@diousk/pi-subagents-fast 0.23.0 → 0.25.1

package/CHANGELOG.md

@@ -7,6 +7,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.25.1] - 2026-10-05
+
+### Fixed
+- **GitHub Copilot agents omit `service_tier`.** Copilot rejects the field even on its Responses API. Configured tiers are ignored for Copilot requests and hidden from agent UI, including after model routing.
+
+## [0.25.0] - 2026-10-02
+
+### Added
+- **Optional Jev model instructions.** Configure `instruction` per model in JSON or `/agents`; successful selection appends it to the child system prompt. Shadow and fallback preserve the original instructions.
+- **Jev selects custom agent profiles.** In `jev` and `shadow` modes, enabled agent files and `jev.models` compete in one comparison; `jev` waits for the decision before starting the selected agent with its prompt, tools and model settings. Agent-only routing needs no duplicated model list, and runtime status shows selected and suggested agents. `auto` priority, low-confidence fallback and nested agent permissions are preserved.
+
+## [0.24.0] - 2026-10-02
+
+### Added
+- **Routing runtime status in `/agents`.** Open Model routing → Runtime status to inspect effective configuration, credential presence, eligible profiles and the latest retained routing decisions, including fallback reasons and actual model/thinking. The refreshable, read-only view hides API keys and makes no classifier request.
+
 ## [0.23.0] - 2026-10-02
 
 ### Changed

package/README.md

@@ -28,7 +28,7 @@ https://github.com/user-attachments/assets/8685261b-9338-4fea-8dfe-1c590d5df543
 - **Graceful turn limits** — agents get a "wrap up" warning before hard abort, producing clean partial results instead of cut-off output
 - **Case-insensitive agent types** — `"explore"`, `"Explore"`, `"EXPLORE"` all work. A type that doesn't resolve to exactly one *enabled* agent — unknown, disabled, or ambiguous between two agents differing only by case — falls back to general-purpose with a note, or is refused outright under [`fallbackSubagent: none`](#persistent-settings)
 - **Fuzzy model selection** — specify models by name (`"haiku"`, `"sonnet"`) instead of full IDs, with automatic filtering to only available/configured models
-- **Model routing** — custom agents first, then a user-supplied Markdown guideline, then optional Jev model selection. Configure it through `/agents → Model routing` or two ordinary JSON fields; see [Model routing](#model-routing)
+- **Model routing** — custom agents first, then a user-supplied Markdown guideline, then optional Jev model selection. In `jev` mode, Jev compares custom agents and model profiles together before starting the child. Configure it through `/agents → Model routing`; see [Model routing](#model-routing)
 - **Context inheritance** — optionally fork the parent conversation into a sub-agent so it knows what's been discussed
 - **Persistent agent memory** — three scopes (project, local, user) with automatic read-only fallback for agents without write tools
 - **Git worktree isolation** — run agents in isolated repo copies; changes auto-committed to branches on completion
@@ -340,7 +340,7 @@ All fields are optional — sensible defaults for everything.
 | `disallowed_tools` | — | Comma-separated tools to deny even if extensions provide them |
 | `isolation` | — | Set to `worktree` to run in an isolated git worktree, or `off` to refuse one even when the caller passes `isolation: "worktree"` (frontmatter is authoritative). `none`, `no`, and `false` are accepted spellings of `off` |
 | `model` | inherit parent | Model — `provider/modelId` or fuzzy name (`"haiku"`, `"sonnet"`). Resolved tolerantly (`.`/`-` and a trailing date stamp are interchangeable) and falls back to the same model under another provider if the named one doesn't have it |
-| `service_tier` | — | OpenAI Responses/Codex processing tier: `auto`, `default`, `flex`, `fast`, `priority`, or `scale`. Applied only to `openai-responses` and `openai-codex-responses`; omitted preserves the provider default, and other APIs ignore it without displaying it as active |
+| `service_tier` | — | OpenAI Responses/Codex processing tier: `auto`, `default`, `flex`, `fast`, `priority`, or `scale`. Applied only to `openai-responses` and `openai-codex-responses`, excluding GitHub Copilot; omitted preserves the provider default. GitHub Copilot and other APIs ignore it without displaying it as active |
 | `thinking` | inherit | off, minimal, low, medium, high, xhigh, max — actual availability depends on your pi version and model; pi maps or clamps unsupported levels |
 | `max_turns` | unlimited | Max agentic turns before graceful shutdown. `0` or omit for unlimited |
 | `persist_session` | `subagents.json` `rememberAgents` (default `true`) | Persist this subagent as a normal pi session instead of keeping the session in memory only; overrides the `rememberAgents` project default in both directions. It records its spawning session as parent, so it nests under it in `/resume`. The subagent's `.output` transcript is still written either way unless `output_transcript: false` |
@@ -353,7 +353,7 @@ All fields are optional — sensible defaults for everything.
 | `isolated` | `false` | Hermetic specialist mode: forces `extensions: false` + `skills: false` + drops `ext:` selectors. Only built-in tools. Distinct from `isolation: worktree` (filesystem) |
 | `enabled` | `true` | Set to `false` to disable an agent (useful for hiding a default agent per-project) |
 
-For an OpenAI Responses or Codex agent, set `service_tier: fast` to request fast processing; `priority` remains a supported alias. Availability depends on the provider and account. The UI shows the requested tier only when the effective model uses one of those APIs. See [OpenAI fast mode](https://developers.openai.com/api/docs/guides/fast-mode).
+For an OpenAI Responses or Codex agent, set `service_tier: fast` to request fast processing; `priority` remains a supported alias. Availability depends on the provider and account. GitHub Copilot rejects `service_tier` even when its model uses `openai-responses`, so the extension skips the field and tier tag for Copilot. The UI shows the requested tier only when the effective model supports it. See [OpenAI fast mode](https://developers.openai.com/api/docs/guides/fast-mode).
 
 The extension forwards both tiers for `gpt-6.1-sol` and `gpt-6-luna` through Pi's OpenAI Responses and Codex transports. Pi 1.0.0's Codex adapter currently estimates a response marked `fast` at the standard rate; forwarding the tier works, but its displayed cost can be understated. This extension reports Pi's cost estimate without recalculating it. OpenAI documents Fast mode as unavailable for these models with EU data residency.
 
@@ -636,6 +636,8 @@ When background agents complete, they notify the main agent. The **join mode** c
 
 Open `/agents → Model routing` to choose a mode and configure a guideline path, models with descriptions and required thinking levels, or a masked TypeSafe API key. The menu shows the active mode and source and saves project settings. For machine-wide defaults, edit `~/.pi/agent/subagents.json`; project overrides go in `.pi/subagents.json`. `PI_CODING_AGENT_DIR` changes the global directory along with Pi's other configuration.
 
+Use `/agents → Model routing → Runtime status` to inspect the effective mode and its project/global source, routing priority and fallback, guideline path, configured model/thinking profiles, current candidate eligibility, classifier availability and credential presence. No API key is displayed and opening the page does not classify a task. Credential presence does not confirm that the service will accept it. The page also shows the latest ten retained agent decisions, including selected versus actual model/thinking, shadow suggestions, confidence, fallback/error reasons and classifier usage. This is an in-memory view, not a durable log: evicted records and previous sessions are not included. Press `r` to refresh, Up/Down to scroll and Esc to return.
+
 Set one top-level field to control routing; omitting it means `auto`:
 
 ```json
@@ -645,12 +647,14 @@ Set one top-level field to control routing; omitting it means `auto`:
 | `routingMode` | Behavior |
 |---|---|
 | `auto` (default) | Use the priority table below |
-| `shadow` | Ask Jev and record its suggestion, but keep the model chosen from agent definitions, the guideline or the existing model. Jev requests can incur charges |
-| `jev` | Jev chooses first for every fresh task, even with custom agents, a guideline, explicit model parameters or agent-file model pins. Invalid configuration, unavailable credentials, low confidence or errors keep the default-priority choice |
+| `shadow` | Compare custom agents and model profiles through Jev and record its suggestion, but keep the original agent, model and thinking. Jev requests can incur charges |
+| `jev` | Wait for Jev to compare all eligible custom agents and `jev.models` together before starting each fresh task. Invalid configuration, unavailable credentials, low confidence or errors keep the default-priority choice |
 | `off` | No custom-agent routing selection guidance, guideline injection or Jev requests. Agents remain callable and existing model/thinking settings still apply |
 
 Mode changes apply to subsequent fresh launches. The main agent's routing guidance and tool description refresh before its next turn, so switching to `off` removes previously injected routing instructions.
 
+In `jev` mode, a background `Agent` call also waits for routing before returning. If it queues behind the concurrency limit, it waits through dequeue and selection. Runtime status shows `pending` while Jev is deciding; no child session or worktree starts during that wait. After selection, normal foreground/background execution applies.
+
 The default priority is:
 
 | Priority | Configuration | Who chooses |
@@ -662,6 +666,22 @@ The default priority is:
 
 **Custom agents:** keep using `~/.pi/agent/agents/<name>.md`, `.pi/agents/<name>.md` or `.agents/agents/<name>.md`. No routing setting is needed. Built-in and disabled agents do not activate priority 1. Agent-file model/thinking pins supply the default choice over `Agent` parameters; a confident Jev choice in `jev` mode can replace both model and thinking level.
 
+**Let Jev choose the agent:** keep your existing agent files and set `routingMode: "jev"`. The main agent submits the task and waits for the extension's routing decision; it does not choose the specialist itself. The required `subagent_type` supplies a fallback (`general-purpose` when available, an allowed type for nested calls). Jev compares each enabled custom agent's `description` with every `jev.models` description in one request. Agents using the same model remain separate candidates. Project definitions override global definitions with the same name; built-in defaults, disabled agents, unavailable models and agents outside the parent's nested allowlist are excluded.
+
+For agent-only routing, this is enough when Pi/environment TypeSafe credentials are available:
+
+```json
+{ "routingMode": "jev" }
+```
+
+Or supply a key without duplicating your agent profiles in JSON:
+
+```json
+{ "routingMode": "jev", "jev": { "TYPESAFE_API_KEY": "your-typesafe-key" } }
+```
+
+A selected agent runs with its own prompt, tools, extensions, skills, memory, session persistence, service tier, model and thinking. An omitted agent model inherits the parent model; an unavailable model pin excludes that candidate. Its configured turn limit, context and isolation settings also apply; unspecified values retain the invocation's defaults. Foreground/background delivery, handles, workflow schema and ownership/depth remain properties of the original invocation. Tool/nested transcripts use the selected agent's `output_transcript`. A selected `jev.models` entry applies model/thinking and its optional `instruction`, retaining the submitted agent. Shadow applies neither kind of selection. Runtime status lists the combined candidates and the selected or suggested agent name.
+
 **Custom guideline:** write your routing rules in `~/.pi/agent/agents/custom-route.md`, then configure:
 
 ```json
@@ -686,7 +706,7 @@ For example, the Markdown can say “Use anthropic/claude-haiku-4-5 for simple e
 }
 ```
 
-`TYPESAFE_API_KEY` is optional when Pi already has TypeSafe credentials or the environment variable is set. A literal key must be a nonempty token without whitespace or control characters and applies only to that classifier request; it is saved in the settings file, masked in the menu and omitted from settings events, prompts and routing records. Without a literal key, Pi's native credential availability check runs before classification. Rejected credentials fall back without retrying. The extension does not change environment variables or register a routing provider. `models` accepts 1–254 unique entries with descriptions of 1–4000 characters. Every entry requires `thinkingLevel`: `minimal`, `low`, `medium`, `high`, `xhigh` or `max`. Existing profiles must add this field; a missing or invalid value disables the whole Jev block and uses the default-priority fallback. Jev entries use exact `provider/model-id` spelling; fuzzy names remain available for explicit `Agent` parameters.
+`TYPESAFE_API_KEY` is optional when Pi already has TypeSafe credentials or the environment variable is set. A literal key must be a nonempty token without whitespace or control characters and applies only to that classifier request; it is saved in the settings file, masked in the menu and omitted from settings events, prompts and routing records. Without a literal key, Pi's native credential availability check runs before classification. Rejected credentials fall back without retrying. The extension does not change environment variables or register a routing provider. `models` is optional (defaults to `[]`) and accepts up to 254 unique entries with descriptions of 1–4000 characters. Every model entry requires `thinkingLevel`: `minimal`, `low`, `medium`, `high`, `xhigh` or `max`. Each model may also set `instruction` (up to 16000 characters; blank means omitted). It is appended to the selected child’s system prompt, preserving existing agent instructions. Only `description` is sent as selection criteria; `instruction` is applied only after a successful model-profile selection, never in shadow or fallback. Edit it through `/agents → Model routing → Jev models and descriptions`. Agent files keep their optional `thinking` field. A missing or invalid model-entry level disables the whole Jev block. The combined eligible agent/model pool must not exceed 254 profiles; an oversized pool falls back with a diagnostic instead of silently dropping profiles. Jev entries use exact `provider/model-id` spelling; fuzzy names remain available in agent files and explicit `Agent` parameters.
 
 Under `auto`, supplying **either** `model` or `thinking` explicitly skips Jev. An agent-file pin also skips it. In workflows, either `model` or `effort` skips it, and workflow options retain their precedence over agent-file defaults. Inherited models remain eligible for Jev. Under `jev`, these choices supply the fallback model, but do not skip classification. Under `shadow`, they remain the actual choice while Jev records a comparison. A selected Jev profile applies both its model and `thinkingLevel`. Pi clamps the requested thinking level to what that model supports. Shadow and fallback preserve the original model and thinking level.
 
@@ -731,7 +751,7 @@ Runtime tuning values set via `/agents` → Settings (max concurrency, max foreg
 
 **Precedence:** project overrides global on any field present in both. Missing fields fall back to the hardcoded defaults (max concurrency `10`, max foreground concurrency `0` = unlimited, default max turns unlimited, grace turns `5`, nested depth `2`, join mode `smart`, defaults enabled).
 
-Routing settings: `routingMode` (`auto | shadow | jev | off`, default `auto`), `customGuideline` (`string | false`, default unset) and `jev` (`{ TYPESAFE_API_KEY?: string, models: { model, description, thinkingLevel }[] } | false`, default unset). See [Model routing](#model-routing) for examples and the whole-block override rule.
+Routing settings: `routingMode` (`auto | shadow | jev | off`, default `auto`), `customGuideline` (`string | false`, default unset) and `jev` (`{ TYPESAFE_API_KEY?: string, models?: { model, description, thinkingLevel, instruction? }[] } | false`, default unset). See [Model routing](#model-routing) for examples and the whole-block override rule.
 
 **Nested depth** (`maxSubagentDepth`, default `2`): the hard ceiling on [nested delegation](#nested-subagents), counted from the main session (main = 0, its subagents = 1). `0` or `1` disables nesting project-wide regardless of any agent's `allowed_subagents`. Read when a subagent session is built, so a change applies to agents started after it.
 
@@ -768,13 +788,13 @@ The `~` marks it as pi's estimate rather than a billed figure. **A cost is shown
 
 Independent of `reportUsage`: this one is what you read, that one is what your session counts. Toggle via `/agents → Settings → Show cost`; applied live.
 
-**Show model** (`showModel`, default `false`): whether the widget's running rows name the model, thinking level, and configured OpenAI service tier when the effective API supports it:
+**Show model** (`showModel`, default `false`): whether the widget's running rows name the model, thinking level, and configured OpenAI service tier when the effective model supports it:
 
 ```text
 ├─ ⠹ Explore  inspect code · gpt-5 · thinking: high · service tier: priority · ↻3 · 8.2k token · 4.1s
 ```
 
-Off by default because the row already carries the description, turns, tool uses, tokens and elapsed time, and every character it gains is one the description loses on a narrow terminal. The other surfaces show the model and thinking level either way: the `Agent` tool result names them beside its tags, and the conversation viewer's `↳` row spells out the canonical `provider/model-id`. They include the configured service tier only when the effective API supports it, so an unsupported provider is not presented as honoring the request.
+Off by default because the row already carries the description, turns, tool uses, tokens and elapsed time, and every character it gains is one the description loses on a narrow terminal. The other surfaces show the model and thinking level either way: the `Agent` tool result names them beside its tags, and the conversation viewer's `↳` row spells out the canonical `provider/model-id`. They include the configured service tier only when the effective model supports it, so an unsupported provider is not presented as honoring the request.
 
 Both places report what the run *actually* used, read back from the child session once pi has resolved its defaults and clamped the level to what the model supports — not what the call asked for. Where those differ, the request is kept beside the effective value rather than dropped, whether pi clamped it or an agent file's frontmatter outranked it:
 
@@ -861,7 +881,7 @@ The four agent-lifecycle events — `subagents:started`, `:completed`, `:failed`
 
 `usage` answers the other question — what was billed — and so does include `cacheRead`, because the prefix really is re-read and re-charged on every call. It is a pi `Usage`, the same shape pi puts on `ToolResultEvent` and `AssistantMessage`, so `usage.cost.total` is where a listener already expects the money and anything pi adds to `Usage` arrives without a change here. Neither field derives from the other; `tokens` is a view model, `usage` is the data.
 
-Completed/failed payloads and persisted `subagents:record` entries also carry `routing` (`source`, `code`, `reason`, optional `model`, `thinkingLevel`, `suggestedModel`, `suggestedThinkingLevel`, supplied `description`, `confidence`, `unpriced`, `guidelinePath`, `guidelineHash`) and optional `routingUsage` (classifier-only Pi `Usage`). The guideline hash is SHA-256 of the original file contents. Coding `tokens` excludes classifier tokens; total cost includes the reported classifier cost once. Settings events omit `jev.TYPESAFE_API_KEY`.
+Completed/failed payloads and persisted `subagents:record` entries also carry `routing` (`source`, `code`, `reason`, optional `agent`, `suggestedAgent`, `model`, `thinkingLevel`, `suggestedModel`, `suggestedThinkingLevel`, supplied `description`, `confidence`, `unpriced`, `guidelinePath`, `guidelineHash`) and optional `routingUsage` (classifier-only Pi `Usage`). The guideline hash is SHA-256 of the original file contents. Coding `tokens` excludes classifier tokens; total cost includes the reported classifier cost once. Settings events omit `jev.TYPESAFE_API_KEY`.
 
 ## Cross-Extension RPC
 
@@ -1117,6 +1137,7 @@ src/
     agent-mention.ts      # `@` roster (running, resumable, and startable agents) + popup rows
     schedule-menu.ts      # /agents → Scheduled jobs submenu
     model-routing-menu.ts # Guideline/model descriptions and masked API-key editor
+    routing-status.ts  # Read-only effective routing configuration and retained decisions
     select-item.ts        # Collision-safe ctx.ui.select wrapper (numbered rows)
     workflow-card.ts      # Inline workflow card (tool result and session entry)
     workflow-dialog.ts    # /agents → Workflows two-pane inspector

package/dist/agent-manager.d.ts

@@ -172,7 +172,7 @@ interface SpawnOptions {
     /** Called on streaming text deltas from the assistant response. */
     onTextDelta?: (delta: string, fullText: string) => void;
     /** Called when the agent session is created (for accessing session stats). */
-    onSessionCreated?: (session: AgentSession) => void;
+    onSessionCreated?: (session: AgentSession, agentConfig?: AgentConfig) => void;
     /** Called at the end of each agentic turn with the cumulative count. */
     onTurnEnd?: (turnCount: number) => void;
     /** Called once per assistant message_end with that message's usage delta. */

package/dist/agent-manager.js

@@ -16,7 +16,7 @@
 import { randomUUID } from "node:crypto";
 import { statSync } from "node:fs";
 import { isAbsolute } from "node:path";
-import { resolveDefaultModel, resumeAgent, runAgent } from "./agent-runner.js";
+import { resolveDefaultModel, resumeAgent, runAgent, supportsServiceTier } from "./agent-runner.js";
 import { getAgentConfig } from "./agent-types.js";
 import { assignHandle, handleBase } from "./mention.js";
 import { describeModel } from "./model-resolver.js";
@@ -472,6 +472,7 @@ export class AgentManager {
             this.runningBackground++;
         else if (pool === "foreground")
             this.runningForeground++;
+        let routingInstruction;
         const config = options.agentConfig;
         const provenance = options.routing;
         const explicit = provenance
@@ -487,6 +488,8 @@ export class AgentManager {
         const route = policy.mode === "jev" || policy.mode === "shadow" ||
             (policy.mode === "auto" && !explicit && !config?.model && !config?.thinking && policy.source === "jev");
         if (!options.resumeSessionFile && provenance?.entrypoint !== "internal" && route) {
+            record.routing.code = "pending";
+            record.routing.reason = "Waiting for Jev to compare agent and model profiles";
             const stop = () => this.abort(id);
             options.signal?.addEventListener("abort", stop, { once: true });
             if (options.signal?.aborted)
@@ -501,16 +504,31 @@ export class AgentManager {
                         current.lifetimeUsage.cost = (current.lifetimeUsage.cost ?? 0) + usage.cost.total;
                         current = current.parentAgentId ? this.agents.get(current.parentAgentId) : undefined;
                     }
-                });
+                }, provenance?.allowedAgentTypes);
                 record.routing = { ...routed.decision, guidelinePath: policy.guidelinePath, guidelineHash: policy.guidelineHash };
                 if (routed.model && policy.mode === "shadow") {
-                    record.routing = { ...record.routing, code: "shadow", model: undefined, thinkingLevel: undefined, suggestedModel: routed.decision.model,
+                    record.routing = { ...record.routing, code: "shadow", model: undefined, agent: undefined, thinkingLevel: undefined, suggestedModel: routed.decision.model,
                         fallbackSource: policy.source, reason: "Jev suggested a model; shadow mode kept the default-priority model" };
                 }
                 else if (routed.model) {
+                    if (routed.agentConfig) {
+                        const selected = routed.agentConfig;
+                        type = selected.name;
+                        record.type = type;
+                        options.agentConfig = selected;
+                        options.maxTurns = selected.maxTurns ?? options.maxTurns;
+                        options.isolated = selected.isolated ?? options.isolated;
+                        options.inheritContext = selected.inheritContext ?? options.inheritContext;
+                        if (selected.isolation !== undefined)
+                            options.isolation = selected.isolation === "worktree" ? "worktree" : undefined;
+                        record.invocation = { ...record.invocation, maxTurns: options.maxTurns, isolated: options.isolated,
+                            inheritContext: options.inheritContext, isolation: options.isolation };
+                    }
+                    routingInstruction = routed.instruction;
                     options.model = routed.model;
                     options.thinkingLevel = routed.thinkingLevel;
                     record.invocation = { ...record.invocation, thinking: routed.thinkingLevel, requestedThinking: undefined };
+                    record.invocation.serviceTier = supportsServiceTier(routed.model) ? options.agentConfig?.serviceTier : undefined;
                 }
             }
             catch {
@@ -584,6 +602,7 @@ export class AgentManager {
             pi,
             agentId: id,
             agentConfig: options.agentConfig,
+            routingInstruction,
             model: options.model,
             maxTurns: options.maxTurns,
             isolated: options.isolated,
@@ -649,6 +668,9 @@ export class AgentManager {
                     // AND, one line later, being replaced by the effective one.
                     const requested = record.invocation.requestedThinking ?? record.invocation.thinking;
                     Object.assign(record.invocation, describeModel(session.model));
+                    if (options.agentConfig?.serviceTier) {
+                        record.invocation.serviceTier = supportsServiceTier(session.model) ? options.agentConfig.serviceTier : undefined;
+                    }
                     // Guarded for the reason above: a session that reports no level keeps
                     // the request rather than losing it. Overwriting unconditionally would
                     // turn an older or stubbed session into a blank `thinking:` tag, which
@@ -667,7 +689,7 @@ export class AgentManager {
                     }
                     record.pendingSteers = undefined;
                 }
-                options.onSessionCreated?.(session);
+                options.onSessionCreated?.(session, options.agentConfig);
             },
         })
             .then(async ({ responseText, session, aborted, steered, failure, structuredJson, structuredRetried }) => {

package/dist/agent-runner.d.ts

@@ -1,7 +1,7 @@
 /**
  * agent-runner.ts — Core execution engine: creates sessions, runs agents, collects results.
  */
-import type { Model } from "@earendil-works/pi-ai";
+import type { Api, Model } from "@earendil-works/pi-ai";
 import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { type AgentSession, DefaultResourceLoader, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { type NestedAgentManager } from "./nested-tools.js";
@@ -20,8 +20,8 @@ export declare const SUBAGENT_TOOL_NAMES: {
     readonly GET_RESULT: "get_subagent_result";
     readonly STEER: "steer_subagent";
 };
-/** Whether an API accepts the OpenAI `service_tier` request field. */
-export declare function isServiceTierApi(api: string | undefined): boolean;
+/** Copilot uses Responses but rejects OpenAI's `service_tier` request field. */
+export declare function supportsServiceTier(model: Pick<Model<Api>, "api" | "provider"> | undefined): boolean;
 /**
  * Add a custom agent's service tier to compatible provider requests.
  *
@@ -157,6 +157,8 @@ export interface ToolActivity {
     toolName: string;
 }
 export interface RunOptions {
+    /** Additional instructions from an applied Jev model profile. */
+    routingInstruction?: string;
     /** Snapshot of the selected definition for this branch. */
     agentConfig?: AgentConfig;
     /** ExtensionAPI instance — used for pi.exec() instead of execSync. */

package/dist/agent-runner.js

@@ -31,9 +31,9 @@ export const SUBAGENT_TOOL_NAMES = {
 const EXCLUDED_TOOL_NAMES = Object.values(SUBAGENT_TOOL_NAMES);
 /** APIs whose request payloads support OpenAI service tiers. */
 const SERVICE_TIER_APIS = new Set(["openai-codex-responses", "openai-responses"]);
-/** Whether an API accepts the OpenAI `service_tier` request field. */
-export function isServiceTierApi(api) {
-    return api !== undefined && SERVICE_TIER_APIS.has(api);
+/** Copilot uses Responses but rejects OpenAI's `service_tier` request field. */
+export function supportsServiceTier(model) {
+    return model !== undefined && model.provider !== "github-copilot" && SERVICE_TIER_APIS.has(model.api);
 }
 function isObjectPayload(payload) {
     return typeof payload === "object" && payload !== null && !Array.isArray(payload);
@@ -54,7 +54,7 @@ export function installServiceTierPayload(session, serviceTier) {
             ? await priorOnPayload(payload, requestModel)
             : undefined;
         const effectivePayload = replacement === undefined ? payload : replacement;
-        if (!isServiceTierApi(requestModel.api) || !isObjectPayload(effectivePayload)) {
+        if (!supportsServiceTier(requestModel) || !isObjectPayload(effectivePayload)) {
             return effectivePayload;
         }
         return { ...effectivePayload, service_tier: serviceTier };
@@ -515,6 +515,8 @@ export async function runAgent(ctx, type, prompt, options) {
             throw new Error(`No fallback config available for unknown type "${type}"`);
         systemPrompt = buildAgentPrompt({ ...fallback, name: type }, effectiveCwd, env, parentSystemPrompt, extras);
     }
+    if (options.routingInstruction)
+        systemPrompt += `\n\n<jev_model_instruction>\n${options.routingInstruction}\n</jev_model_instruction>`;
     // When skills is string[], we've already preloaded them into the prompt.
     // Still pass noSkills: true since we don't need the skill loader to load them again.
     const noSkills = skills === false || Array.isArray(skills);

package/dist/index.js

@@ -18,7 +18,7 @@ import { abortable } from "./abortable.js";
 import { hasAgentBadge, renderAgentName } from "./agent-color.js";
 import { buildNewAgentFile, disableInContent, enableInContent, isEmptyStub, locateAgentFile, personalAgentsDir, projectAgentsDir, serializeAgentFile } from "./agent-file-toggle.js";
 import { AgentManager, isTopLevelAgent } from "./agent-manager.js";
-import { getAgentConversation, getDefaultMaxTurns, getGraceTurns, getRememberAgents, isServiceTierApi, normalizeMaxTurns, resolveEffectiveMaxTurns, SUBAGENT_TOOL_NAMES, setDefaultMaxTurns, setGraceTurns, setRememberAgents, steerAgent } from "./agent-runner.js";
+import { getAgentConversation, getDefaultMaxTurns, getGraceTurns, getRememberAgents, normalizeMaxTurns, resolveEffectiveMaxTurns, SUBAGENT_TOOL_NAMES, setDefaultMaxTurns, setGraceTurns, setRememberAgents, steerAgent, supportsServiceTier } from "./agent-runner.js";
 import { BUILTIN_TOOL_NAMES, getAgentConfig, getAllTypes, getAvailableTypes, getConfig, getFallbackSubagent, isDefaultsDisabled, NO_FALLBACK, registerAgents, resolveSpawnType, resolveType, setDefaultsDisabled, setFallbackSubagent } from "./agent-types.js";
 import { inChildSessionContext } from "./child-context.js";
 import { registerRpcHandlers } from "./cross-extension-rpc.js";
@@ -1711,8 +1711,10 @@ Terse command-style prompts produce shallow, generic work.
             // downstream consumer keys off record.outputFile being set, so no spawn
             // path can re-enable the transcript by accident.
             const outputTranscript = customConfig?.outputTranscript ?? getOutputTranscriptDefault();
-            const attachTranscript = (rec, agentId) => {
-                if (!rec || !outputTranscript)
+            const attachTranscript = (rec, agentId, config = customConfig) => {
+                if (!rec || rec.outputFile || (!rec.session && (rec.status === "queued" || routingPolicy.mode === "jev" || rec.routing?.mode === "jev")))
+                    return;
+                if (!(config?.outputTranscript ?? getOutputTranscriptDefault()))
                     return;
                 rec.outputFile = createOutputFilePath(ctx.cwd, agentId, ctx.sessionManager.getSessionId());
                 writeInitialEntry(rec.outputFile, agentId, params.prompt, ctx.cwd);
@@ -1741,7 +1743,7 @@ Terse command-style prompts produce shallow, generic work.
                 modelName,
                 modelId,
                 thinking,
-                serviceTier: model && isServiceTierApi(model.api) ? customConfig?.serviceTier : undefined,
+                serviceTier: supportsServiceTier(model) ? customConfig?.serviceTier : undefined,
                 // Only set where the agent file outranked the caller, so the surfaces can
                 // disclose a parameter that was accepted but could not take effect (#182).
                 requestedThinking: resolvedConfig.overridden?.thinking,
@@ -1899,9 +1901,11 @@ Terse command-style prompts produce shallow, generic work.
                 // rather than closing over a value that doesn't exist yet.
                 let id;
                 const origBgOnSession = bgCallbacks.onSessionCreated;
-                bgCallbacks.onSessionCreated = (session) => {
+                bgCallbacks.onSessionCreated = (session, config) => {
                     origBgOnSession(session);
                     const rec = manager.getRecord(id);
+                    attachTranscript(rec, id, config);
+                    bgState.maxTurns = normalizeMaxTurns(rec?.invocation?.maxTurns ?? getDefaultMaxTurns());
                     if (rec?.outputFile) {
                         rec.outputCleanup = streamToOutputFile(session, rec.outputFile, id, ctx.cwd);
                     }
@@ -1938,6 +1942,8 @@ Terse command-style prompts produce shallow, generic work.
                 // copy is an awaited git call. Wait for it here, after the synchronous
                 // wiring above, so a strict-isolation failure still fails THIS tool
                 // call instead of being reported as a subagent that ran (#179).
+                if (routingPolicy.mode === "jev" && record?.startGate)
+                    await record.startGate;
                 await manager.awaitStartup(id);
                 if (joinMode == null || joinMode === 'async') {
                     // Foreground/no join mode or explicit async — not part of any batch
@@ -1959,14 +1965,14 @@ Terse command-style prompts produce shallow, generic work.
                 // Emit created event
                 pi.events.emit("subagents:created", {
                     id,
-                    type: subagentType,
+                    type: record?.type ?? subagentType,
                     description: params.description,
                     isBackground: true,
                 });
                 const isQueued = record?.status === "queued";
                 return textResult(`${fallbackNote}Agent ${isQueued ? "queued" : "started"} in background.\n` +
                     `Agent ID: ${id}\n` +
-                    `Type: ${displayName}\n` +
+                    `Type: ${record ? getDisplayName(record.type) : displayName}\n` +
                     `Description: ${params.description}\n` +
                     (record?.outputFile ? `Output file: ${record.outputFile}\n` : "") +
                     (isQueued ? `Position: queued (max ${manager.getMaxConcurrent()} concurrent)\n` : "") +
@@ -2016,7 +2022,7 @@ Terse command-style prompts produce shallow, generic work.
             // The output file path is set synchronously after spawn (below),
             // before onSessionCreated fires — same pattern as background agents.
             const origOnSession = fgCallbacks.onSessionCreated;
-            fgCallbacks.onSessionCreated = (session) => {
+            fgCallbacks.onSessionCreated = (session, config) => {
                 origOnSession(session);
                 // It really started — stop reporting it as queued, and repaint now
                 // rather than leaving the stale line up for the next spinner tick.
@@ -2038,6 +2044,8 @@ Terse command-style prompts produce shallow, generic work.
                 // Stream conversation to output file (foreground agent logging)
                 if (fgId) {
                     const rec = manager.getRecord(fgId);
+                    attachTranscript(rec, fgId, config);
+                    fgState.maxTurns = normalizeMaxTurns(rec?.invocation?.maxTurns ?? getDefaultMaxTurns());
                     if (rec?.outputFile) {
                         rec.outputCleanup = streamToOutputFile(session, rec.outputFile, fgId, ctx.cwd);
                     }
@@ -2752,7 +2760,7 @@ Terse command-style prompts produce shallow, generic work.
             await showAgentsMenu(ctx);
         }
         else if (choice === "Model routing") {
-            const patch = await showRoutingMenu(ctx);
+            const patch = await showRoutingMenu(ctx, () => manager.listAgents());
             if (patch) {
                 const toast = saveAndEmitChanged({ ...snapshotSettings(), ...patch }, "Model routing settings updated", (event, payload) => pi.events.emit(event, payload), ctx.cwd);
                 ctx.ui.notify(toast.message, toast.level);
@@ -3066,7 +3074,7 @@ description: <one-line description shown in UI>
 color: <optional agent name badge color: red, blue, green, yellow, purple, orange, pink, cyan, an Agency Agents alias, or quoted "#RRGGBB">
 tools: <comma-separated built-in tools: read, bash, edit, write, grep, find, ls. Use "none" for no tools. Omit for all tools>
 model: <optional model as "provider/modelId", e.g. "anthropic/claude-haiku-4-5". Omit to inherit parent model>
-service_tier: <optional OpenAI Responses/Codex processing tier: auto, default, flex, priority, or scale. Use only with those APIs; omit for other APIs or the provider default>
+service_tier: <optional OpenAI Responses/Codex processing tier: auto, default, flex, priority, or scale. Omit for GitHub Copilot, other APIs, or the provider default>
 thinking: <optional thinking level: ${THINKING_LEVELS.join(", ")}. Omit to inherit>
 max_turns: <optional max agentic turns. 0 or omit for unlimited (default)>
 prompt_mode: <"replace" (body IS the full system prompt) or "append" (body is appended to default prompt). Default: replace>
@@ -3097,7 +3105,7 @@ Guidelines for choosing settings:
 - Use prompt_mode: replace for fully custom agents with their own personality/instructions
 - Set inherit_context: true if the agent needs to know what was discussed in the parent conversation
 - Set isolated: true if the agent should NOT have access to MCP servers or other extensions
-- Set service_tier: priority only when the model uses an OpenAI Responses/Codex API; omit it for other providers
+- Set service_tier: priority only when the model uses an OpenAI Responses/Codex API; omit it for GitHub Copilot and other providers that do not support it
 - Set output_transcript: false to skip writing this agent's transcript; this alone doesn't keep the run off disk (persist_session, isolation: worktree commits, and memory still write) — set those too if that's the goal
 - Only include frontmatter fields that differ from defaults — omit fields where the default is fine
 
@@ -3397,7 +3405,7 @@ Write the file using the write tool. Only write the file, nothing else.`;
                 {
                     id: "showModel",
                     label: "Show model",
-                    description: "Name the model driving each agent, its thinking level, and configured OpenAI service tier when the effective API supports it, on the widget's running rows. The Agent tool result and the conversation viewer show these details when applicable — this adds them to the widget, where the row is already dense.",
+                    description: "Name the model driving each agent, its thinking level, and configured OpenAI service tier when the effective model supports it, on the widget's running rows. The Agent tool result and the conversation viewer show these details when applicable — this adds them to the widget, where the row is already dense.",
                     currentValue: isShowModelEnabled() ? "on" : "off",
                     values: ["on", "off"],
                 },

package/dist/model-routing.d.ts

@@ -11,6 +11,7 @@ export interface RoutingPolicy {
         name: string;
         description: string;
     }[];
+    agentProfiles?: AgentConfig[];
     guideline?: string;
     guidelinePath?: string;
     guidelineHash?: string;
@@ -21,6 +22,8 @@ export interface RoutingPolicy {
 export interface RoutingInput {
     /** Private launch snapshot; never accepted from external callers. */
     policy?: RoutingPolicy;
+    /** Parent permission boundary, supplied only by the nested tool. */
+    allowedAgentTypes?: string[];
     modelExplicit: boolean;
     thinkingExplicit: boolean;
     entrypoint: "agent" | "nested" | "workflow" | "schedule" | "internal";
@@ -30,9 +33,11 @@ export interface RoutingDecision {
     source: RoutingSource;
     fallbackSource?: RoutingSource;
     reason: string;
-    code: "baseline" | "explicit" | "off" | "shadow" | "config_unavailable" | "credentials_unavailable" | "guideline_unavailable" | "no_candidates" | "classifier_unavailable" | "cancelled" | "timeout" | "invalid_answer" | "abstained" | "unavailable_choice" | "selected" | "classifier_error";
+    code: "baseline" | "pending" | "explicit" | "off" | "shadow" | "config_unavailable" | "credentials_unavailable" | "guideline_unavailable" | "no_candidates" | "classifier_unavailable" | "cancelled" | "timeout" | "invalid_answer" | "abstained" | "unavailable_choice" | "selected" | "classifier_error";
     model?: string;
     suggestedModel?: string;
+    agent?: string;
+    suggestedAgent?: string;
     thinkingLevel?: ThinkingLevel;
     suggestedThinkingLevel?: ThinkingLevel;
     description?: string;
@@ -44,14 +49,26 @@ export interface RoutingDecision {
 export declare function loadRoutingPolicy(cwd: string, loadedAgents?: Map<string, AgentConfig>): RoutingPolicy;
 /** Added in every description mode and refreshed before each main-agent turn. */
 export declare function routingGuidance(policy: RoutingPolicy): string;
+export declare function eligibleModels(ctx: ExtensionContext): Map<string, Model<Api>>;
+export interface RoutingCandidate {
+    instruction?: string;
+    model: string;
+    description: string;
+    thinkingLevel?: ThinkingLevel;
+    agentConfig?: AgentConfig;
+}
+/** Keep profiles distinct even when multiple specialists use the same model. */
+export declare function routingCandidates(ctx: ExtensionContext, policy: RoutingPolicy, baseline: Model<Api> | undefined, allowedAgentTypes?: readonly string[]): RoutingCandidate[];
 /** Bounded classifier pool, independent of agent concurrency and nesting. */
 export declare class ModelRouter {
     private active;
     private waiters;
     private acquire;
-    choose(ctx: ExtensionContext, policy: RoutingPolicy, prompt: string, description: string, baseline: Model<Api> | undefined, signal: AbortSignal, onUsage: (usage: Usage) => void): Promise<{
+    choose(ctx: ExtensionContext, policy: RoutingPolicy, prompt: string, description: string, baseline: Model<Api> | undefined, signal: AbortSignal, onUsage: (usage: Usage) => void, allowedAgentTypes?: readonly string[]): Promise<{
         model?: Model<Api>;
         thinkingLevel?: ThinkingLevel;
+        instruction?: string;
+        agentConfig?: AgentConfig;
         decision: RoutingDecision;
     }>;
 }