@oh-my-pi/pi-coding-agent 16.1.22 → 16.2.0

package/dist/types/advisor/__tests__/emission-guard.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/types/advisor/emission-guard.d.ts

@@ -0,0 +1,73 @@
+/**
+ * Per-session policy gate for advisor `advise()` calls.
+ *
+ * The advisor system prompt tells the watcher model:
+ *
+ * > at most one `advise` per update
+ * > NEVER repeat advice you already gave, and NEVER send the same advice twice
+ *
+ * Real advisor models violate this. Issue #3520 captured a session where
+ * `__advisor.jsonl` recorded 309 `advise` calls covering 92 unique notes —
+ * 114× `Stop.`, 52× `No issue; continue.`, 41× `Done.` — flooding the primary
+ * transcript with `<advisory severity="blocker">Stop.</advisory>` after the
+ * task was already complete. The fix is to make the rules load-bearing in code
+ * instead of prose: silently drop duplicates, content-free self-talk, and
+ * over-budget calls at the `enqueueAdvice` boundary so the primary stays
+ * clean even when the advisor misbehaves.
+ *
+ * The gate is intentionally invisible to the advisor model — `AdviseTool`
+ * still returns `Recorded.` for a suppressed call. Surfacing "suppressed"
+ * back into advisor context risks the model rephrasing the same useless note
+ * to bypass the dedupe ("Stop.", then "Halt." then "Stop now.").
+ */
+/**
+ * Case-insensitive, punctuation-folded normalization. Collapses every run of
+ * non-letter / non-digit characters into a single space and trims, so
+ * `"Stop."`, `"*Stop*"`, and `"  stop  "` all key to `stop`, while
+ * `"No issue; continue."` keys to `no issue continue`.
+ *
+ * Exported for tests.
+ */
+export declare function normalizeAdvisorNote(note: string): string;
+/**
+ * Decides whether an advisor `advise()` call should reach the primary agent.
+ *
+ * Enforces — in this order — the noise filter, session-scoped exact-text
+ * dedupe (FIFO-evicted at {@link DEFAULT_HISTORY_CAPACITY}), and a per-update
+ * rate limit of one accepted note per advisor model prompt. Suppressed calls
+ * never consume the per-update budget — a noise call doesn't burn the slot
+ * for a real concern that follows in the same update.
+ *
+ * Reset on advisor reset (compaction, session switch, `/new`) via
+ * {@link reset}. Per-update gate is cleared at the start of every advisor
+ * `agent.prompt()` cycle via {@link beginUpdate}.
+ */
+export declare class AdvisorEmissionGuard {
+    #private;
+    constructor(opts?: {
+        capacity?: number;
+    });
+    /**
+     * Drop all dedupe and per-update state. Called from
+     * `AgentSession#resetAdvisorSessionState()` whenever the advisor runtime is
+     * reset — same boundary as `yieldQueue.clear("advisor")`, so a re-primed
+     * advisor can re-raise old issues (the primary transcript was rewritten).
+     */
+    reset(): void;
+    /**
+     * Clear the per-update rate-limit gate. Called by `AdvisorRuntime` right
+     * before each `agent.prompt(batch)` invocation so the next advisor model
+     * cycle starts with a fresh budget of one advise.
+     */
+    beginUpdate(): void;
+    /**
+     * Whether the proposed note should reach the primary. On `true` the gate
+     * has already recorded the note (consumed the per-update budget and added
+     * it to the dedupe history) — caller delivers the note. On `false` the
+     * caller drops it.
+     *
+     * Empty / whitespace-only notes are suppressed; the model's
+     * tool-args contract still requires a non-empty string but defense-in-depth.
+     */
+    accept(note: string): boolean;
+}

package/dist/types/advisor/index.d.ts

@@ -1,4 +1,5 @@
 export * from "./advise-tool";
+export * from "./emission-guard";
 export * from "./runtime";
 export * from "./transcript-recorder";
 export * from "./watchdog";

package/dist/types/advisor/runtime.d.ts

@@ -26,6 +26,13 @@ export interface AdvisorRuntimeHost {
      * the primary's next compaction triggers {@link AdvisorRuntime.reset}).
      */
     maintainContext?(incomingTokens: number): Promise<boolean>;
+    /**
+     * Called immediately before each `agent.prompt(batch)` cycle. Lets the host
+     * clear per-update advisor state — currently the one-advise-per-update gate
+     * in {@link AdvisorEmissionGuard}, which the host owns because it is the
+     * one that routes `advise()` results back to the primary.
+     */
+    beginAdvisorUpdate?(): void;
 }
 export declare class AdvisorRuntime {
     #private;

package/dist/types/advisor/watchdog.d.ts

@@ -1,3 +1,5 @@
+import type { ActiveRepoContext } from "../utils/active-repo-context";
+export declare function formatActiveRepoWatchdogPrompt(activeRepoContext: ActiveRepoContext): string;
 /**
  * Discover and load WATCHDOG.md files walking up from cwd, project .omp folder, and user agent dir.
  * Returns formatted watchdog file blocks ready to be appended to the advisor system prompt.

package/dist/types/cli/flag-tables.d.ts

@@ -46,6 +46,7 @@ export interface ParseDeps {
     };
     parseThinking: (value: string | null | undefined) => ConfiguredThinkingLevel | undefined;
     builtinToolNames: readonly string[];
+    normalizeToolNames: (values: Iterable<string>) => string[];
     thinkingEfforts: readonly string[];
 }
 export type StringSetter = (result: Args, value: string, deps: ParseDeps) => void;

package/dist/types/cli/gallery-fixtures/search.d.ts

@@ -1,3 +1,3 @@
-/** Gallery fixtures for the search tools (search, search_tool_bm25, ast_grep). */
+/** Gallery fixtures for the search tools (grep, search_tool_bm25, ast_grep). */
 import type { GalleryFixture } from "./types";
 export declare const searchFixtures: Record<string, GalleryFixture>;

package/dist/types/cli/gc-cli.d.ts

@@ -0,0 +1,58 @@
+export interface GcCommandFlags {
+    apply?: boolean;
+    json?: boolean;
+    agentDir?: string;
+    blobs?: boolean;
+    archive?: boolean;
+    wal?: boolean;
+    coldArchiveAfterDays?: number;
+    retainNewestGlobal?: number;
+    retainNewestPerCwd?: number;
+}
+export interface GcCommandArgs {
+    flags: GcCommandFlags;
+}
+export interface BlobGcResult {
+    referenced: number;
+    candidates: number;
+    wouldDelete: number;
+    deleted: number;
+    bytes: number;
+    errors: string[];
+}
+export interface ArchiveGcResult {
+    scanned: number;
+    skippedActive: number;
+    keptNewestGlobal: number;
+    keptNewestPerCwd: number;
+    wouldArchive: number;
+    archived: number;
+    historyRowsDeleted: number;
+    ftsRebuilt: boolean;
+    errors: string[];
+}
+export interface WalCheckpointResult {
+    dbPath: string;
+    walBytes: number;
+    wouldCheckpoint: boolean;
+    checkpointed: boolean;
+    busy: number;
+    log: number;
+    checkpointedFrames: number;
+}
+export interface WalGcResult {
+    databases: WalCheckpointResult[];
+    walBytes: number;
+    wouldCheckpoint: boolean;
+    checkpointed: boolean;
+}
+export interface GcResult {
+    agentDir: string;
+    apply: boolean;
+    blobs?: BlobGcResult;
+    archive?: ArchiveGcResult;
+    wal?: WalGcResult;
+    lockPath: string;
+}
+export declare function collectGcErrors(result: GcResult): string[];
+export declare function runGcCommand(args: GcCommandArgs): Promise<GcResult>;

package/dist/types/collab/display-name.d.ts

@@ -0,0 +1,3 @@
+import type { InteractiveModeContext } from "../modes/types";
+/** Display name for this process's user in collab sessions. */
+export declare function collabDisplayName(ctx: InteractiveModeContext): string;

package/dist/types/collab/host.d.ts

@@ -10,8 +10,6 @@
  */
 import type { InteractiveModeContext } from "../modes/types";
 import { type CollabParticipant } from "./protocol";
-/** Display name for this process's user in collab sessions. */
-export declare function collabDisplayName(ctx: InteractiveModeContext): string;
 export declare class CollabHost {
     #private;
     constructor(ctx: InteractiveModeContext);

package/dist/types/commands/gc.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Run on-disk storage maintenance.
+ */
+import { Command } from "@oh-my-pi/pi-utils/cli";
+export default class Gc extends Command {
+    static description: string;
+    static flags: {
+        apply: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"boolean"> & {
+            description: string;
+        };
+        json: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"boolean"> & {
+            description: string;
+        };
+        "agent-dir": import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"string"> & {
+            description: string;
+        };
+        blobs: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"boolean"> & {
+            description: string;
+        };
+        archive: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"boolean"> & {
+            description: string;
+        };
+        wal: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"boolean"> & {
+            description: string;
+        };
+        "cold-archive-after-days": import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"integer"> & {
+            description: string;
+        };
+        "retain-newest-global": import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"integer"> & {
+            description: string;
+        };
+        "retain-newest-per-cwd": import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"integer"> & {
+            description: string;
+        };
+    };
+    run(): Promise<void>;
+}

package/dist/types/commands/worktree.d.ts

@@ -1,6 +1,3 @@
-/**
- * List and clean up agent-managed git worktrees under `~/.omp/wt`.
- */
 import { Command } from "@oh-my-pi/pi-utils/cli";
 export default class Worktree extends Command {
     static description: string;

package/dist/types/config/inline-tool-descriptors-mode.d.ts

@@ -0,0 +1,13 @@
+/**
+ * Resolves whether full tool descriptors should be inlined into the system
+ * prompt (and stripped from provider tool schemas) for a given model and
+ * setting.
+ *
+ * `auto` enforces a per-model policy: inline for Gemini models, off otherwise.
+ * Gemini benefits from descriptors in-prompt; other providers keep them in the
+ * tool schemas. `on`/`off` are explicit user overrides.
+ *
+ * @param modelId Canonical model id (e.g. `gemini-3-pro`); resolve aliases via
+ *   `ModelRegistry.getCanonicalId` before calling so `auto` classifies correctly.
+ */
+export declare function shouldInlineToolDescriptors(setting: "auto" | "on" | "off" | undefined, modelId: string | undefined): boolean;

package/dist/types/config/model-discovery.d.ts

@@ -6,9 +6,10 @@
  * discovery lives in pi-catalog's provider-models.
  */
 import { type ApiKey, type FetchImpl } from "@oh-my-pi/pi-ai";
-import type { Api, Model } from "@oh-my-pi/pi-ai/types";
+import type { Api, Model, RemoteCompactionConfig } from "@oh-my-pi/pi-ai/types";
 import type { ModelSpec } from "@oh-my-pi/pi-catalog/types";
 import type { ProviderDiscovery } from "./models-config-schema";
+export declare const DISCOVERY_DEFAULT_CONTEXT_WINDOW = 128000;
 export declare const DISCOVERY_DEFAULT_MAX_TOKENS = 32768;
 export declare function getImplicitOllamaBaseUrl(): string;
 export declare function getOllamaContextLengthOverride(): number | undefined;
@@ -20,6 +21,7 @@ export interface DiscoveryProviderConfig {
     baseUrl?: string;
     headers?: Record<string, string>;
     compat?: ModelSpec<Api>["compat"];
+    remoteCompaction?: RemoteCompactionConfig<Api>;
     discovery: ProviderDiscovery;
     optional?: boolean;
 }
@@ -40,6 +42,7 @@ export declare function discoverOllamaModels(providerConfig: DiscoveryProviderCo
 export declare function discoverLlamaCppModels(providerConfig: DiscoveryProviderConfig, ctx: DiscoveryContext): Promise<Model<Api>[]>;
 export declare function discoverLlamaCppModelContextWindow(model: Pick<Model<Api>, "provider" | "id" | "baseUrl" | "headers">, ctx: DiscoveryContext): Promise<number | undefined>;
 export declare function discoverOpenAIModelsList(providerConfig: DiscoveryProviderConfig, ctx: DiscoveryContext): Promise<Model<Api>[]>;
+export declare function discoverLiteLLMModels(providerConfig: DiscoveryProviderConfig, ctx: DiscoveryContext): Promise<Model<Api>[]>;
 /**
  * Discover models from an Anthropic+OpenAI-compatible reseller proxy that
  * exposes both `/v1/messages` and `/v1/chat/completions`, advertising each
@@ -56,3 +59,5 @@ export declare function discoverOpenAIModelsList(providerConfig: DiscoveryProvid
  * URL (which ends in `/v1`) round-trips correctly.
  */
 export declare function discoverProxyModels(providerConfig: DiscoveryProviderConfig, ctx: DiscoveryContext): Promise<Model<Api>[]>;
+export declare function normalizeLiteLLMDiscoveryBaseUrl(baseUrl?: string): string;
+export declare function normalizeOpenAIModelsListBaseUrl(baseUrl?: string): string;

package/dist/types/config/model-registry.d.ts

@@ -1,4 +1,4 @@
-import type { Api, Context, Model, ModelSpec, SimpleStreamOptions, ThinkingConfig } from "@oh-my-pi/pi-ai/types";
+import type { Api, Context, Model, ModelSpec, RemoteCompactionConfig, SimpleStreamOptions, ThinkingConfig } from "@oh-my-pi/pi-ai/types";
 import type { AssistantMessageEventStream } from "@oh-my-pi/pi-ai/utils/event-stream";
 import { type ModelRefreshStrategy } from "@oh-my-pi/pi-catalog/model-manager";
 import type { ApiKeyResolver, FetchImpl } from "@oh-my-pi/pi-ai";
@@ -17,6 +17,7 @@ interface ProviderOverride {
     apiKey?: string;
     authHeader?: boolean;
     compat?: ModelSpec<Api>["compat"];
+    remoteCompaction?: RemoteCompactionConfig<Api>;
     transport?: Model<Api>["transport"];
 }
 /**
@@ -45,7 +46,7 @@ interface ProviderOverride {
  * See `xiaomi-tp-discovery-merge.test.ts` and the `refresh()` baseUrl-override
  * regression in `model-registry.test.ts`.
  */
-export declare function mergeDiscoveredModel<TApi extends Api>(model: Model<TApi>, existing: Model<Api> | undefined, providerOverride?: Pick<ProviderOverride, "baseUrl" | "headers" | "transport">): Model<TApi>;
+export declare function mergeDiscoveredModel<TApi extends Api>(model: Model<TApi>, existing: Model<Api> | undefined, providerOverride?: Pick<ProviderOverride, "baseUrl" | "headers" | "remoteCompaction" | "transport">): Model<TApi>;
 export type ProviderDiscoveryStatus = "idle" | "ok" | "empty" | "cached" | "unavailable" | "unauthenticated";
 export interface ProviderDiscoveryState {
     provider: string;
@@ -227,6 +228,7 @@ export interface ProviderConfigInput {
     streamSimple?: (model: Model<Api>, context: Context, options?: SimpleStreamOptions) => AssistantMessageEventStream;
     headers?: Record<string, string>;
     compat?: ModelSpec<Api>["compat"];
+    remoteCompaction?: RemoteCompactionConfig<Api>;
     authHeader?: boolean;
     /** Streaming transport override — see {@link Model.transport}. */
     transport?: Model<Api>["transport"];
@@ -264,6 +266,8 @@ export interface ProviderConfigInput {
         headers?: Record<string, string>;
         compat?: ModelSpec<Api>["compat"];
         contextPromotionTarget?: string;
+        compactionModel?: string;
+        remoteCompaction?: RemoteCompactionConfig<Api>;
         premiumMultiplier?: number;
     }>;
 }

package/dist/types/config/model-resolver.d.ts

@@ -181,6 +181,18 @@ export declare function resolveRoleSelection(roles: readonly string[], settings:
     model: Model<Api>;
     thinkingLevel?: ThinkingLevel;
 } | undefined;
+/**
+ * Resolve the model for the `advisor` role. A configured `modelRoles.advisor`
+ * wins outright (a bad override surfaces as no model rather than silently
+ * running something else); when unset it falls back to the `slow` priority
+ * chain via {@link ROLE_PRIORITY_ALIAS} — a strong reasoning model that, unlike
+ * the `slow` role itself, never inherits the primary's model. Returns undefined
+ * only when no candidate in the resolved chain is available.
+ */
+export declare function resolveAdvisorRoleSelection(settings: Settings, availableModels: Model<Api>[], modelRegistry?: CanonicalModelRegistry): {
+    model: Model<Api>;
+    thinkingLevel?: ThinkingLevel;
+} | undefined;
 /**
  * Resolve model patterns to actual Model objects with optional thinking levels
  * Format: "pattern:level" where :level is optional

package/dist/types/config/models-config-schema.d.ts

@@ -239,10 +239,17 @@ export declare const ModelOverrideSchema: import("arktype/internal/variants/obje
         } | undefined;
     } | undefined;
     contextPromotionTarget?: string | undefined;
+    compactionModel?: string | undefined;
+    remoteCompaction?: {
+        enabled?: boolean | undefined;
+        api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+        endpoint?: string | undefined;
+        model?: string | undefined;
+    } | undefined;
 }, {}>;
 export type ModelOverride = typeof ModelOverrideSchema.infer;
 export declare const ProviderDiscoverySchema: import("arktype/internal/variants/object.ts").ObjectType<{
-    type: "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
+    type: "litellm" | "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
 }, {}>;
 export declare const ProviderAuthSchema: import("arktype/internal/variants/string.ts").StringType<"apiKey" | "none" | "oauth", {}>;
 export type ProviderAuthMode = typeof ProviderAuthSchema.infer;
@@ -356,10 +363,16 @@ export declare const ModelsConfigSchema: import("arktype/internal/variants/objec
                     replayUnsignedThinking?: boolean | undefined;
                 } | undefined;
             } | undefined;
+            remoteCompaction?: {
+                enabled?: boolean | undefined;
+                api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                endpoint?: string | undefined;
+                model?: string | undefined;
+            } | undefined;
             authHeader?: boolean | undefined;
             auth?: "apiKey" | "none" | "oauth" | undefined;
             discovery?: {
-                type: "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
+                type: "litellm" | "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
             } | undefined;
             models?: {
                 id: string;
@@ -505,6 +518,13 @@ export declare const ModelsConfigSchema: import("arktype/internal/variants/objec
                     } | undefined;
                 } | undefined;
                 contextPromotionTarget?: string | undefined;
+                compactionModel?: string | undefined;
+                remoteCompaction?: {
+                    enabled?: boolean | undefined;
+                    api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                    endpoint?: string | undefined;
+                    model?: string | undefined;
+                } | undefined;
             }[] | undefined;
             modelOverrides?: {
                 [x: string]: {
@@ -648,6 +668,13 @@ export declare const ModelsConfigSchema: import("arktype/internal/variants/objec
                         } | undefined;
                     } | undefined;
                     contextPromotionTarget?: string | undefined;
+                    compactionModel?: string | undefined;
+                    remoteCompaction?: {
+                        enabled?: boolean | undefined;
+                        api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                        endpoint?: string | undefined;
+                        model?: string | undefined;
+                    } | undefined;
                 };
             } | undefined;
             disableStrictTools?: boolean | undefined;

package/dist/types/config/models-config.d.ts

@@ -21,6 +21,7 @@ export interface ProviderValidationConfig {
     oauthConfigured?: boolean;
     discovery?: ProviderDiscovery;
     compat?: ModelSpec<Api>["compat"];
+    remoteCompaction?: unknown;
     disableStrictTools?: boolean;
     modelOverrides?: Record<string, unknown>;
     models: ProviderValidationModel[];
@@ -135,10 +136,16 @@ export declare const ModelsConfigFile: ConfigFile<{
                     replayUnsignedThinking?: boolean | undefined;
                 } | undefined;
             } | undefined;
+            remoteCompaction?: {
+                enabled?: boolean | undefined;
+                api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                endpoint?: string | undefined;
+                model?: string | undefined;
+            } | undefined;
             authHeader?: boolean | undefined;
             auth?: "apiKey" | "none" | "oauth" | undefined;
             discovery?: {
-                type: "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
+                type: "litellm" | "llama.cpp" | "lm-studio" | "ollama" | "openai-models-list" | "proxy";
             } | undefined;
             models?: {
                 id: string;
@@ -269,6 +276,13 @@ export declare const ModelsConfigFile: ConfigFile<{
                     } | undefined;
                 } | undefined;
                 contextPromotionTarget?: string | undefined;
+                compactionModel?: string | undefined;
+                remoteCompaction?: {
+                    enabled?: boolean | undefined;
+                    api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                    endpoint?: string | undefined;
+                    model?: string | undefined;
+                } | undefined;
             }[] | undefined;
             modelOverrides?: {
                 [x: string]: {
@@ -397,6 +411,13 @@ export declare const ModelsConfigFile: ConfigFile<{
                         } | undefined;
                     } | undefined;
                     contextPromotionTarget?: string | undefined;
+                    compactionModel?: string | undefined;
+                    remoteCompaction?: {
+                        enabled?: boolean | undefined;
+                        api?: "anthropic-messages" | "azure-openai-responses" | "google-gemini-cli" | "google-generative-ai" | "google-vertex" | "openai-codex-responses" | "openai-completions" | "openai-responses" | undefined;
+                        endpoint?: string | undefined;
+                        model?: string | undefined;
+                    } | undefined;
                 };
             } | undefined;
             disableStrictTools?: boolean | undefined;