| 1 | //! The **Adaptive Reasoning Router** — a saved, reusable *service*, not a Fleet |
| 2 | //! member. |
| 3 | //! |
| 4 | //! A Fleet answers *who*: the exact provider/model assignments an operator |
| 5 | //! saved for each worker. A Reasoning Router answers a much smaller question, |
| 6 | //! for an already frozen worker route: *how hard should this already-chosen |
| 7 | //! model think on this task?* Those are different kinds of thing, so they are |
| 8 | //! different kinds of value here: |
| 9 | //! |
| 10 | //! - A Router is **never dispatchable**. It has no role, no tools, no shell, no |
| 11 | //! write authority, and no delegation budget. It cannot be named by a task. |
| 12 | //! - A Router is **referenced, not embedded**. It is saved once at |
| 13 | //! `routers/<name>.toml` and referenced by name from any number of Fleets, so |
| 14 | //! two Fleets can share one Router configuration without duplicating it. |
| 15 | //! - A Router **never changes a route**. Provider, model, member, role, tools, |
| 16 | //! and permissions are all frozen before it is called and are not among the |
| 17 | //! things it is allowed to answer. |
| 18 | //! |
| 19 | //! ## Cheap by construction |
| 20 | //! |
| 21 | //! A Router call is a per-task tax on someone's tokens, so its own reasoning is |
| 22 | //! capped at [`RouterCallReasoning`] — `off` or `low`, nothing else. `medium`, |
| 23 | //! `high`, and `max` are **rejected at parse time rather than silently clamped**: |
| 24 | //! an operator who wrote `high` asked for something this service will not do, |
| 25 | //! and quietly running at `off` while the file says `high` is exactly the kind |
| 26 | //! of invisible substitution receipts exist to prevent. |
| 27 | //! |
| 28 | //! The reverse lie is equally forbidden. A profile that asks for `low` is |
| 29 | //! *called* at `low` wherever the route can express it; nothing here forces |
| 30 | //! `off` and then reports `low`. Normalization against the route's real |
| 31 | //! capability is recorded on the receipt (see |
| 32 | //! [`crate::fleet_reasoning::RouterCallDisclosure`]). |
| 33 | //! |
| 34 | //! ## Legacy inline routers |
| 35 | //! |
| 36 | //! The prototype form — a `[[members]]` entry with `kind = "router"` inside the |
| 37 | //! Fleet file — still parses, is labelled `legacy_inline`, and is **normalized |
| 38 | //! into the same [`CapturedReasoningRouter`]** the named store produces. There |
| 39 | //! is one runtime representation of the service, whichever way it was written. |
| 40 | |
| 41 | use std::path::PathBuf; |
| 42 | |
| 43 | use serde::{Deserialize, Serialize}; |
| 44 | use thiserror::Error; |
| 45 | |
| 46 | use crate::fleet_exact::{FrozenRoute, PermissionCeiling, ReasoningTier, RouterMember}; |
| 47 | use crate::named_fleet::FleetSearchRoot; |
| 48 | |
| 49 | /// Directory (under each search root) that holds saved Router profiles. |
| 50 | pub const REASONING_ROUTER_DIR: &str = "routers"; |
| 51 | /// Wire value of the `schema` key that selects a Router profile document. |
| 52 | pub const REASONING_ROUTER_SCHEMA_KIND: &str = "reasoning_router"; |
| 53 | /// Current revision of the Router profile schema. |
| 54 | pub const REASONING_ROUTER_SCHEMA_REVISION: u32 = 1; |
| 55 | /// Stable service label a receipt prints so the reader can tell at a glance |
| 56 | /// that this is the reasoning service and not a Fleet member. |
| 57 | pub const REASONING_ROUTER_SERVICE_KIND: &str = "reasoning_router"; |
| 58 | /// Origin recorded for a Router that was written inline in a Fleet file. |
| 59 | pub const LEGACY_INLINE_ROUTER_ORIGIN: &str = "legacy_inline"; |
| 60 | |
| 61 | /// The reasoning a **Router call itself** may run at. |
| 62 | /// |
| 63 | /// Deliberately not [`ReasoningTier`]: this type exists precisely so that |
| 64 | /// `medium`/`high`/`max` are unrepresentable. A Router emits one ~15-byte JSON |
| 65 | /// object; anything above `low` spends a user's tokens on thinking about a |
| 66 | /// question that does not need thought. |
| 67 | #[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)] |
| 68 | #[serde(rename_all = "snake_case")] |
| 69 | pub enum RouterCallReasoning { |
| 70 | #[default] |
| 71 | Off, |
| 72 | Low, |
| 73 | } |
| 74 | |
| 75 | impl RouterCallReasoning { |
| 76 | #[must_use] |
| 77 | pub const fn as_str(self) -> &'static str { |
| 78 | match self { |
| 79 | Self::Off => "off", |
| 80 | Self::Low => "low", |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | /// The concrete tier this maps onto for capability normalization. |
| 85 | #[must_use] |
| 86 | pub const fn tier(self) -> ReasoningTier { |
| 87 | match self { |
| 88 | Self::Off => ReasoningTier::Off, |
| 89 | Self::Low => ReasoningTier::Low, |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | /// Parse a configured value. |
| 94 | /// |
| 95 | /// `medium`/`high`/`max` are a distinct, named error rather than a clamp — |
| 96 | /// see the module docs. |
| 97 | pub fn parse(value: &str, router: &str) -> Result<Self, ReasoningRouterError> { |
| 98 | let trimmed = value.trim(); |
| 99 | match trimmed.to_ascii_lowercase().as_str() { |
| 100 | "off" | "none" | "disabled" => Ok(Self::Off), |
| 101 | "low" | "minimal" => Ok(Self::Low), |
| 102 | "medium" | "mid" | "high" | "max" | "maximum" | "xhigh" => { |
| 103 | Err(ReasoningRouterError::CallReasoningTooExpensive { |
| 104 | router: router.to_string(), |
| 105 | value: trimmed.to_string(), |
| 106 | }) |
| 107 | } |
| 108 | _ => Err(ReasoningRouterError::InvalidCallReasoning { |
| 109 | router: router.to_string(), |
| 110 | value: trimmed.to_string(), |
| 111 | }), |
| 112 | } |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | /// A Router identity qualified by the origin its definition came from. |
| 117 | /// |
| 118 | /// Path-free for the same reason [`crate::QualifiedFleetId`] is: an absolute |
| 119 | /// path in a durable receipt leaks the operator's home directory. |
| 120 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 121 | pub struct QualifiedRouterId { |
| 122 | pub name: String, |
| 123 | pub origin: String, |
| 124 | } |
| 125 | |
| 126 | impl QualifiedRouterId { |
| 127 | #[must_use] |
| 128 | pub fn qualified(&self) -> String { |
| 129 | format!("{}/{}", self.origin, self.name) |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | /// A saved Router profile: one exact provider/model plus a cheap call ceiling. |
| 134 | /// |
| 135 | /// This is *the* reusable unit. Any number of Fleets may reference the same |
| 136 | /// profile by name; none of them owns it. |
| 137 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 138 | pub struct ReasoningRouterProfile { |
| 139 | pub name: String, |
| 140 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 141 | pub description: Option<String>, |
| 142 | pub schema_revision: u32, |
| 143 | /// Exact configured provider id. |
| 144 | pub provider: String, |
| 145 | /// Exact model id. |
| 146 | pub model: String, |
| 147 | /// What the Router's own call runs at. `off` or `low` only. |
| 148 | pub call_reasoning: RouterCallReasoning, |
| 149 | } |
| 150 | |
| 151 | impl ReasoningRouterProfile { |
| 152 | /// Parse a Router profile document. |
| 153 | pub fn parse(text: &str) -> Result<Self, ReasoningRouterError> { |
| 154 | let doc: RouterProfileToml = |
| 155 | toml::from_str(text).map_err(|error| ReasoningRouterError::Parse(error.to_string()))?; |
| 156 | if !doc |
| 157 | .schema |
| 158 | .trim() |
| 159 | .eq_ignore_ascii_case(REASONING_ROUTER_SCHEMA_KIND) |
| 160 | { |
| 161 | return Err(ReasoningRouterError::UnknownSchema { |
| 162 | schema: doc.schema.trim().to_string(), |
| 163 | }); |
| 164 | } |
| 165 | if doc.schema_revision != REASONING_ROUTER_SCHEMA_REVISION { |
| 166 | return Err(ReasoningRouterError::UnsupportedRevision { |
| 167 | revision: doc.schema_revision, |
| 168 | supported: REASONING_ROUTER_SCHEMA_REVISION, |
| 169 | }); |
| 170 | } |
| 171 | let name = crate::role_resolve::normalize_token(&doc.name).ok_or_else(|| { |
| 172 | ReasoningRouterError::InvalidToken { |
| 173 | field: "name".to_string(), |
| 174 | value: doc.name.trim().to_string(), |
| 175 | } |
| 176 | })?; |
| 177 | let provider = exact_token(&doc.provider, &name, "provider")?; |
| 178 | let model = exact_token(&doc.model, &name, "model")?; |
| 179 | let call_reasoning = match doc.call_reasoning.as_deref() { |
| 180 | None => RouterCallReasoning::default(), |
| 181 | Some(value) => RouterCallReasoning::parse(value, &name)?, |
| 182 | }; |
| 183 | |
| 184 | Ok(Self { |
| 185 | name, |
| 186 | description: doc.description, |
| 187 | schema_revision: doc.schema_revision, |
| 188 | provider, |
| 189 | model, |
| 190 | call_reasoning, |
| 191 | }) |
| 192 | } |
| 193 | |
| 194 | /// Load one profile from a labelled search root set. |
| 195 | /// |
| 196 | /// A bare name present under more than one origin is **ambiguous**: a |
| 197 | /// personal `~/.codewhale` Router silently shadowing a project Router would |
| 198 | /// change which provider sees every task's routing summary. Naming the |
| 199 | /// origin (`codewhale_home/fast`) resolves it. |
| 200 | pub fn load_by_name( |
| 201 | name: &str, |
| 202 | search_roots: &[FleetSearchRoot], |
| 203 | ) -> Result<(Self, QualifiedRouterId), ReasoningRouterError> { |
| 204 | let (requested_origin, bare) = split_qualified(name); |
| 205 | if bare.is_empty() { |
| 206 | return Err(ReasoningRouterError::InvalidToken { |
| 207 | field: "router reference".to_string(), |
| 208 | value: name.trim().to_string(), |
| 209 | }); |
| 210 | } |
| 211 | let file_name = format!("{bare}.toml"); |
| 212 | |
| 213 | let mut candidates: Vec<(&FleetSearchRoot, PathBuf)> = Vec::new(); |
| 214 | for root in search_roots { |
| 215 | if let Some(origin) = requested_origin |
| 216 | && !root.origin.eq_ignore_ascii_case(origin) |
| 217 | { |
| 218 | continue; |
| 219 | } |
| 220 | let path = root.root.join(REASONING_ROUTER_DIR).join(&file_name); |
| 221 | if path.is_file() { |
| 222 | candidates.push((root, path)); |
| 223 | } |
| 224 | } |
| 225 | |
| 226 | let Some((first_root, first_path)) = candidates.first() else { |
| 227 | return Err(ReasoningRouterError::NotFound { |
| 228 | name: name.trim().to_string(), |
| 229 | }); |
| 230 | }; |
| 231 | if candidates.len() > 1 { |
| 232 | // Unlike legacy Fleet role maps, there is no first-hit-wins fallback |
| 233 | // here: every Router profile names an exact provider/model, so a |
| 234 | // shadowed one always changes behavior. |
| 235 | return Err(ReasoningRouterError::AmbiguousRouter { |
| 236 | name: bare.to_string(), |
| 237 | origins: candidates |
| 238 | .iter() |
| 239 | .map(|(root, _)| format!("{}/{bare}", root.origin)) |
| 240 | .collect(), |
| 241 | }); |
| 242 | } |
| 243 | |
| 244 | let text = |
| 245 | std::fs::read_to_string(first_path).map_err(|error| ReasoningRouterError::Io { |
| 246 | path: first_path.display().to_string(), |
| 247 | message: error.to_string(), |
| 248 | })?; |
| 249 | let profile = Self::parse(&text)?; |
| 250 | if profile.name != bare { |
| 251 | return Err(ReasoningRouterError::NameMismatch { |
| 252 | declared: profile.name.clone(), |
| 253 | expected: bare.to_string(), |
| 254 | }); |
| 255 | } |
| 256 | let id = QualifiedRouterId { |
| 257 | name: profile.name.clone(), |
| 258 | origin: first_root.origin.clone(), |
| 259 | }; |
| 260 | Ok((profile, id)) |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | /// The Router service as frozen into a Workflow snapshot. |
| 265 | /// |
| 266 | /// Whether it came from a named profile or from the legacy inline member, this |
| 267 | /// is the single runtime representation. Nothing downstream branches on which |
| 268 | /// form the operator wrote. |
| 269 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 270 | pub struct CapturedReasoningRouter { |
| 271 | /// Always [`REASONING_ROUTER_SERVICE_KIND`]. Recorded explicitly so a |
| 272 | /// receipt states what kind of thing this is instead of implying it. |
| 273 | #[serde(default = "default_service_kind")] |
| 274 | pub service_kind: String, |
| 275 | /// The Router's id: the profile name, or the inline member's id. |
| 276 | pub id: String, |
| 277 | /// The origin the definition came from, or [`LEGACY_INLINE_ROUTER_ORIGIN`]. |
| 278 | #[serde(default = "default_router_origin")] |
| 279 | pub origin: String, |
| 280 | /// True when this was written inline in the Fleet file rather than saved as |
| 281 | /// a reusable profile. |
| 282 | #[serde(default)] |
| 283 | pub legacy_inline: bool, |
| 284 | /// The Router's own exact provider/model. |
| 285 | pub route: FrozenRoute, |
| 286 | /// What the operator configured this Router's call to run at. |
| 287 | #[serde(default)] |
| 288 | pub requested_call_reasoning: RouterCallReasoning, |
| 289 | /// Always `false`. Stated rather than implied. |
| 290 | #[serde(default)] |
| 291 | pub dispatchable: bool, |
| 292 | /// Always [`PermissionCeiling::ROUTER`]. |
| 293 | #[serde(default = "router_permissions")] |
| 294 | pub permissions: PermissionCeiling, |
| 295 | } |
| 296 | |
| 297 | fn default_service_kind() -> String { |
| 298 | REASONING_ROUTER_SERVICE_KIND.to_string() |
| 299 | } |
| 300 | |
| 301 | fn default_router_origin() -> String { |
| 302 | LEGACY_INLINE_ROUTER_ORIGIN.to_string() |
| 303 | } |
| 304 | |
| 305 | fn router_permissions() -> PermissionCeiling { |
| 306 | PermissionCeiling::ROUTER |
| 307 | } |
| 308 | |
| 309 | impl CapturedReasoningRouter { |
| 310 | /// Capture a saved, reusable profile. |
| 311 | #[must_use] |
| 312 | pub fn from_profile(profile: &ReasoningRouterProfile, origin: impl Into<String>) -> Self { |
| 313 | Self { |
| 314 | service_kind: default_service_kind(), |
| 315 | id: profile.name.clone(), |
| 316 | origin: origin.into(), |
| 317 | legacy_inline: false, |
| 318 | route: FrozenRoute { |
| 319 | provider: profile.provider.clone(), |
| 320 | model: profile.model.clone(), |
| 321 | }, |
| 322 | requested_call_reasoning: profile.call_reasoning, |
| 323 | dispatchable: false, |
| 324 | permissions: PermissionCeiling::ROUTER, |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | /// Normalize the prototype inline form into the same captured service. |
| 329 | /// |
| 330 | /// The inline member's own `reasoning` was a full [`ReasoningTier`]; it is |
| 331 | /// mapped onto the cheap call ceiling here, and anything above `low` is |
| 332 | /// rejected by the Fleet parser rather than clamped silently. |
| 333 | #[must_use] |
| 334 | pub fn from_legacy_inline(member: &RouterMember) -> Self { |
| 335 | Self { |
| 336 | service_kind: default_service_kind(), |
| 337 | id: member.id.clone(), |
| 338 | origin: default_router_origin(), |
| 339 | legacy_inline: true, |
| 340 | route: member.frozen_route(), |
| 341 | requested_call_reasoning: member.call_reasoning, |
| 342 | dispatchable: false, |
| 343 | permissions: PermissionCeiling::ROUTER, |
| 344 | } |
| 345 | } |
| 346 | |
| 347 | /// `origin/id` — the stable display form a receipt prints. |
| 348 | #[must_use] |
| 349 | pub fn qualified(&self) -> String { |
| 350 | format!("{}/{}", self.origin, self.id) |
| 351 | } |
| 352 | |
| 353 | /// A Router is never a worker. Constant, not a policy lookup. |
| 354 | #[must_use] |
| 355 | pub const fn is_dispatchable(&self) -> bool { |
| 356 | false |
| 357 | } |
| 358 | |
| 359 | /// The Router's tool surface is empty, always. |
| 360 | #[must_use] |
| 361 | pub const fn tool_surface(&self) -> &'static [&'static str] { |
| 362 | &[] |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | /// How a Fleet points at its Router. |
| 367 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 368 | #[serde(rename_all = "snake_case", tag = "kind")] |
| 369 | pub enum FleetRouterRef { |
| 370 | /// A saved, reusable profile named by `reasoning_router = "<name>"`. |
| 371 | Profile { name: String }, |
| 372 | /// The prototype inline `[[members]] kind = "router"` form. |
| 373 | LegacyInline(Box<RouterMember>), |
| 374 | } |
| 375 | |
| 376 | fn split_qualified(name: &str) -> (Option<&str>, &str) { |
| 377 | let trimmed = name.trim(); |
| 378 | match trimmed.split_once('/') { |
| 379 | Some((origin, bare)) if !origin.trim().is_empty() && !bare.trim().is_empty() => { |
| 380 | (Some(origin.trim()), bare.trim()) |
| 381 | } |
| 382 | _ => (None, trimmed), |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | fn exact_token(value: &str, router: &str, field: &str) -> Result<String, ReasoningRouterError> { |
| 387 | let trimmed = value.trim(); |
| 388 | if trimmed.is_empty() |
| 389 | || trimmed |
| 390 | .chars() |
| 391 | .any(|ch| ch.is_whitespace() || matches!(ch, '"' | '\'' | '`' | '=')) |
| 392 | { |
| 393 | return Err(ReasoningRouterError::InvalidToken { |
| 394 | field: format!("{router}.{field}"), |
| 395 | value: trimmed.to_string(), |
| 396 | }); |
| 397 | } |
| 398 | Ok(trimmed.to_string()) |
| 399 | } |
| 400 | |
| 401 | #[derive(Debug, Deserialize)] |
| 402 | #[serde(deny_unknown_fields)] |
| 403 | struct RouterProfileToml { |
| 404 | name: String, |
| 405 | #[serde(default)] |
| 406 | description: Option<String>, |
| 407 | schema: String, |
| 408 | #[serde(default = "default_router_revision")] |
| 409 | schema_revision: u32, |
| 410 | provider: String, |
| 411 | model: String, |
| 412 | #[serde(default, alias = "reasoning")] |
| 413 | call_reasoning: Option<String>, |
| 414 | } |
| 415 | |
| 416 | const fn default_router_revision() -> u32 { |
| 417 | REASONING_ROUTER_SCHEMA_REVISION |
| 418 | } |
| 419 | |
| 420 | #[derive(Debug, Clone, PartialEq, Eq, Error)] |
| 421 | pub enum ReasoningRouterError { |
| 422 | #[error("failed to parse reasoning router profile: {0}")] |
| 423 | Parse(String), |
| 424 | #[error("failed to read reasoning router profile `{path}`: {message}")] |
| 425 | Io { path: String, message: String }, |
| 426 | #[error("reasoning router `{name}` was not found in any configured origin")] |
| 427 | NotFound { name: String }, |
| 428 | #[error( |
| 429 | "reasoning router `{name}` is defined in more than one place ({}); a router names an \ |
| 430 | exact provider/model, so shadowing would silently change which provider sees every \ |
| 431 | routing summary. Name one explicitly as `origin/{name}`.", |
| 432 | origins.join(", ") |
| 433 | )] |
| 434 | AmbiguousRouter { name: String, origins: Vec<String> }, |
| 435 | #[error("unknown reasoning router schema `{schema}`; expected `reasoning_router`")] |
| 436 | UnknownSchema { schema: String }, |
| 437 | #[error( |
| 438 | "reasoning router schema revision {revision} is not supported (this build reads {supported})" |
| 439 | )] |
| 440 | UnsupportedRevision { revision: u32, supported: u32 }, |
| 441 | #[error("{field} must be a non-empty token without whitespace, quotes, or `=` (got `{value}`)")] |
| 442 | InvalidToken { field: String, value: String }, |
| 443 | #[error("reasoning router name mismatch: file declares `{declared}`, expected `{expected}`")] |
| 444 | NameMismatch { declared: String, expected: String }, |
| 445 | #[error( |
| 446 | "reasoning router `{router}` requests call reasoning `{value}`; a router may only run at \ |
| 447 | `off` or `low`. This is rejected rather than clamped: a router answers one tiny JSON \ |
| 448 | object per task, and running it at `{value}` would spend your tokens on thinking nobody \ |
| 449 | asked for. Set `call_reasoning` to `off` or `low`." |
| 450 | )] |
| 451 | CallReasoningTooExpensive { router: String, value: String }, |
| 452 | #[error( |
| 453 | "reasoning router `{router}` has invalid call reasoning `{value}`; expected off or low" |
| 454 | )] |
| 455 | InvalidCallReasoning { router: String, value: String }, |
| 456 | } |
| 457 | |
| 458 | #[cfg(test)] |
| 459 | mod tests { |
| 460 | use super::*; |
| 461 | |
| 462 | const LUNA: &str = r#" |
| 463 | name = "luna-low" |
| 464 | description = "GPT-5.6 Luna, called at low" |
| 465 | schema = "reasoning_router" |
| 466 | schema_revision = 1 |
| 467 | provider = "openai" |
| 468 | model = "gpt-5.6-luna" |
| 469 | call_reasoning = "low" |
| 470 | "#; |
| 471 | |
| 472 | #[test] |
| 473 | fn a_profile_parses_its_exact_route_and_cheap_call_tier() { |
| 474 | let profile = ReasoningRouterProfile::parse(LUNA).expect("parse"); |
| 475 | assert_eq!(profile.name, "luna-low"); |
| 476 | assert_eq!(profile.provider, "openai"); |
| 477 | assert_eq!(profile.model, "gpt-5.6-luna"); |
| 478 | assert_eq!(profile.call_reasoning, RouterCallReasoning::Low); |
| 479 | assert_eq!(profile.schema_revision, REASONING_ROUTER_SCHEMA_REVISION); |
| 480 | } |
| 481 | |
| 482 | #[test] |
| 483 | fn call_reasoning_defaults_to_off_when_unset() { |
| 484 | let text = LUNA.replace("call_reasoning = \"low\"\n", ""); |
| 485 | let profile = ReasoningRouterProfile::parse(&text).expect("parse"); |
| 486 | assert_eq!(profile.call_reasoning, RouterCallReasoning::Off); |
| 487 | } |
| 488 | |
| 489 | /// The whole point of the cheap ceiling: an expensive tier is an error the |
| 490 | /// operator can see, never a clamp they cannot. |
| 491 | #[test] |
| 492 | fn medium_high_and_max_are_rejected_not_clamped() { |
| 493 | for value in ["medium", "high", "max", "xhigh"] { |
| 494 | let text = LUNA.replace("\"low\"", &format!("\"{value}\"")); |
| 495 | let err = ReasoningRouterProfile::parse(&text) |
| 496 | .expect_err("an expensive router tier must be rejected"); |
| 497 | assert!( |
| 498 | matches!(err, ReasoningRouterError::CallReasoningTooExpensive { .. }), |
| 499 | "value={value} err={err:?}" |
| 500 | ); |
| 501 | let message = err.to_string(); |
| 502 | assert!(message.contains("off"), "{message}"); |
| 503 | assert!(message.contains("low"), "{message}"); |
| 504 | assert!( |
| 505 | !message.contains("clamped to"), |
| 506 | "the error must not describe a clamp: {message}" |
| 507 | ); |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | #[test] |
| 512 | fn an_unknown_tier_is_its_own_error() { |
| 513 | let text = LUNA.replace("\"low\"", "\"turbo\""); |
| 514 | assert!(matches!( |
| 515 | ReasoningRouterProfile::parse(&text).expect_err("garbage"), |
| 516 | ReasoningRouterError::InvalidCallReasoning { .. } |
| 517 | )); |
| 518 | } |
| 519 | |
| 520 | #[test] |
| 521 | fn an_unknown_schema_or_future_revision_fails_closed() { |
| 522 | let wrong_schema = LUNA.replace("\"reasoning_router\"", "\"exact\""); |
| 523 | assert!(matches!( |
| 524 | ReasoningRouterProfile::parse(&wrong_schema).expect_err("schema"), |
| 525 | ReasoningRouterError::UnknownSchema { .. } |
| 526 | )); |
| 527 | |
| 528 | let future = LUNA.replace("schema_revision = 1", "schema_revision = 99"); |
| 529 | assert!(matches!( |
| 530 | ReasoningRouterProfile::parse(&future).expect_err("revision"), |
| 531 | ReasoningRouterError::UnsupportedRevision { revision: 99, .. } |
| 532 | )); |
| 533 | } |
| 534 | |
| 535 | #[test] |
| 536 | fn a_captured_profile_holds_no_authority_and_is_never_dispatchable() { |
| 537 | let profile = ReasoningRouterProfile::parse(LUNA).expect("parse"); |
| 538 | let captured = CapturedReasoningRouter::from_profile(&profile, "workspace"); |
| 539 | |
| 540 | assert_eq!(captured.qualified(), "workspace/luna-low"); |
| 541 | assert_eq!(captured.service_kind, REASONING_ROUTER_SERVICE_KIND); |
| 542 | assert!(!captured.legacy_inline); |
| 543 | assert!(!captured.is_dispatchable()); |
| 544 | assert!(captured.tool_surface().is_empty()); |
| 545 | assert!(!captured.permissions.tools); |
| 546 | assert!(!captured.permissions.write); |
| 547 | assert!(!captured.permissions.network_tool); |
| 548 | assert_eq!(captured.permissions.delegation_depth, 0); |
| 549 | } |
| 550 | |
| 551 | /// One saved profile, two Fleets. The service is referenced, not owned. |
| 552 | #[test] |
| 553 | fn one_saved_profile_serves_more_than_one_fleet() { |
| 554 | let tmp = tempfile::tempdir().expect("tmp"); |
| 555 | std::fs::create_dir_all(tmp.path().join(REASONING_ROUTER_DIR)).expect("dir"); |
| 556 | std::fs::write( |
| 557 | tmp.path().join(REASONING_ROUTER_DIR).join("luna-low.toml"), |
| 558 | LUNA, |
| 559 | ) |
| 560 | .expect("write"); |
| 561 | |
| 562 | let roots = vec![FleetSearchRoot::new("workspace", tmp.path())]; |
| 563 | let (first, first_id) = |
| 564 | ReasoningRouterProfile::load_by_name("luna-low", &roots).expect("load"); |
| 565 | let (second, second_id) = |
| 566 | ReasoningRouterProfile::load_by_name("workspace/luna-low", &roots).expect("qualified"); |
| 567 | |
| 568 | assert_eq!(first, second); |
| 569 | assert_eq!(first_id, second_id); |
| 570 | assert_eq!(first_id.qualified(), "workspace/luna-low"); |
| 571 | |
| 572 | // Two independent captures of the same saved service agree exactly. |
| 573 | let a = CapturedReasoningRouter::from_profile(&first, &first_id.origin); |
| 574 | let b = CapturedReasoningRouter::from_profile(&second, &second_id.origin); |
| 575 | assert_eq!(a, b); |
| 576 | } |
| 577 | |
| 578 | /// Bare-name ambiguity across origins must fail; a qualified origin works. |
| 579 | #[test] |
| 580 | fn a_bare_name_defined_in_two_origins_is_ambiguous_until_qualified() { |
| 581 | let tmp = tempfile::tempdir().expect("tmp"); |
| 582 | let home = tmp.path().join("home"); |
| 583 | let workspace = tmp.path().join("workspace"); |
| 584 | for root in [&home, &workspace] { |
| 585 | std::fs::create_dir_all(root.join(REASONING_ROUTER_DIR)).expect("dir"); |
| 586 | } |
| 587 | std::fs::write( |
| 588 | home.join(REASONING_ROUTER_DIR).join("luna-low.toml"), |
| 589 | LUNA.replace("gpt-5.6-luna", "gpt-5.6-luna-mini"), |
| 590 | ) |
| 591 | .expect("home"); |
| 592 | std::fs::write( |
| 593 | workspace.join(REASONING_ROUTER_DIR).join("luna-low.toml"), |
| 594 | LUNA, |
| 595 | ) |
| 596 | .expect("workspace"); |
| 597 | |
| 598 | let roots = vec![ |
| 599 | FleetSearchRoot::new("codewhale_home", &home), |
| 600 | FleetSearchRoot::new("workspace", &workspace), |
| 601 | ]; |
| 602 | |
| 603 | let err = ReasoningRouterProfile::load_by_name("luna-low", &roots) |
| 604 | .expect_err("bare name must not be resolved by shadowing"); |
| 605 | assert!( |
| 606 | matches!(err, ReasoningRouterError::AmbiguousRouter { .. }), |
| 607 | "{err:?}" |
| 608 | ); |
| 609 | let message = err.to_string(); |
| 610 | assert!(message.contains("codewhale_home"), "{message}"); |
| 611 | assert!(message.contains("workspace"), "{message}"); |
| 612 | |
| 613 | let (workspace_profile, id) = |
| 614 | ReasoningRouterProfile::load_by_name("workspace/luna-low", &roots).expect("qualified"); |
| 615 | assert_eq!(id.qualified(), "workspace/luna-low"); |
| 616 | assert_eq!(workspace_profile.model, "gpt-5.6-luna"); |
| 617 | |
| 618 | let (home_profile, home_id) = |
| 619 | ReasoningRouterProfile::load_by_name("codewhale_home/luna-low", &roots) |
| 620 | .expect("qualified"); |
| 621 | assert_eq!(home_id.qualified(), "codewhale_home/luna-low"); |
| 622 | assert_eq!(home_profile.model, "gpt-5.6-luna-mini"); |
| 623 | } |
| 624 | |
| 625 | #[test] |
| 626 | fn a_missing_profile_is_a_named_error() { |
| 627 | let tmp = tempfile::tempdir().expect("tmp"); |
| 628 | let roots = vec![FleetSearchRoot::new("workspace", tmp.path())]; |
| 629 | assert!(matches!( |
| 630 | ReasoningRouterProfile::load_by_name("nope", &roots).expect_err("missing"), |
| 631 | ReasoningRouterError::NotFound { .. } |
| 632 | )); |
| 633 | } |
| 634 | |
| 635 | /// A captured router serializes into a durable snapshot with no secrets and |
| 636 | /// no paths, and older records without the newer fields still read. |
| 637 | #[test] |
| 638 | fn a_captured_router_is_durable_and_backward_compatible() { |
| 639 | let profile = ReasoningRouterProfile::parse(LUNA).expect("parse"); |
| 640 | let captured = CapturedReasoningRouter::from_profile(&profile, "workspace"); |
| 641 | let json = serde_json::to_string(&captured).expect("serialize"); |
| 642 | let lowered = json.to_ascii_lowercase(); |
| 643 | for forbidden in ["api_key", "secret", "token", "bearer", "/users/", ".toml"] { |
| 644 | assert!(!lowered.contains(forbidden), "{forbidden} in {json}"); |
| 645 | } |
| 646 | let back: CapturedReasoningRouter = serde_json::from_str(&json).expect("round-trip"); |
| 647 | assert_eq!(back, captured); |
| 648 | |
| 649 | let older = r#"{"id":"router","route":{"provider":"zai","model":"glm-5-turbo"}}"#; |
| 650 | let legacy: CapturedReasoningRouter = serde_json::from_str(older).expect("serde defaults"); |
| 651 | assert_eq!(legacy.service_kind, REASONING_ROUTER_SERVICE_KIND); |
| 652 | assert_eq!(legacy.origin, LEGACY_INLINE_ROUTER_ORIGIN); |
| 653 | assert_eq!(legacy.requested_call_reasoning, RouterCallReasoning::Off); |
| 654 | assert!(!legacy.dispatchable); |
| 655 | } |
| 656 | } |
| 657 |