@signalridge/pi-subagents 1.8.0 → 1.9.0

package/src/types.ts

@@ -4,13 +4,11 @@
 
 import type { ThinkingLevel } from "@earendil-works/pi-ai";
 import type { AgentSession } from "@earendil-works/pi-coding-agent";
-import type { WorkflowTier } from "@signalridge/pi-subagents-protocol";
 import type { AgentTierResolutionSnapshot } from "./agent-tiers.js";
 import type { LifetimeUsage } from "./usage.js";
-import type { WorkflowTierResolutionSnapshot } from "./workflow-tiers.js";
 import type { WorktreeCleanupResult, WorktreeInfo } from "./worktree.js";
 
-export type { ThinkingLevel, WorkflowTier };
+export type { ThinkingLevel };
 
 /** Agent type: any string name (built-in defaults or user-defined). */
 export type SubagentType = string;
@@ -237,18 +235,18 @@ export interface ResumableAgentEntry {
 }
 
 export interface AgentInvocation {
-  /** Short display name, e.g. "haiku" — only set when different from parent. */
+  /**
+   * Short display name, e.g. "haiku". Set only when the spawn is not simply
+   * running the parent's model: an untiered spawn that resolved a different one,
+   * or a tier whose profile pins one. A tier that inherits leaves this absent,
+   * which is how the UI says "same model as the parent".
+   */
   modelName?: string;
   thinking?: ThinkingLevel;
-  /** Semantic workflow tier selected by the workflow definition. */
-  tier?: WorkflowTier;
-  /** Immutable model/thinking resolution captured by pi-subagents. */
-  tierSnapshot?: WorkflowTierResolutionSnapshot;
   /**
-   * User-named tier applied to an ordinary spawn. Deliberately a separate field
-   * from `tier` above: that one is the workflow protocol's small/medium/large
-   * union, and widening it to hold arbitrary names would take the protocol's
-   * exhaustiveness with it.
+   * The tier applied to this spawn, from any caller — the Agent tool, nested
+   * delegation, the scheduler, or a managed workflow call. All four name a key
+   * from the one `agentTiers` catalogue.
    */
   agentTier?: string;
   /** Immutable model/thinking resolution for `agentTier`. */
@@ -269,7 +267,6 @@ export type AgentRecordSnapshot = Omit<Readonly<AgentRecord>, "session" | "abort
   readonly lifetimeUsage: Readonly<LifetimeUsage>;
   readonly pendingSteers?: readonly string[];
   readonly invocation?: Readonly<AgentInvocation> & {
-    readonly tierSnapshot?: Readonly<NonNullable<AgentInvocation["tierSnapshot"]>>;
     readonly agentTierSnapshot?: Readonly<NonNullable<AgentInvocation["agentTierSnapshot"]>>;
   };
 };

package/src/ui/agent-display.ts

@@ -174,7 +174,7 @@ export function buildInvocationTags(
 ): { modelName?: string; tags: string[] } {
   const tags: string[] = [];
   if (!invocation) return { tags };
-  if (invocation.tier) tags.push(`tier: ${invocation.tier}`);
+  if (invocation.agentTier) tags.push(`tier: ${invocation.agentTier}`);
   if (invocation.thinking) tags.push(`thinking: ${invocation.thinking}`);
   if (invocation.isolated) tags.push("isolated");
   if (invocation.isolation === "worktree") tags.push("worktree");

package/src/workflow-tiers.ts

@@ -1,202 +0,0 @@
-/** Resolution and audit snapshots for semantic workflow model tiers. */
-import {
-  type Api,
-  clampThinkingLevel,
-  getSupportedThinkingLevels,
-  type Model,
-} from "@earendil-works/pi-ai";
-import type { WorkflowTier } from "@signalridge/pi-subagents-protocol";
-import { type ModelRegistry, resolveModel } from "./model-resolver.js";
-import {
-  DEFAULT_WORKFLOW_TIER_PROFILES,
-  type WorkflowSettings,
-  type WorkflowThinking,
-  type WorkflowTierProfile,
-} from "./settings.js";
-import type { AgentConfig, ThinkingLevel } from "./types.js";
-
-let workflowSettings: WorkflowSettings = {};
-
-export type { WorkflowTier };
-
-export type WorkflowResolutionSource = "frontmatter" | "tier" | "parent";
-
-/** Durable, JSON-safe explanation of the policy used for one managed spawn. */
-export interface WorkflowTierResolutionSnapshot {
-  tier: WorkflowTier;
-  /** Effective provider/model id, after resolving or inheriting the parent. */
-  model?: string;
-  /** Effective thinking level; omitted when the selected model supports no level. */
-  thinking?: ThinkingLevel;
-  configuredModel?: string;
-  configuredThinking?: WorkflowThinking;
-  requestedThinking?: ThinkingLevel;
-  modelSource: WorkflowResolutionSource;
-  thinkingSource: WorkflowResolutionSource;
-  /** True when pi-ai lowered the requested level for the selected model. */
-  clamped?: boolean;
-  diagnostic?: string;
-}
-
-export interface WorkflowTierResolution {
-  model?: Model<Api>;
-  thinkingLevel?: ThinkingLevel;
-  snapshot?: WorkflowTierResolutionSnapshot;
-}
-
-export interface ResolveWorkflowTierInput {
-  /** Explicit semantic tier from a workflow task. */
-  tier?: WorkflowTier;
-  /** Settings are passed by pi-subagents; workflows never resolve policy. */
-  settings?: WorkflowSettings;
-  agentConfig?: AgentConfig;
-  /** Direct model objects take precedence over agent frontmatter and tiers. */
-  directModel?: Model<Api>;
-  /** Direct model references take precedence over agent frontmatter and tiers. */
-  modelOverride?: string;
-  thinkingOverride?: ThinkingLevel;
-  parentModel?: Model<Api>;
-  parentThinking?: ThinkingLevel;
-  modelRegistry: ModelRegistry<Model<Api>>;
-}
-
-export function isWorkflowTier(value: unknown): value is WorkflowTier {
-  return value === "small" || value === "medium" || value === "large";
-}
-
-export function getWorkflowSettings(): WorkflowSettings {
-  return structuredClone(workflowSettings);
-}
-
-export function setWorkflowSettings(settings: WorkflowSettings): void {
-  workflowSettings = structuredClone(settings);
-}
-
-
-function effectiveModelId(model: Model<Api> | undefined): string | undefined {
-  return model ? `${model.provider}/${model.id}` : undefined;
-}
-
-/**
- * Resolve model and thinking independently, preserving agent frontmatter as the
- * authoritative override. A tier only fills fields omitted by frontmatter; the
- * parent session is the final fallback. An unavailable model explicitly selected
- * by a tier fails closed instead of silently changing the model.
- */
-function isCompleteProfile(value: unknown): value is WorkflowTierProfile {
-  if (!value || typeof value !== "object" || Array.isArray(value)) return false;
-  const profile = value as Record<string, unknown>;
-  return (
-    typeof profile.model === "string" &&
-    profile.model.length > 0 &&
-    typeof profile.thinking === "string" &&
-    ["inherit", "minimal", "low", "medium", "high", "xhigh", "max"].includes(profile.thinking)
-  );
-}
-
-/**
- * Resolve model and thinking independently, preserving agent frontmatter as the
- * authoritative override. A tier only fills fields omitted by frontmatter; the
- * parent session is the final fallback. The configured default tier is applied
- * only inside pi-subagents, never by the workflow package.
- */
-export function resolveWorkflowTier(input: ResolveWorkflowTierInput): WorkflowTierResolution {
-  if (input.tier !== undefined && !isWorkflowTier(input.tier)) {
-    throw new Error("workflow tier must be one of small, medium, or large");
-  }
-
-  const workflow = input.settings ?? workflowSettings;
-  if (input.tier === undefined && workflow.blockedDefaultTier) {
-    throw new Error("workflow defaultTier is blocked by malformed configuration");
-  }
-  const tier = input.tier ?? workflow.defaultTier;
-  if (tier !== undefined && !isWorkflowTier(tier)) {
-    throw new Error("workflow defaultTier must be one of small, medium, or large");
-  }
-
-  const defaultProfile = tier === undefined ? undefined : DEFAULT_WORKFLOW_TIER_PROFILES[tier];
-  const configuredProfile = tier === undefined ? undefined : workflow.tiers?.[tier];
-  if (tier !== undefined && workflow.blockedTiers?.includes(tier)) {
-    throw new Error(`workflow tier "${tier}" is blocked by malformed configuration`);
-  }
-  if (configuredProfile !== undefined && !isCompleteProfile(configuredProfile)) {
-    // This should be unreachable after settings validation. Throwing here keeps
-    // a forged in-process settings object from degrading into parent policy.
-    throw new Error(`workflow tier "${tier}" has an incomplete model/thinking profile`);
-  }
-  const profile = tier === undefined
-    ? undefined
-    : { ...defaultProfile, ...(configuredProfile ?? {}) };
-
-  const tierConfiguredModel = profile?.model;
-  const configuredModel = input.modelOverride ?? input.agentConfig?.model ?? tierConfiguredModel;
-  const modelSource: WorkflowResolutionSource = input.directModel !== undefined || input.modelOverride !== undefined || input.agentConfig?.model !== undefined
-    ? "frontmatter"
-    : configuredModel !== undefined && configuredModel !== "inherit"
-      ? "tier"
-      : "parent";
-  let model = input.directModel ?? input.parentModel;
-  let diagnostic: string | undefined;
-
-  if (input.directModel === undefined && configuredModel !== undefined && configuredModel !== "inherit") {
-    const resolved = resolveModel(configuredModel, input.modelRegistry);
-    if (typeof resolved === "string") {
-      const tierOwnsModel =
-        input.modelOverride === undefined &&
-        input.agentConfig?.model === undefined &&
-        tierConfiguredModel !== undefined &&
-        tierConfiguredModel !== "inherit";
-      if (tierOwnsModel) {
-        throw new Error(`workflow tier "${tier}" has an unavailable model: ${resolved}`);
-      }
-      diagnostic = `${resolved} Falling back to the parent model.`;
-    } else {
-      model = resolved;
-    }
-  }
-
-  const tierConfiguredThinking = profile?.thinking;
-  const configuredThinking = input.thinkingOverride ?? input.agentConfig?.thinking ?? tierConfiguredThinking;
-  const thinkingSource: WorkflowResolutionSource = input.thinkingOverride !== undefined || input.agentConfig?.thinking !== undefined
-    ? "frontmatter"
-    : tierConfiguredThinking !== undefined && tierConfiguredThinking !== "inherit"
-      ? "tier"
-      : "parent";
-  const requestedThinking = configuredThinking === "inherit" ? input.parentThinking : configuredThinking ?? input.parentThinking;
-  let thinkingLevel = requestedThinking;
-  let clamped = false;
-
-  if (model && requestedThinking) {
-    const supported = getSupportedThinkingLevels(model);
-    const clampedLevel = clampThinkingLevel(model, requestedThinking);
-    if (clampedLevel !== requestedThinking) {
-      clamped = true;
-      const supportedText = supported.join(", ");
-      const clampDiagnostic = `Thinking level "${requestedThinking}" is not supported by ${effectiveModelId(model) ?? "the selected model"}; using "${clampedLevel}" (supported: ${supportedText}).`;
-      diagnostic = diagnostic ? `${diagnostic} ${clampDiagnostic}` : clampDiagnostic;
-    }
-    // "off" is a ModelThinkingLevel sentinel, not a ThinkingLevel accepted by
-    // AgentSession options. Omitting the option preserves the provider's off behavior.
-    thinkingLevel = clampedLevel === "off" ? undefined : clampedLevel as ThinkingLevel;
-  }
-
-  if (tier === undefined) return { model, thinkingLevel };
-
-  const snapshot: WorkflowTierResolutionSnapshot = {
-    tier,
-    ...(effectiveModelId(model) ? { model: effectiveModelId(model) } : {}),
-    ...(thinkingLevel ? { thinking: thinkingLevel } : {}),
-    ...(configuredModel !== undefined ? { configuredModel } : {}),
-    ...(configuredThinking !== undefined ? { configuredThinking } : {}),
-    ...(requestedThinking !== undefined ? { requestedThinking } : {}),
-    modelSource,
-    thinkingSource,
-    ...(clamped ? { clamped: true } : {}),
-    ...(diagnostic ? { diagnostic } : {}),
-  };
-
-  return { model, thinkingLevel, snapshot };
-}
-
-/** Compatibility alias for callers that use the shorter tier terminology. */
-export const resolveTier = resolveWorkflowTier;