — ok?")

optimize → /goal loop (no dedicated skill)

• Extract: metric (latency/throughput/bundle/memory), direction (minimize for size/time/cost; maximize for score/rate), goal value + unit, target files.
• Invoke: compose a /goal loop via /ork:prd-to-goal → references/recipe-library.md. Be explicit that this is a /goal-driven loop, not a /ork:experiment skill (which doesn't exist).
• Edges: "make it faster" with no metric → ask what to measure (response time? build time? bundle?). Multiple metrics → pick the emphasized one, note the rest as constraints.

cover → /ork:cover

• Extract: target % ("90%" → 90, "above 85" → 85), scope ("the auth module" → src/auth/).
• Invoke: /ork:cover --target \{N\}
• Edges: "write more tests" with no target → ask the target %. "test the new feature" is build/verify (functional tests), not cover (coverage %). A surface called out as "untested" is cover even with no % target ("the payments service is untested, fix that") — the "fix" there repairs a coverage gap, not a bug; ask the target % at invoke time.

design → /ork:brainstorm

• Extract: topic (full goal). Deep mode if the goal says "thorough/comprehensive/deep dive" or spans multiple systems.
• Invoke: /ork:brainstorm \{topic\}
• Edges: "how should we…" is design, not build. "Design AND build…" → start design, offer build after.

build → /ork:implement

• Extract: feature description, ticket ID, mode (greenfield/brownfield/refactor/bugfix).
• Invoke: /ork:implement \{description\}
• Edges: "implement the design from the brainstorm" → check for recent brainstorm state first.

review → /ork:review-pr

• Extract: PR/MR number (#123 → 123), scope filter if named.
• Invoke: /ork:review-pr \{number or branch\}
• Edges: "review my code" with no PR → ask which PR/branch. "review the design" is design, not review.

verify → /ork:verify

• Extract: checks (tests/lint/typecheck/all), scope.
• Invoke: /ork:verify
• Edges: "make sure it works" → all checks. "check tests pass" → tests-focused.

improve-skill → holdout-promotion gate

• Extract: which SKILL.md, quality metric (else task-completion against test cases).
• Invoke: the champion/challenger holdout-promotion gate (/ork:assess evals + evolution-engine). Requires a benchmark + holdout set to exist first — if missing, help the user define 5–10 cases before looping.
• Edges: "optimize my prompt" (not a skill file) → route to optimize with the prompt as the target. When the improvement target IS a skill or agent — a SKILL.md, a named skill, an agent prompt — it is improve-skill regardless of the verb: "optimize the prompt for the security-auditor skill" and "make the brainstorm SKILL.md produce better ideas" are both improve-skill, not optimize/design.

Model weight (second, orthogonal dimension)

Intent answers who does the work. Weight answers how expensive that worker should be. Both are decided in the same Classify pass. Weight never changes the intent, and intent never fixes the weight: a route is \{intent\} @ \{weight\}.

The tiers are the ones already declared in src/agents/*.md frontmatter. There is no separate router taxonomy and no new vocabulary to learn (36 agents):

The rubric (one pass, decidable)

Read the goal once against both signal columns:

Resolution is asymmetric on purpose:

1. ANY heavy signal makes it Heavy (opus tier).
2. Light requires ALL light signals (haiku tier).
3. Everything else is Standard (sonnet tier). This is the default, and the correct answer for most work.

The asymmetry is not laziness. Under-powering a security review or a schema migration produces a confident wrong answer that nobody catches; over-powering a file rename only wastes money. Bias the rare direction toward the recoverable failure.

Weight is per leg, not per route

A single route often fans out into legs of different weight. Classify the legs, not the headline. /ork:review-pr on a PR touching auth is a Heavy security leg (security-auditor, opus) plus a Standard quality leg plus a Light lint leg. Do not promote the whole route to Heavy because one leg is.

Default weight per intent

Defaults, not verdicts. The rubric overrides any row here when the goal carries a heavy signal.

Premium spend needs explicit consent

Routing down is silent. Spending less is never a decision the user has to approve.

Routing up to Heavy is premium spend (opus / fable) and MUST be surfaced in the confirm line, naming the heavy signal that triggered it. Never upgrade silently, never mid-handoff, never inside a spawned agent that the user did not see. If the user declines the upgrade, run Standard and state plainly which check is being weakened.

Composition with the spend cap (they are different mechanisms)

routing        (ex ante,  per leg)     -> SELECTS the tier
team-size-gate (ex post,  per session) -> COUNTS what was selected, caps it

src/hooks/src/pretool/task/team-size-gate.ts is a backstop, not a selector. It reads toolInput.model exactly once, read-only, to classify premium vs non-premium, then increments a session ledger and warns or denies past ORK_TEAM_OPUS_MAX (default 8). It has no model-override surface and is ADVISORY by default, so its "Use sonnet/haiku" message is advice to a human that it cannot itself enact. Routing is what enacts it.

Rules that keep the two composed rather than duplicated:

• The router never reads, sets, or suggests raising ORK_TEAM_OPUS_MAX. The cap is the user's budget, not a router variable.
• A gate denial is not a re-route trigger. Do not relabel a Heavy leg as Standard to slip under the cap. Report the denial and let the user choose.
• The router does not re-implement counting. It has no session ledger and makes no claim about remaining budget.

Honest limits of this dimension

Tier is bound statically to agent identity in each agent's frontmatter, and the spawn happens downstream in whichever skill the router handed off to. The router therefore has exactly two levers, and neither is a model override:

1. Selection: which skill and which agent gets the work (choosing eval-runner over a sonnet agent is a real tier decision).
2. The caller's model: the 13 inherit agents (36% of the catalog) adopt it, so the weight the router hands off at propagates to that bloc.

Everything else is a documented hint carried into the handoff. Per the no-bypass guardrail, the router does not override a target skill's own agent selection.

Disambiguation (when multiple categories match)

1. Explicit verb wins. "Fix the slow query" → fix.
2. Metric + direction → optimize.
3. Percentage in a test context → cover.
4. Question form → design ("how should") or diagnose ("why") — and this beats symptom words: "why isn't the build green" is diagnose, not fix.
5. PR/MR/#N → review.
6. Ticket reference → build.
7. Truly ambiguous → ask ONE question naming the two candidate routes. Do not guess when two DIFFERENT skills are plausible and the goal names no artifact, metric, or number. Canonical fallbacks: "fix performance on the dashboard" (fix vs optimize), "review my code" with no PR/branch named (review vs verify), "test the new feature" (cover vs verify vs build), "the queries are slow and maybe vulnerable to injection" (optimize vs review).