opencode-matrixx 2.4.0 → 2.6.0

package/dist/create-tools.d.ts

@@ -2,11 +2,11 @@ import type { AvailableCategory, AvailableSkill } from "./agents/dynamic-agent-p
 import type { MatrixxConfig } from "./config";
 import type { BrowserAutomationProvider } from "./config/schema/browser-automation";
 import type { Managers } from "./create-managers";
-import type { LoadedSkill } from "./features/opencode-skill-loader/types";
+import type { BuiltinSkill } from "./features/builtin-skills";
 import type { PluginContext, ToolsRecord } from "./plugin/types";
 type CreateToolsResult = {
     filteredTools: ToolsRecord;
-    mergedSkills: LoadedSkill[];
+    builtinSkills: BuiltinSkill[];
     availableSkills: AvailableSkill[];
     availableCategories: AvailableCategory[];
     browserProvider: BrowserAutomationProvider;
@@ -16,6 +16,6 @@ type CreateToolsResult = {
 export declare function createTools(args: {
     ctx: PluginContext;
     pluginConfig: MatrixxConfig;
-    managers: Pick<Managers, "backgroundManager" | "tmuxSessionManager" | "skillMcpManager">;
+    managers: Pick<Managers, "backgroundManager" | "tmuxSessionManager">;
 }): Promise<CreateToolsResult>;
 export {};

package/dist/features/background-agent/error-helpers.d.ts

@@ -1,6 +1,4 @@
 import type { EventProperties } from "./manager";
-export declare function formatDuration(start: Date, end?: Date): string;
 export declare function getErrorText(error: unknown): string;
 export declare function isAbortedSessionError(error: unknown): boolean;
-export declare function isRecord(value: unknown): value is Record<string, unknown>;
 export declare function getSessionErrorMessage(properties: EventProperties): string | undefined;

package/dist/features/builtin-commands/templates/dcp-profile.d.ts

@@ -1 +1 @@
-export declare const DCP_PROFILE_TEMPLATE = "You are switching the active DCP (Dynamic Context Pruning) profile tier.\n\nThis command is provided by the Matrixx plugin to call a user-managed shell script that swaps the active DCP configuration. The script lives in the user's home directory and is not part of the Matrixx codebase.\n\n## Step 1: Verify DCP is installed\n\nCheck whether the DCP plugin is installed at the standard OpenCode plugin location:\n\n```bash\nif [ ! -d \"$HOME/.config/opencode/node_modules/@tarquinen/opencode-dcp\" ]; then\n  echo \"DCP is not installed at the standard OpenCode plugin location.\" >&2\n  echo \"Install it with: npm install --prefix ~/.config/opencode @tarquinen/opencode-dcp\" >&2\n  exit 1\nfi\n```\n\nIf the directory does not exist, stop immediately and report the error to the user. Do not proceed.\n\n## Step 2: Resolve the switch script path\n\nThe switch script location is configurable via the `dcp.switch_script` key in the user's `matrixx.jsonc` (read from `~/.config/opencode/matrixx.jsonc` or the project-level file). If that key is absent, use the default:\n\n```\n~/.myopencode/dcp/switch-profile.sh\n```\n\nIf the resolved script does not exist or is not executable, stop and report the path to the user.\n\n## Step 3: Determine the target profile\n\nParse the arguments passed to this command. The user invoked `/dcp-profile <arguments>` where `<arguments>` is the first positional argument.\n\n- If the argument is a known profile name (one of: economy, balanced, performance, ultimate), use it directly.\n- If the argument is empty or missing, read `dcp.default_profile` from the user's config; if absent, default to `balanced`.\n- If the argument is not a recognized profile name, list the available profiles and stop. Do NOT guess or pass invalid names to the script.\n\n## Step 4: Execute the switch script\n\nRun the script via the bash tool. Capture both stdout and stderr. If the script exits with a non-zero status, show the captured stderr output to the user verbatim and stop. Do not swallow errors.\n\n```bash\n\"<resolved-switch-script>\" \"<profile-name>\"\n```\n\n## Step 5: Confirm and instruct\n\nAfter a successful switch:\n\n1. Report the script's stdout to the user.\n2. Tell the user that the new DCP configuration will take effect after they restart their OpenCode session (the active session has already loaded the previous config into memory).\n3. Do not attempt to reload DCP in-place; a session restart is required.\n\n## Important constraints\n\n- This command is a thin wrapper. All real logic lives in the user's switch script. Never bypass the script by writing to `dcp.jsonc` directly.\n- Do not install, upgrade, or modify the DCP plugin from this command. If the user needs to install or upgrade DCP, instruct them to run `opencode plugin @tarquinen/opencode-dcp@<version>` (or use `npm install --prefix ~/.config/opencode` for cached installs).\n- Do not edit any DCP config files (`dcp-base.jsonc`, `dcp-<profile>.jsonc`, `dcp-generated-*.jsonc`) as part of switching.";
+export declare const DCP_PROFILE_TEMPLATE = "You are switching the active DCP (Dynamic Context Pruning) profile tier.\n\nThis command is provided by the Matrixx plugin. It uses the built-in `dcp_switch_profile` tool to apply DCP profile configurations \u2014 no external scripts needed.\n\n## Step 1: Verify DCP is installed\n\nCheck whether the DCP plugin is installed at the standard OpenCode plugin location:\n\n```bash\nif [ ! -d \"$HOME/.config/opencode/node_modules/@tarquinen/opencode-dcp\" ]; then\n  echo \"DCP is not installed at the standard OpenCode plugin location.\" >&2\n  echo \"Install it with: npm install --prefix ~/.config/opencode @tarquinen/opencode-dcp\" >&2\n  exit 1\nfi\n```\n\nIf the directory does not exist, stop immediately and report the error to the user. Do not proceed.\n\n## Step 2: Determine the target profile\n\nParse the arguments passed to this command. The user invoked `/dcp-profile <arguments>` where `<arguments>` is the first positional argument.\n\n- If the argument is a known profile name (one of: economy, balanced, performance, ultimate), use it directly.\n- If the argument is empty or missing, read `dcp.default_profile` from the user's `matrixx.jsonc` config; if absent, default to `balanced`.\n- If the argument is not a recognized profile name, list the available profiles and stop. Do NOT guess or pass invalid names.\n\n## Step 3: Call the built-in `dcp_switch_profile` tool\n\nUse the `dcp_switch_profile` tool with the resolved profile name. This tool reads profile parameters from the Matrixx plugin configuration and writes the full inline DCP config to `~/.config/opencode/dcp.jsonc`.\n\nThe tool will handle all file operations \u2014 you do NOT need to run any external scripts or edit DCP config files directly.\n\n## Step 4: Confirm and instruct\n\nAfter a successful switch:\n\n1. Report the tool's output to the user.\n2. Tell the user that the new DCP configuration will take effect after they restart their OpenCode session (the active session has already loaded the previous config into memory).\n3. Do not attempt to reload DCP in-place; a session restart is required.\n\n## Important constraints\n\n- Use the built-in `dcp_switch_profile` tool. Do NOT bypass it by editing DCP config files directly.\n- Do not install, upgrade, or modify the DCP plugin from this command. If the user needs to install or upgrade DCP, instruct them to run `opencode plugin @tarquinen/opencode-dcp@<version>` (or use `npm install --prefix ~/.config/opencode` for cached installs).\n";

package/dist/features/builtin-commands/templates/evolution.d.ts

@@ -0,0 +1 @@
+export declare const EVOLUTION_TEMPLATE = "# Evolution Command\n\n## Purpose\n\nManage self-evolution proposals staged in `.matrixx/evolution/pending`.\n\n## Commands\n\n- `/evolution list` \u2014 list pending proposals\n- `/evolution approve <slug>` \u2014 promote pending skill to `.opencode/skills/<slug>/SKILL.md`\n- `/evolution reject <slug>` \u2014 discard pending proposal\n- `/evolution audit` \u2014 show last 20 audit entries\n\n---\n\n# PHASE 1: PARSE ARGUMENTS\n\nArguments: $ARGUMENTS\n\n- If empty or \"list\": list pending\n- If \"approve <slug>\" or \"approve <slug> --global\": promote\n- If \"reject <slug>\": discard\n- If \"audit\" or \"log\": show audit\n- Otherwise: show help and list pending\n\n---\n\n# PHASE 2: EXECUTE\n\n## List\n\nUse bash: `ls -1 .matrixx/evolution/pending/*.md 2>/dev/null | xargs -I {} basename {} .md` or `rtk ls .matrixx/evolution/pending`\nFor each slug, read `.matrixx/evolution/pending/<slug>.meta.json` to show confidence, version, derived_from.\nIf none: \"No pending evolution proposals.\"\n\n## Approve\n\n1. Verify `.matrixx/evolution/pending/<slug>.md` exists \u2014 if not, error: \"Pending <slug> not found. Run /evolution list to see available.\"\n2. Copy staged skill:\n   - `mkdir -p .opencode/skills/<slug>`\n   - `cp .matrixx/evolution/pending/<slug>.md .opencode/skills/<slug>/SKILL.md`\n   - Also preserve versioned copy: `mkdir -p .matrixx/evolution/skills/<slug>/versions && cp .matrixx/evolution/pending/<slug>.md .matrixx/evolution/skills/<slug>/SKILL.md`\n3. Append audit: `echo '{\"action\":\"promoted\",\"slug\":\"<slug>\",\"timestamp\":\"'$(date -Iseconds)'\"}' >> .matrixx/evolution/audit.log`\n4. Remove pending: `rm .matrixx/evolution/pending/<slug>.md .matrixx/evolution/pending/<slug>.meta.json`\n5. If argument includes --global, also copy to `~/.agents/skills/<slug>/SKILL.md`\n6. Confirm: \"Promoted <slug> to .opencode/skills/<slug>/SKILL.md \u2014 will be loaded on next session start.\"\n\n## Reject\n\n1. Verify pending exists\n2. `rm .matrixx/evolution/pending/<slug>.md .matrixx/evolution/pending/<slug>.meta.json`\n3. Append audit with action rejected\n4. Confirm: \"Rejected <slug>.\"\n\n## Audit\n\nRead `.matrixx/evolution/audit.log` last 20 lines: `tail -n 20 .matrixx/evolution/audit.log 2>/dev/null || echo \"No audit entries.\"`\n\n---\n\n# CONSTRAINTS\n\n- Use bash/rtk for file ops \u2014 no dedicated evolution tool yet\n- Never invent slug \u2014 read from pending dir\n- Keep operations atomic; report errors clearly\n- Do not modify pending content \u2014 promote as-is after human review\n";

package/dist/features/builtin-commands/types.d.ts

@@ -1,3 +1,3 @@
 import type { CommandDefinition } from "../command-loader";
-export type BuiltinCommandName = "init-deep" | "matrix-loop" | "cancel-loop" | "ulw-loop" | "refactor" | "start-work" | "stop-continuation" | "handoff" | "pickup" | "remove-deadcode" | "profile" | "end-ultrawork" | "research" | "assembly" | "ultrawork" | "bdd-backend" | "bdd-contract" | "bdd-frontend" | "bdd-pipeline" | "bdd-tests" | "dcp-profile";
+export type BuiltinCommandName = "init-deep" | "matrix-loop" | "cancel-loop" | "ulw-loop" | "refactor" | "start-work" | "stop-continuation" | "handoff" | "pickup" | "remove-deadcode" | "profile" | "end-ultrawork" | "research" | "assembly" | "ultrawork" | "bdd-backend" | "bdd-contract" | "bdd-frontend" | "bdd-pipeline" | "bdd-tests" | "dcp-profile" | "evolution";
 export type BuiltinCommands = Record<string, CommandDefinition>;

package/dist/features/builtin-skills/types.d.ts

@@ -1,4 +1,13 @@
-import type { SkillMcpConfig } from "../skill-mcp-manager/types";
+export interface McpServerDefinition {
+    type?: "http" | "sse" | "stdio";
+    url?: string;
+    command?: string;
+    args?: string[];
+    env?: Record<string, string>;
+    headers?: Record<string, string>;
+    disabled?: boolean;
+}
+export type SkillMcpConfig = Record<string, McpServerDefinition>;
 export interface BuiltinSkill {
     name: string;
     description: string;

package/dist/features/command-loader/index.d.ts

@@ -1,2 +1,2 @@
-export * from "./loader";
+export * from "./types";
 export * from "./types";

package/dist/features/evolution/compressor/index.d.ts

@@ -0,0 +1,5 @@
+export * from "./interface";
+export * from "./llm";
+import type { EvolutionCompressorConfig } from "../../../config/schema/evolution";
+import type { Compressor } from "./interface";
+export declare function createCompressor(config: EvolutionCompressorConfig, llmCall?: (prompt: string) => Promise<string>): Compressor;

package/dist/features/evolution/compressor/interface.d.ts

@@ -0,0 +1,5 @@
+import type { CompressionInput, DistilledKnowledge } from "../types";
+export interface Compressor {
+    compress(input: CompressionInput): Promise<DistilledKnowledge>;
+}
+export type { CompressionInput, DistilledKnowledge, TraceRecord } from "../types";

package/dist/features/evolution/compressor/llm.d.ts

@@ -0,0 +1,12 @@
+import type { EvolutionCompressorConfig } from "../../../config/schema/evolution";
+import type { CompressionInput, DistilledKnowledge } from "../types";
+import type { Compressor } from "./interface";
+export declare class LlmCompressor implements Compressor {
+    private config;
+    private llmCall?;
+    constructor(options: {
+        config: EvolutionCompressorConfig;
+        llmCall?: (prompt: string) => Promise<string>;
+    });
+    compress(input: CompressionInput): Promise<DistilledKnowledge>;
+}

package/dist/features/evolution/evaluator.d.ts

@@ -0,0 +1,11 @@
+import type { SkillMeta } from "./types";
+export declare function containsSecretForEval(text: string): boolean;
+export type EvalResult = {
+    score: number;
+    demoted: boolean;
+};
+export declare function readMeta(slug: string, projectRoot?: string): Promise<SkillMeta | null>;
+export declare function evaluateSkill(slug: string, opts?: {
+    threshold?: number;
+    projectRoot?: string;
+}): Promise<EvalResult>;

package/dist/features/evolution/index.d.ts

@@ -0,0 +1,7 @@
+export { createCompressor } from "./compressor";
+export { LlmCompressor } from "./compressor/llm";
+export * from "./evaluator";
+export * from "./pipeline";
+export * from "./store";
+export * from "./types";
+export * from "./writer";

package/dist/features/evolution/pipeline.d.ts

@@ -0,0 +1,7 @@
+import type { EvolutionConfig } from "../../config/schema/evolution";
+import type { CompressionInput } from "./types";
+export declare function runEvolutionPipeline(input: CompressionInput, config: EvolutionConfig): Promise<{
+    staged?: string;
+    promoted?: string;
+    reason?: string;
+}>;

package/dist/features/evolution/store.d.ts

@@ -0,0 +1,22 @@
+import type { EvolutionState, TraceRecord } from "./types";
+export declare const EVOLUTION_DIR = ".matrixx/evolution";
+export declare const TRACES_DIR = ".matrixx/evolution/traces";
+export declare const DISTILLED_DIR = ".matrixx/evolution/distilled";
+export declare const SKILLS_DIR = ".matrixx/evolution/skills";
+export declare const PENDING_DIR = ".matrixx/evolution/pending";
+export declare const STATE_FILE = ".matrixx/evolution/state.json";
+export declare const AUDIT_FILE = ".matrixx/evolution/audit.log";
+export declare class TraceStore {
+    private ringBuffer;
+    private evolutionDir;
+    constructor(evolutionDir?: string);
+    append(record: TraceRecord): Promise<void>;
+    getRecent(count?: number): TraceRecord[];
+    readSession(sessionID: string): Promise<TraceRecord[]>;
+    getAllTraces(sessionID: string): Promise<TraceRecord[]>;
+    cleanup(retentionDays: number): Promise<void>;
+    getState(): Promise<EvolutionState>;
+    updateState(patch: Partial<EvolutionState>): Promise<void>;
+    appendAudit(entry: Record<string, unknown>): Promise<void>;
+}
+export declare const traceStore: TraceStore;

package/dist/features/evolution/types.d.ts

@@ -0,0 +1,51 @@
+export type TraceRecord = {
+    id: string;
+    sessionID: string;
+    callID: string;
+    timestamp: string;
+    agent: string;
+    tool: string;
+    args: unknown;
+    output: string;
+    durationMs: number;
+    success: boolean;
+    errorType?: string;
+    model?: string;
+};
+export type DistilledKnowledge = {
+    title: string;
+    summary: string;
+    patterns: string[];
+    pitfalls: string[];
+    prerequisites: string[];
+    skillDraft?: string;
+    confidence: number;
+    sourceSessionIDs: string[];
+};
+export type CompressionInput = {
+    sessionID: string;
+    traces: TraceRecord[];
+    messages?: unknown[];
+    notepads?: string[];
+    handoff?: unknown;
+    taskHistory?: unknown[];
+};
+export interface Compressor {
+    compress(input: CompressionInput): Promise<DistilledKnowledge>;
+}
+export type EvolutionState = {
+    totalTraces: number;
+    totalCompressions: number;
+    lastCompressionAt?: string;
+    lastPromptAt?: string;
+};
+export type SkillMeta = {
+    name: string;
+    version: string;
+    derived_from: string[];
+    created_at: string;
+    confidence: number;
+    eval_score?: number | null;
+    tags?: string[];
+    prerequisites?: string[];
+};

package/dist/features/evolution/writer.d.ts

@@ -0,0 +1,20 @@
+import type { EvolutionWriterConfig } from "../../config/schema/evolution";
+import type { DistilledKnowledge } from "./types";
+export declare class EvolutionWriter {
+    private pendingDir;
+    private skillsDir;
+    private promotedBase;
+    private globalBase?;
+    constructor(config: EvolutionWriterConfig, projectRoot?: string);
+    stage(knowledge: DistilledKnowledge): Promise<{
+        slug: string;
+        pendingPath: string;
+        metaPath: string;
+    }>;
+    promote(slug: string): Promise<{
+        promotedPath: string;
+    }>;
+    reject(slug: string): Promise<void>;
+    listPending(): Promise<string[]>;
+    listPromoted(): Promise<string[]>;
+}

package/dist/features/task-storage/types.d.ts

@@ -1,8 +1,8 @@
 import { z } from "zod";
 export declare const TaskStatusSchema: z.ZodEnum<{
     deleted: "deleted";
-    completed: "completed";
     pending: "pending";
+    completed: "completed";
     in_progress: "in_progress";
 }>;
 export type TaskStatus = z.infer<typeof TaskStatusSchema>;
@@ -12,8 +12,8 @@ export declare const TaskSchema: z.ZodObject<{
     description: z.ZodString;
     status: z.ZodEnum<{
         deleted: "deleted";
-        completed: "completed";
         pending: "pending";
+        completed: "completed";
         in_progress: "in_progress";
     }>;
     activeForm: z.ZodOptional<z.ZodString>;

package/dist/features/task-toast-manager/types.d.ts

@@ -1,9 +1,9 @@
-import type { ModelSource } from "../../shared/model-resolver";
+import type { ModelResolutionProvenance } from "../../shared/model-resolution-pipeline";
 export type TaskStatus = "running" | "queued" | "completed" | "error";
 export interface ModelFallbackInfo {
     model: string;
     type: "user-defined" | "inherited" | "category-default" | "system-default";
-    source?: ModelSource;
+    source?: ModelResolutionProvenance;
 }
 export interface TrackedTask {
     id: string;

package/dist/hooks/anthropic-context-window-limit-recovery/message-builder.d.ts

@@ -2,6 +2,5 @@ import type { PluginInput } from "@opencode-ai/plugin";
 export declare const PLACEHOLDER_TEXT = "[user interrupted]";
 type OpencodeClient = PluginInput["client"];
 export declare function sanitizeEmptyMessagesBeforeSummarize(sessionID: string, client?: OpencodeClient): Promise<number>;
-export declare function formatBytes(bytes: number): string;
 export declare function getLastAssistant(sessionID: string, client: OpencodeClient, directory: string): Promise<Record<string, unknown> | null>;
 export {};

package/dist/hooks/anthropic-context-window-limit-recovery/tool-part-types.d.ts

@@ -1,3 +1,4 @@
+import type { TaskStatus } from "../../shared/status-types";
 export interface StoredToolPart {
     id: string;
     sessionID: string;
@@ -6,7 +7,7 @@ export interface StoredToolPart {
     callID: string;
     tool: string;
     state: {
-        status: "pending" | "running" | "completed" | "error";
+        status: TaskStatus;
         input: Record<string, unknown>;
         output?: string;
         error?: string;

package/dist/hooks/auto-slash-command/executor.d.ts

@@ -1,7 +1,10 @@
-import { type LoadedSkill } from "../../features/opencode-skill-loader";
+import type { createOpencodeClient } from "@opencode-ai/sdk";
+import type { BuiltinSkill } from "../../features/builtin-skills";
 import type { ParsedSlashCommand } from "./types";
 export interface ExecutorOptions {
-    skills?: LoadedSkill[];
+    skills?: BuiltinSkill[];
+    /** OpenCode SDK client for discovering plugin-registered commands */
+    client?: ReturnType<typeof createOpencodeClient>;
 }
 interface ExecuteResult {
     success: boolean;

package/dist/hooks/auto-slash-command/hook.d.ts

@@ -1,7 +1,9 @@
-import type { LoadedSkill } from "../../features/opencode-skill-loader";
+import type { BuiltinSkill } from "../../features/builtin-skills";
 import type { AutoSlashCommandHookInput, AutoSlashCommandHookOutput, CommandExecuteBeforeInput, CommandExecuteBeforeOutput } from "./types";
 export interface AutoSlashCommandHookOptions {
-    skills?: LoadedSkill[];
+    skills?: BuiltinSkill[];
+    /** OpenCode SDK client for discovering plugin-registered commands */
+    client?: ReturnType<typeof import("@opencode-ai/sdk").createOpencodeClient>;
 }
 export declare function createAutoSlashCommandHook(options?: AutoSlashCommandHookOptions): {
     "chat.message": (input: AutoSlashCommandHookInput, output: AutoSlashCommandHookOutput) => Promise<void>;

package/dist/hooks/evolution-compressor/index.d.ts

@@ -0,0 +1,15 @@
+import type { EvolutionConfig } from "../../config/schema/evolution";
+import type { PluginContext } from "../../plugin/types";
+export declare function createEvolutionCompressorHook(a?: PluginContext | EvolutionConfig, b?: EvolutionConfig): {
+    event: (input: {
+        event: {
+            type: string;
+            properties?: Record<string, unknown>;
+        };
+    }) => Promise<void>;
+    "experimental.session.compacting": (input: {
+        sessionID: string;
+    }, _output: {
+        context: string[];
+    }) => Promise<void>;
+};

package/dist/hooks/evolution-hitl/index.d.ts

@@ -0,0 +1,12 @@
+import type { Message, Part } from "@opencode-ai/sdk";
+type MessageWithParts = {
+    info: Message;
+    parts: Part[];
+};
+type TransformOutput = {
+    messages: MessageWithParts[];
+};
+export declare function createEvolutionHitlHook(a?: unknown, b?: unknown): {
+    "experimental.chat.messages.transform": (_input: Record<string, never>, output: TransformOutput) => Promise<void>;
+};
+export {};

package/dist/hooks/evolution-quality-gate/index.d.ts

@@ -0,0 +1,6 @@
+import type { EvolutionGovernanceConfig } from "../../config/schema/evolution";
+import type { DistilledKnowledge } from "../../features/evolution/types";
+export declare function passesQualityGate(knowledge: DistilledKnowledge, config: EvolutionGovernanceConfig): {
+    pass: boolean;
+    reason?: string;
+};

package/dist/hooks/evolution-watcher/index.d.ts

@@ -0,0 +1,21 @@
+import type { EvolutionConfig } from "../../config/schema/evolution";
+import type { PluginContext } from "../../plugin/types";
+export { classifySuccess, truncate } from "./utils";
+export declare function createEvolutionWatcherHook(a?: PluginContext | EvolutionConfig, b?: EvolutionConfig): {
+    "tool.execute.before": (input: {
+        tool: string;
+        sessionID: string;
+        callID: string;
+    }, output: {
+        args: Record<string, unknown>;
+    }) => Promise<void>;
+    "tool.execute.after": (input: {
+        tool: string;
+        sessionID: string;
+        callID: string;
+    }, output: {
+        title: string;
+        output: string;
+        metadata: Record<string, unknown>;
+    }) => Promise<void>;
+};

package/dist/hooks/evolution-watcher/utils.d.ts

@@ -0,0 +1,5 @@
+export declare function truncate(value: string, maxChars: number): string;
+export declare function classifySuccess(tool: string, output: string): {
+    success: boolean;
+    errorType?: string;
+};

package/dist/hooks/index.d.ts

@@ -18,17 +18,22 @@ export { createEditErrorRecoveryHook } from "./edit-error-recovery";
 export { createEmptyTaskResponseDetectorHook } from "./empty-task-response-detector";
 export { createEnvContextInjectorHook } from "./env-context-injector";
 export { createEnvFileWriteGuardHook } from "./env-file-write-guard";
+export { createEvolutionCompressorHook } from "./evolution-compressor";
+export { createEvolutionHitlHook } from "./evolution-hitl";
+export { passesQualityGate } from "./evolution-quality-gate";
+export { createEvolutionWatcherHook } from "./evolution-watcher";
 export { createHashlineEditDiffEnhancerHook } from "./hashline-edit-diff-enhancer";
 export { createHashlineReadEnhancerHook } from "./hashline-read-enhancer";
 export { createInteractiveBashSessionHook } from "./interactive-bash-session";
 export { createJsonErrorRecoveryHook } from "./json-error-recovery";
 export { createKeywordDetectorHook } from "./keyword-detector";
 export { createMatrixLoopHook, type MatrixLoopHook } from "./matrix-loop";
+export { createMcpStartupNotificationHook } from "./mcp-startup-notification";
 export { createMouseNotepadHook } from "./mouse-notepad";
 export { createNonInteractiveEnvHook } from "./non-interactive-env";
+export { createOracleMdOnlyHook } from "./oracle-md-only";
 export { createPlanPersister } from "./plan-persister";
 export { createPreemptiveCompactionHook } from "./preemptive-compaction";
-export { createOracleMdOnlyHook } from "./prometheus-md-only";
 export { createQualityGateHook } from "./quality-gate/hook";
 export { createQuestionLabelTruncatorHook } from "./question-label-truncator";
 export { createReadImageResizerHook } from "./read-image-resizer";
@@ -46,6 +51,7 @@ export { createStartWorkHook } from "./start-work";
 export { createStopContinuationGuardHook, type StopContinuationGuard } from "./stop-continuation-guard";
 export { createTaskNotepadHook } from "./task-notepad";
 export { createTaskResumeInfoHook } from "./task-resume-info";
+export { createTaskTodoMirrorHook } from "./task-todo-mirror";
 export { createTasksTodowriteDisablerHook } from "./tasks-todowrite-disabler";
 export { createThinkModeHook } from "./think-mode";
 export { createThinkingBlockValidatorHook } from "./thinking-block-validator";

package/dist/hooks/keyword-detector/analyze/default.d.ts

@@ -9,4 +9,4 @@
  * - Vietnamese: phân tích, điều tra, nghiên cứu, kiểm tra, xem xét, chẩn đoán, giải thích, tìm hiểu, gỡ lỗi, tại sao
  */
 export declare const ANALYZE_PATTERN: RegExp;
-export declare const ANALYZE_MESSAGE = "[analyze-mode]\nANALYSIS MODE. Gather context before diving deep:\n\nCONTEXT GATHERING (parallel):\n- 1-2 explore agents (codebase patterns, implementations)\n- 1-2 librarian agents (if external library involved)\n- Direct tools: Grep, AST-grep, LSP for targeted searches\n\nIF COMPLEX - DO NOT STRUGGLE ALONE. Consult specialists:\n- **Oracle**: Conventional problems (architecture, debugging, complex logic)\n- **Artistry**: Non-conventional problems (different approach needed)\n\nSYNTHESIZE findings before proceeding.";
+export declare const ANALYZE_MESSAGE = "[analyze-mode]\nANALYSIS MODE. Gather context before diving deep:\n\nCONTEXT GATHERING (parallel):\n- 1-2 trinity agents (codebase patterns, implementations)\n- 1-2 operator agents (if external library involved)\n- Direct tools: Grep, AST-grep, LSP for targeted searches\n\nIF COMPLEX - DO NOT STRUGGLE ALONE. Consult specialists:\n- **Oracle**: Conventional problems (architecture, debugging, complex logic)\n- **Matrix-bend**: Non-conventional problems (different approach needed)\n\nSYNTHESIZE findings before proceeding.";

package/dist/hooks/keyword-detector/constants.d.ts

@@ -1,4 +1,4 @@
-export { CODE_BLOCK_PATTERN, INLINE_CODE_PATTERN } from "../../shared";
+export { CODE_BLOCK_PATTERN, INLINE_CODE_PATTERN } from "../../shared/code-patterns";
 export { ANALYZE_MESSAGE, ANALYZE_PATTERN } from "./analyze";
 export { SEARCH_MESSAGE, SEARCH_PATTERN } from "./search";
 export { getUltraworkMessage, isPlannerAgent } from "./ultrawork";

package/dist/hooks/keyword-detector/search/default.d.ts

@@ -9,4 +9,4 @@
  * - Vietnamese: tìm kiếm, tra cứu, định vị, quét, phát hiện, truy tìm, tìm ra, ở đâu, liệt kê
  */
 export declare const SEARCH_PATTERN: RegExp;
-export declare const SEARCH_MESSAGE = "[search-mode]\nMAXIMIZE SEARCH EFFORT. Launch multiple background agents IN PARALLEL:\n- explore agents (codebase patterns, file structures, ast-grep)\n- librarian agents (remote repos, official docs, GitHub examples)\nPlus direct tools: Grep, ripgrep (rg), ast-grep (sg)\nNEVER stop at first result - be exhaustive.";
+export declare const SEARCH_MESSAGE = "[search-mode]\nMAXIMIZE SEARCH EFFORT. Launch multiple background agents IN PARALLEL:\n- trinity agents (codebase patterns, file structures, ast-grep)\n- operator agents (remote repos, official docs, GitHub examples)\nPlus direct tools: Grep, ripgrep (rg), ast-grep (sg)\nNEVER stop at first result - be exhaustive.";

package/dist/hooks/keyword-detector/ultrawork/deepseek.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Ultrawork message optimized for DeepSeek V4 Flash models.
+ *
+ * Key characteristics (from DeepSeek V4 Flash research):
+ * - Thinking/reasoning mode (enabled by default, explicitly disable when not needed)
+ * - Strong on code generation, simple agent tasks, long-context understanding
+ * - Responds well to structured prompts with XML-style section boundaries
+ * - Task sandwich pattern (task before AND after long context)
+ * - Output anchors to reduce preamble drift
+ * - Self-check instructions to catch ~70% of errors
+ * - Temperature 0.0 for deterministic code output
+ * - 1M token context window
+ * - Preserve reasoning_content in tool-call assistant messages across turns
+ */
+export declare const ULTRAWORK_DEEPSEEK_MESSAGE = "<ultrawork-mode>\n\n**MANDATORY**: You MUST say \"ULTRAWORK MODE ENABLED!\" to the user as your first response when this mode activates. This is non-negotiable.\n\n<role>\n  You are a senior engineering agent. Ship verified work. No process narration.\n</role>\n\n<thinking_mode>\n  Thinking mode is ON by default on DeepSeek V4 Flash. For trivial tasks (single-file edit, typo fix, simple lookup), explicitly request thinking OFF. For complex tasks (architecture, multi-file, debugging, planning), keep thinking ON at high effort. When thinking is enabled, temperature/penalty parameters are ignored \u2014 tune the prompt instead. Never strip reasoning_content from assistant messages that contain tool_calls.\n</thinking_mode>\n\n<certainty_protocol>\n  ## Absolute Certainty Required\n  You MUST NOT start implementation until you are 100% certain.\n\n  Before you write code:\n  - Fully understand the user's actual intent\n  - Explore the codebase to understand patterns and architecture\n  - Have a clear work plan\n  - Resolve ambiguities through exploration, not guessing\n\n  When uncertain:\n  1. Fire trinity agents for codebase exploration (run_in_background=true)\n  2. Fire operator agents for external research (run_in_background=true)\n  3. Consult oracle for architecture/debugging after 2+ attempts\n  4. Only ask the user as last resort\n\n  Signs you are NOT ready: making assumptions, unsure which files, plan has \"maybe\", can't explain exact steps.\n</certainty_protocol>\n\n<task>\n  Deliver EXACTLY what the user asked, end-to-end working, with captured evidence: a failing-first proof that went RED to GREEN, plus real-surface proof sized by the tier below. Tests alone never prove done.\n</task>\n\n<quality_tiers>\n  LIGHT: Known pattern, no open design decisions (bugfix following existing pattern, query tweak, copy/constants). Plan directly in notepad. 1-2 success criteria. One real-surface proof. Self-review.\n\n  HEAVY: New module/layer/abstraction, auth/security, external integration, DB schema, concurrency, cross-boundary refactor, or user signals care. 3+ success criteria (happy, edge, regression). Reviewer loop until approval. Full evidence gates.\n</quality_tiers>\n\n<delegation_framework>\n  ## Agents / Categories + Skills\n\n  DEFAULT: Delegate. Do not work yourself.\n\n  | Task | Action |\n  |------|--------|\n  | Codebase exploration | task(subagent_type=\"trinity\", load_skills=[], run_in_background=true) |\n  | Documentation lookup | task(subagent_type=\"operator\", load_skills=[], run_in_background=true) |\n  | Planning (2+ steps) | task(subagent_type=\"plan\", load_skills=[]) |\n  | Hard problem (conventional) | task(subagent_type=\"oracle\", load_skills=[]) |\n  | Hard problem (non-conventional) | task(category=\"matrix-bend\", load_skills=[...]) |\n  | Implementation | task(category=\"...\", load_skills=[...]) |\n\n  Do it yourself only when: trivial (<10 lines), you have full context, delegation overhead exceeds task complexity.\n</delegation_framework>\n\n<plan_agent_rule>\n  ## Plan Agent Invocation (Non-Negotiable)\n\n  Size the scope first. Count distinct surfaces, files, steps. If 2+ steps, unclear scope, implementation required, or architecture decision needed: MUST call plan agent.\n\n  After plan returns: execute in EXACT wave order and parallel grouping it specifies. Run verification IT defines per task.\n</plan_agent_rule>\n\n<verification_guarantee>\n  ## Verification Guarantee\n\n  Nothing is done without proof.\n\n  ### Goal Registration (BINDING)\n  Register the goal with todowrite BEFORE any implementation: objective, scenario contract, and WHEN TO STOP line.\n\n  ### Scenario Contract (BINDING)\n  Define 3+ scenarios before coding: happy path, edge (boundary/empty/malformed/concurrent), adjacent-surface regression. Each has a binary pass condition, real surface proof, and test id.\n\n  ### Acceptance Criteria + QA\n  Output an acceptance criteria block before any code. Each criterion: binary PASS/FAIL, verifiable via command. Run every verification command. Report results. Fix failures, re-run all.\n\n  | Evidence Gate | Required |\n  |---|---|\n  | RED | Failing assertion before production code |\n  | GREEN | Same test passing |\n  | Surface | CLI/curl/browser artifact |\n  | Build | Exit code 0 |\n  | Suite | All green, no skip/.only/xfail |\n  | Lint | lsp_diagnostics clean |\n\n  **NO EVIDENCE = NOT VERIFIED = NOT DONE.**\n\n  ### Durable Notepad\n  Create a notepad file with sections: Plan, Scenarios, Now, Todo, Findings (file:line), Learnings. Append only. If context is lost, re-read and resume.\n\n  ### TDD Workflow (Mandatory)\n  Every production change follows RED \u2192 GREEN \u2192 SURFACE \u2192 REFACTOR \u2192 REGRESSION. Write failing test FIRST. Capture RED. Write smallest change to flip GREEN. Exercise real surface. Refactor if needed. Re-run full scenario list.\n\n  ### Commit Discipline\n  One atomic commit per verified increment. Before composing, read git log and match conventions.\n\n  ### Reviewer Gate\n  Trigger when: user demands review, 3+ files, 20+ turns, 30+ minutes, refactor/migration/perf/security. Spawn reviewer via task with goal + scenarios + evidence + diff.\n</verification_guarantee>\n\n<execution_rules>\n  ## Execution Rules\n  - TODO format: path: <action> for <scenario> \u2014 verify by <check>\n  - Mark in_progress/completed INSTANTLY. Never batch.\n  - Parallel independent agents. Never parallelise RED and GREEN of same scenario.\n  - Background first: 10+ concurrent agents if needed.\n  - Verify after every increment. Re-read request before final answer.\n</execution_rules>\n\n<output_discipline>\n  ## Output Discipline\n  - First line literally: \"ULTRAWORK MODE ENABLED!\"\n  - During execution: surface only state changes and evidence.\n  - Final message: outcome + criteria checklist with evidence refs + notepad path.\n  - No file-by-file changelog unless asked.\n  - Lead with the result, then the evidence, then remaining blockers.\n</output_discipline>\n\n<stop_rules>\n  ## Stop Rules\n  - After each result, ask: can the user's request be answered now with evidence? If yes, answer now.\n  - STOP GOAL: every scenario PASSES, evidence captured, cleanup done, reviewer approved. Above all: is the user's problem ACTUALLY SOLVED? If yes, deliver and stop.\n  - After 2 identical failed attempts at one step, surface and ask user.\n  - After 2 exploration waves with no new facts, stop exploring.\n</stop_rules>\n\n<zero_tolerance>\n  ## Zero Tolerance Failures\n  - No scope reduction\n  - No mock implementations\n  - No partial completion\n  - No unverified success claims\n  - No deleted/skipped failing tests\n  - No fabricated evidence\n</zero_tolerance>\n\n</ultrawork-mode>\n\n---\n\n";
+export declare function getDeepseekUltraworkMessage(): string;

package/dist/hooks/keyword-detector/ultrawork/default.d.ts

@@ -4,7 +4,10 @@
  * Key characteristics:
  * - Natural tool-like usage of explore/librarian agents (run_in_background=true)
  * - Parallel execution emphasized - fire agents and continue working
- * - Simple workflow: EXPLORES → GATHER → PLAN → DELEGATE
+ * - Survey skills first methodology
+ * - Goal registration, scenario contracts, durable notepad
+ * - TDD workflow with RED→GREEN→SURFACE→REFACTOR→REGRESSION
+ * - Manual QA mandate with cleanup receipts
  */
-export declare const ULTRAWORK_DEFAULT_MESSAGE = "<ultrawork-mode>\n\n**MANDATORY**: You MUST say \"ULTRAWORK MODE ENABLED!\" to the user as your first response when this mode activates. This is non-negotiable.\n\n[CODE RED] Maximum precision required. Ultrathink before acting.\n\n## **ABSOLUTE CERTAINTY REQUIRED - DO NOT SKIP THIS**\n\n**YOU MUST NOT START ANY IMPLEMENTATION UNTIL YOU ARE 100% CERTAIN.**\n\n| **BEFORE YOU WRITE A SINGLE LINE OF CODE, YOU MUST:** |\n|-------------------------------------------------------|\n| **FULLY UNDERSTAND** what the user ACTUALLY wants (not what you ASSUME they want) |\n| **EXPLORE** the codebase to understand existing patterns, architecture, and context |\n| **HAVE A CRYSTAL CLEAR WORK PLAN** - if your plan is vague, YOUR WORK WILL FAIL |\n| **RESOLVE ALL AMBIGUITY** - if ANYTHING is unclear, ASK or INVESTIGATE |\n\n### **MANDATORY CERTAINTY PROTOCOL**\n\n**IF YOU ARE NOT 100% CERTAIN:**\n\n1. **THINK DEEPLY** - What is the user's TRUE intent? What problem are they REALLY trying to solve?\n2. **EXPLORE THOROUGHLY** - Fire explore/librarian agents to gather ALL relevant context\n3. **CONSULT SPECIALISTS** - For hard/complex tasks, DO NOT struggle alone. Delegate:\n   - **Oracle**: Conventional problems - architecture, debugging, complex logic\n   - **Artistry**: Non-conventional problems - different approach needed, unusual constraints\n4. **ASK THE USER** - If ambiguity remains after exploration, ASK. Don't guess.\n\n**SIGNS YOU ARE NOT READY TO IMPLEMENT:**\n- You're making assumptions about requirements\n- You're unsure which files to modify\n- You don't understand how existing code works\n- Your plan has \"probably\" or \"maybe\" in it\n- You can't explain the exact steps you'll take\n\n**WHEN IN DOUBT:**\n```\ntask(subagent_type=\"trinity\", load_skills=[], prompt=\"I'm implementing [TASK DESCRIPTION] and need to understand [SPECIFIC KNOWLEDGE GAP]. Find [X] patterns in the codebase \u2014 show file paths, implementation approach, and conventions used. I'll use this to [HOW RESULTS WILL BE USED]. Focus on src/ directories, skip test files unless test patterns are specifically needed. Return concrete file paths with brief descriptions of what each file does.\", run_in_background=true)\ntask(subagent_type=\"operator\", load_skills=[], prompt=\"I'm working with [LIBRARY/TECHNOLOGY] and need [SPECIFIC INFORMATION]. Find official documentation and production-quality examples for [Y] \u2014 specifically: API reference, configuration options, recommended patterns, and common pitfalls. Skip beginner tutorials. I'll use this to [DECISION THIS WILL INFORM].\", run_in_background=true)\ntask(subagent_type=\"oracle\", load_skills=[], prompt=\"I need architectural review of my approach to [TASK]. Here's my plan: [DESCRIBE PLAN WITH SPECIFIC FILES AND CHANGES]. My concerns are: [LIST SPECIFIC UNCERTAINTIES]. Please evaluate: correctness of approach, potential issues I'm missing, and whether a better alternative exists.\", run_in_background=false)\n```\n\n**ONLY AFTER YOU HAVE:**\n- Gathered sufficient context via agents\n- Resolved all ambiguities\n- Created a precise, step-by-step work plan\n- Achieved 100% confidence in your understanding\n\n**...THEN AND ONLY THEN MAY YOU BEGIN IMPLEMENTATION.**\n\n---\n\n## **NO EXCUSES. NO COMPROMISES. DELIVER WHAT WAS ASKED.**\n\n**THE USER'S ORIGINAL REQUEST IS SACRED. YOU MUST FULFILL IT EXACTLY.**\n\n| VIOLATION | CONSEQUENCE |\n|-----------|-------------|\n| \"I couldn't because...\" | **UNACCEPTABLE.** Find a way or ask for help. |\n| \"This is a simplified version...\" | **UNACCEPTABLE.** Deliver the FULL implementation. |\n| \"You can extend this later...\" | **UNACCEPTABLE.** Finish it NOW. |\n| \"Due to limitations...\" | **UNACCEPTABLE.** Use agents, tools, whatever it takes. |\n| \"I made some assumptions...\" | **UNACCEPTABLE.** You should have asked FIRST. |\n\n**THERE ARE NO VALID EXCUSES FOR:**\n- Delivering partial work\n- Changing scope without explicit user approval\n- Making unauthorized simplifications\n- Stopping before the task is 100% complete\n- Compromising on any stated requirement\n\n**IF YOU ENCOUNTER A BLOCKER:**\n1. **DO NOT** give up\n2. **DO NOT** deliver a compromised version\n3. **DO** consult specialists (oracle for conventional, matrix-bend for non-conventional)\n4. **DO** ask the user for guidance\n5. **DO** explore alternative approaches\n\n**THE USER ASKED FOR X. DELIVER EXACTLY X. PERIOD.**\n\n---\n\nYOU MUST LEVERAGE ALL AVAILABLE AGENTS / **CATEGORY + SKILLS** TO THEIR FULLEST POTENTIAL.\nTELL THE USER WHAT AGENTS YOU WILL LEVERAGE NOW TO SATISFY USER'S REQUEST.\n\n## MANDATORY: PLAN AGENT INVOCATION (NON-NEGOTIABLE)\n\n**YOU MUST ALWAYS INVOKE THE PLAN AGENT FOR ANY NON-TRIVIAL TASK.**\n\n| Condition | Action |\n|-----------|--------|\n| Task has 2+ steps | MUST call plan agent |\n| Task scope unclear | MUST call plan agent |\n| Implementation required | MUST call plan agent |\n| Architecture decision needed | MUST call plan agent |\n\n```\ntask(subagent_type=\"plan\", load_skills=[], prompt=\"<gathered context + user request>\")\n```\n\n**WHY PLAN AGENT IS MANDATORY:**\n- Plan agent analyzes dependencies and parallel execution opportunities\n- Plan agent outputs a **parallel task graph** with waves and dependencies\n- Plan agent provides structured TODO list with category + skills per task\n- YOU are an orchestrator, NOT an implementer\n\n### SESSION CONTINUITY WITH PLAN AGENT (CRITICAL)\n\n**Plan agent returns a session_id. USE IT for follow-up interactions.**\n\n| Scenario | Action |\n|----------|--------|\n| Plan agent asks clarifying questions | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"<your answer>\")` |\n| Need to refine the plan | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"Please adjust: <feedback>\")` |\n| Plan needs more detail | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"Add more detail to Task N\")` |\n\n**WHY SESSION_ID IS CRITICAL:**\n- Plan agent retains FULL conversation context\n- No repeated exploration or context gathering\n- Saves 70%+ tokens on follow-ups\n- Maintains interview continuity until plan is finalized\n\n```\n// WRONG: Starting fresh loses all context\ntask(subagent_type=\"plan\", load_skills=[], prompt=\"Here's more info...\")\n\n// CORRECT: Resume preserves everything\ntask(session_id=\"ses_abc123\", load_skills=[], prompt=\"Here's my answer to your question: ...\")\n```\n\n**FAILURE TO CALL PLAN AGENT = INCOMPLETE WORK.**\n\n---\n\n## AGENTS / **CATEGORY + SKILLS** UTILIZATION PRINCIPLES\n\n**DEFAULT BEHAVIOR: DELEGATE. DO NOT WORK YOURSELF.**\n\n| Task Type | Action | Why |\n|-----------|--------|-----|\n| Codebase exploration | task(subagent_type=\"trinity\", load_skills=[], run_in_background=true) | Parallel, context-efficient |\n| Documentation lookup | task(subagent_type=\"operator\", load_skills=[], run_in_background=true) | Specialized knowledge |\n| Planning | task(subagent_type=\"plan\", load_skills=[]) | Parallel task graph + structured TODO list |\n| Hard problem (conventional) | task(subagent_type=\"oracle\", load_skills=[]) | Architecture, debugging, complex logic |\n| Hard problem (non-conventional) | task(category=\"matrix-bend\", load_skills=[...]) | Different approach needed |\n| Implementation | task(category=\"...\", load_skills=[...]) | Domain-optimized models |\n\n**CATEGORY + SKILL DELEGATION:**\n```\n// Frontend work\ntask(category=\"construct\", load_skills=[\"frontend-ui-ux\"])\n\n// Complex logic\ntask(category=\"source\", load_skills=[\"typescript-programmer\"])\n\n// Quick fixes\ntask(category=\"bullet-time\", load_skills=[\"git-master\"])\n```\n\n**YOU SHOULD ONLY DO IT YOURSELF WHEN:**\n- Task is trivially simple (1-2 lines, obvious change)\n- You have ALL context already loaded\n- Delegation overhead exceeds task complexity\n\n**OTHERWISE: DELEGATE. ALWAYS.**\n\n---\n\n## MANDATORY: ACCEPTANCE CRITERIA DEFINITION (NON-NEGOTIABLE)\n\n**BEFORE writing ANY code, you MUST output an Acceptance Criteria block.**\n\nThis is NOT optional. Implementation without defined acceptance criteria = REJECTED.\n\n### Required Format:\n```\n## Acceptance Criteria\n1. [CRITERION]: [Observable, binary pass/fail condition]\n2. [CRITERION]: [Observable, binary pass/fail condition]\n...\n### Verification Commands:\n- [Exact command to run] \u2192 [Expected output]\n- [Exact command to run] \u2192 [Expected output]\n```\n\n### Rules:\n- Each criterion MUST be binary (PASS or FAIL \u2014 no \"mostly works\")\n- Each criterion MUST be verifiable via a specific command or observable behavior\n- Minimum 3 criteria for any non-trivial task\n- Criteria MUST cover: functional correctness, no regressions, code quality (typecheck/lint)\n- Include verification commands that will be executed during QA\n\n### Example:\n```\n## Acceptance Criteria\n1. User registration endpoint returns 201 on valid input\n2. Duplicate email returns 409 with error message\n3. All existing tests pass (bun test)\n4. TypeScript typecheck passes (bun run typecheck)\n5. No new lint errors introduced\n### Verification Commands:\n- bun test \u2192 All tests pass, 0 failures\n- bun run typecheck \u2192 Exit code 0, no errors\n- curl -X POST /api/register ... \u2192 201 Created\n```\n\n**FAILURE TO OUTPUT ACCEPTANCE CRITERIA = YOU MUST STOP AND DEFINE THEM.**\n\n---\n\n## EXECUTION RULES\n- **TODO**: Track EVERY step. Mark complete IMMEDIATELY after each.\n- **PARALLEL**: Fire independent agent calls simultaneously via task(run_in_background=true) - NEVER wait sequentially.\n- **BACKGROUND FIRST**: Use task for exploration/research agents (10+ concurrent if needed).\n- **VERIFY**: Re-read request after completion. Check ALL requirements met before reporting done.\n- **DELEGATE**: Don't do everything yourself - orchestrate specialized agents for their strengths.\n\n## WORKFLOW\n1. Analyze the request and identify required capabilities\n2. Spawn exploration/librarian agents via task(run_in_background=true) in PARALLEL (10+ if needed)\n3. Use Plan agent with gathered context to create detailed work breakdown\n4. Execute with continuous verification against original requirements\n\n---\n\n## MANDATORY: QA EXECUTION (NON-NEGOTIABLE)\n\n**AFTER implementation, you MUST execute ALL verification commands from your Acceptance Criteria.**\n\nThis is NOT optional. Claiming completion without QA execution = REJECTED.\n\n### QA Protocol:\n1. **Run every verification command** listed in your Acceptance Criteria\n2. **Report results** for each criterion: \u2705 PASS or \u274C FAIL\n3. **If ANY criterion fails**: Fix the issue, re-run ALL verification commands\n4. **Output a QA Report** in this exact format:\n\n```\n## QA Report\n| # | Criterion | Result | Evidence |\n|---|-----------|--------|----------|\n| 1 | [criterion] | \u2705 PASS | [what you observed] |\n| 2 | [criterion] | \u274C FAIL | [error output] |\n| 3 | [criterion] | \u2705 PASS | [what you observed] |\n\n**Overall: [X/Y PASS]** \u2014 [ACCEPTED if all pass / NEEDS FIX if any fail]\n```\n\n### Rules:\n- You MUST actually RUN the commands \u2014 not just say \"it should work\"\n- You MUST show evidence (command output, test results)\n- If ANY criterion fails, you MUST fix and re-run ALL criteria\n- You MUST NOT report completion until ALL criteria pass\n- If you cannot achieve a criterion, explain WHY and propose an alternative\n\n**NO EVIDENCE = NOT VERIFIED = NOT DONE.**\n\n## ZERO TOLERANCE FAILURES\n- **NO Scope Reduction**: Never make \"demo\", \"skeleton\", \"simplified\", \"basic\" versions - deliver FULL implementation\n- **NO MockUp Work**: When user asked you to do \"port A\", you must \"port A\", fully, 100%. No Extra feature, No reduced feature, no mock data, fully working 100% port.\n- **NO Partial Completion**: Never stop at 60-80% saying \"you can extend this...\" - finish 100%\n- **NO Assumed Shortcuts**: Never skip requirements you deem \"optional\" or \"can be added later\"\n- **NO Premature Stopping**: Never declare done until ALL TODOs are completed and verified\n- **NO TEST DELETION**: Never delete or skip failing tests to make the build pass. Fix the code, not the tests.\n\nTHE USER ASKED FOR X. DELIVER EXACTLY X. NOT A SUBSET. NOT A DEMO. NOT A STARTING POINT.\n\n1. EXPLORES + LIBRARIANS\n2. GATHER -> PLAN AGENT SPAWN\n3. WORK BY DELEGATING TO ANOTHER AGENTS\n\nNOW.\n\n</ultrawork-mode>\n\n---\n\n";
+export declare const ULTRAWORK_DEFAULT_MESSAGE: string;
 export declare function getDefaultUltraworkMessage(): string;

package/dist/hooks/keyword-detector/ultrawork/gemini.d.ts

@@ -0,0 +1,12 @@
+/**
+ * Ultrawork message optimized for Gemini series models.
+ *
+ * Key characteristics:
+ * - Strong intent classification gate (Gemini models benefit from explicit classification)
+ * - Survey skills first methodology
+ * - Goal registration, scenario contracts, durable notepad
+ * - TDD workflow with RED→GREEN→SURFACE→REFACTOR→REGRESSION
+ * - Manual QA mandate with cleanup receipts
+ */
+export declare const ULTRAWORK_GEMINI_MESSAGE = "<ultrawork-mode>\n\n**MANDATORY**: You MUST say \"ULTRAWORK MODE ENABLED!\" to the user as your first response when this mode activates. This is non-negotiable.\n\n[CODE RED] Maximum precision required. Ultrathink before acting.\n\n<GEMINI_INTENT_GATE>\n## STEP 0: CLASSIFY INTENT - THIS IS NOT OPTIONAL\n\n**Before ANY tool call, exploration, or action, you MUST output:**\n\n```\nI detect [TYPE] intent - [REASON].\nMy approach: [ROUTING DECISION].\n```\n\nWhere TYPE is one of: research | implementation | investigation | evaluation | fix | open-ended\n\n**SELF-CHECK (answer each before proceeding):**\n\n1. Did the user EXPLICITLY ask me to build/create/implement something? \u2192 If NO, do NOT implement.\n2. Did the user say \"look into\", \"check\", \"investigate\", \"explain\"? \u2192 RESEARCH only. Do not code.\n3. Did the user ask \"what do you think?\" \u2192 EVALUATE and propose. Do NOT execute.\n4. Did the user report an error/bug? \u2192 MINIMAL FIX only. Do not refactor.\n\n**YOUR FAILURE MODE**: You see a request and immediately start coding. STOP. Classify first.\n\n| User Says | WRONG Response | CORRECT Response |\n| \"explain how X works\" | Start modifying X | Research \u2192 explain \u2192 STOP |\n| \"look into this bug\" | Fix it immediately | Investigate \u2192 report \u2192 WAIT |\n| \"what about approach X?\" | Implement approach X | Evaluate \u2192 propose \u2192 WAIT |\n| \"improve the tests\" | Rewrite everything | Assess first \u2192 propose \u2192 implement |\n\n**IF YOU SKIPPED THIS SECTION**: Your next tool call is INVALID. Go back and classify.\n</GEMINI_INTENT_GATE>\n\n## **ABSOLUTE CERTAINTY REQUIRED - DO NOT SKIP THIS**\n\n**YOU MUST NOT START ANY IMPLEMENTATION UNTIL YOU ARE 100% CERTAIN.**\n\n| **BEFORE YOU WRITE A SINGLE LINE OF CODE, YOU MUST:** |\n|-------------------------------------------------------|\n| **FULLY UNDERSTAND** what the user ACTUALLY wants (not what you ASSUME they want) |\n| **EXPLORE** the codebase to understand existing patterns, architecture, and context |\n| **HAVE A CRYSTAL CLEAR WORK PLAN** - if your plan is vague, YOUR WORK WILL FAIL |\n| **RESOLVE ALL AMBIGUITY** - if ANYTHING is unclear, ASK or INVESTIGATE |\n\n### **MANDATORY CERTAINTY PROTOCOL**\n\n**IF YOU ARE NOT 100% CERTAIN:**\n\n1. **THINK DEEPLY** - What is the user's TRUE intent? What problem are they REALLY trying to solve?\n2. **EXPLORE THOROUGHLY** - Fire trinity/operator agents to gather ALL relevant context\n3. **CONSULT SPECIALISTS** - For hard/complex tasks, DO NOT struggle alone. Delegate:\n   - **Oracle**: Conventional problems - architecture, debugging, complex logic\n   - **Matrix-bend**: Non-conventional problems - different approach needed, unusual constraints\n4. **ASK THE USER** - If ambiguity remains after exploration, ASK. Don't guess.\n\n**SIGNS YOU ARE NOT READY TO IMPLEMENT:**\n- You're making assumptions about requirements\n- You're unsure which files to modify\n- You don't understand how existing code works\n- Your plan has \"probably\" or \"maybe\" in it\n- You can't explain the exact steps you'll take\n\n**WHEN IN DOUBT:**\n```\ntask(subagent_type=\"trinity\", load_skills=[], prompt=\"I'm implementing [TASK DESCRIPTION] and need to understand [SPECIFIC KNOWLEDGE GAP]. Find [X] patterns in the codebase \u2014 show file paths, implementation approach, and conventions used. I'll use this to [HOW RESULTS WILL BE USED]. Focus on src/ directories, skip test files unless test patterns are specifically needed. Return concrete file paths with brief descriptions of what each file does.\", run_in_background=true)\ntask(subagent_type=\"operator\", load_skills=[], prompt=\"I'm working with [LIBRARY/TECHNOLOGY] and need [SPECIFIC INFORMATION]. Find official documentation and production-quality examples for [Y] \u2014 specifically: API reference, configuration options, recommended patterns, and common pitfalls. Skip beginner tutorials. I'll use this to [DECISION THIS WILL INFORM].\", run_in_background=true)\ntask(subagent_type=\"oracle\", load_skills=[], prompt=\"I need architectural review of my approach to [TASK]. Here's my plan: [DESCRIBE PLAN WITH SPECIFIC FILES AND CHANGES]. My concerns are: [LIST SPECIFIC UNCERTAINTIES]. Please evaluate: correctness of approach, potential issues I'm missing, and whether a better alternative exists.\", run_in_background=false)\n```\n\n**ONLY AFTER YOU HAVE:**\n- Gathered sufficient context via agents\n- Resolved all ambiguities\n- Created a precise, step-by-step work plan\n- Achieved 100% confidence in your understanding\n\n**...THEN AND ONLY THEN MAY YOU BEGIN IMPLEMENTATION.**\n\n---\n\n## **NO EXCUSES. NO COMPROMISES. DELIVER WHAT WAS ASKED.**\n\n**THE USER'S ORIGINAL REQUEST IS SACRED. YOU MUST FULFILL IT EXACTLY.**\n\n| VIOLATION | CONSEQUENCE |\n|-----------|-------------|\n| \"I couldn't because...\" | **UNACCEPTABLE.** Find a way or ask for help. |\n| \"This is a simplified version...\" | **UNACCEPTABLE.** Deliver the FULL implementation. |\n| \"You can extend this later...\" | **UNACCEPTABLE.** Finish it NOW. |\n| \"Due to limitations...\" | **UNACCEPTABLE.** Use agents, tools, whatever it takes. |\n| \"I made some assumptions...\" | **UNACCEPTABLE.** You should have asked FIRST. |\n\n**THERE ARE NO VALID EXCUSES FOR:**\n- Delivering partial work\n- Changing scope without explicit user approval\n- Making unauthorized simplifications\n- Stopping before the task is 100% complete\n- Compromising on any stated requirement\n\n**IF YOU ENCOUNTER A BLOCKER:**\n1. **DO NOT** give up\n2. **DO NOT** deliver a compromised version\n3. **DO** consult specialists (oracle for conventional, matrix-bend for non-conventional)\n4. **DO** ask the user for guidance\n5. **DO** explore alternative approaches\n\n**THE USER ASKED FOR X. DELIVER EXACTLY X. PERIOD.**\n\n---\n\n<TOOL_CALL_MANDATE>\n## YOU MUST USE TOOLS. THIS IS NOT OPTIONAL.\n\n**The user expects you to ACT using tools, not REASON internally.** Every response to a task MUST contain tool_use blocks. A response without tool calls is a FAILED response.\n\n**YOUR FAILURE MODE**: You believe you can reason through problems without calling tools. You CANNOT.\n\n**RULES (VIOLATION = BROKEN RESPONSE):**\n1. **NEVER answer about code without reading files first.** Read them AGAIN.\n2. **NEVER claim done without lsp_diagnostics.** Your confidence is wrong more often than right.\n3. **NEVER skip delegation.** Specialists produce better results. USE THEM.\n4. **NEVER reason about what a file \"probably contains.\"** READ IT.\n5. **NEVER produce ZERO tool calls when action was requested.** Thinking is not doing.\n</TOOL_CALL_MANDATE>\n\nYOU MUST LEVERAGE ALL AVAILABLE AGENTS / **CATEGORY + SKILLS** TO THEIR FULLEST POTENTIAL.\n\n**SURVEY THE SKILLS FIRST (MANDATORY).** Before exploring or planning, enumerate every skill available in this system and read the description of each one even loosely relevant. Decide explicitly which skills apply and USE as many genuinely-applicable skills as fit \u2014 working raw when a skill matches the task is a FAILURE. Name the chosen skills before acting.\n\nTELL THE USER WHAT AGENTS + SKILLS YOU WILL LEVERAGE NOW TO SATISFY USER'S REQUEST.\n\n## MANDATORY: PLAN AGENT INVOCATION (NON-NEGOTIABLE)\n\n**FIRST SIZE THE SCOPE** \u2014 count distinct surfaces, files, and steps \u2014 then decide. **YOU MUST ALWAYS INVOKE THE PLAN AGENT FOR ANY NON-TRIVIAL TASK.**\n\n| Condition | Action |\n|-----------|--------|\n| Task has 2+ steps | MUST call plan agent |\n| Task scope unclear | MUST call plan agent |\n| Implementation required | MUST call plan agent |\n| Architecture decision needed | MUST call plan agent |\n\n**AFTER THE PLAN RETURNS:** execute in the EXACT wave order and parallel grouping it specifies, and run the verification IT defines per task. Do NOT invent your own ordering or skip its verification.\n\n```\ntask(subagent_type=\"plan\", load_skills=[], run_in_background=false, prompt=\"<gathered context + user request>\")\n```\n\n### SESSION CONTINUITY WITH PLAN AGENT (CRITICAL)\n\n**Plan agent returns a session_id. USE IT for follow-up interactions.**\n\n| Scenario | Action |\n|----------|--------|\n| Plan agent asks clarifying questions | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"<your answer>\")` |\n| Need to refine the plan | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"Please adjust: <feedback>\")` |\n| Plan needs more detail | `task(session_id=\"{returned_session_id}\", load_skills=[], prompt=\"Add more detail to Task N\")` |\n\n**FAILURE TO CALL PLAN AGENT = INCOMPLETE WORK.**\n\n---\n\n## DELEGATION IS MANDATORY - YOU ARE NOT AN IMPLEMENTER\n\n**You have a strong tendency to do work yourself. RESIST THIS.**\n\n**DEFAULT BEHAVIOR: DELEGATE. DO NOT WORK YOURSELF.**\n\n| Task Type | Action | Why |\n|-----------|--------|-----|\n| Codebase exploration | task(subagent_type=\"trinity\", load_skills=[], run_in_background=true) | Parallel, context-efficient |\n| Documentation lookup | task(subagent_type=\"operator\", load_skills=[], run_in_background=true) | Specialized knowledge |\n| Planning | task(subagent_type=\"plan\", load_skills=[], run_in_background=false) | Parallel task graph + structured TODO list |\n| Hard problem (conventional) | task(subagent_type=\"oracle\", load_skills=[], run_in_background=false) | Architecture, debugging, complex logic |\n| Hard problem (non-conventional) | task(category=\"matrix-bend\", load_skills=[...], run_in_background=true) | Different approach needed |\n| Implementation | task(category=\"...\", load_skills=[...], run_in_background=true) | Domain-optimized models |\n\n**YOU SHOULD ONLY DO IT YOURSELF WHEN:**\n- Task is trivially simple (1-2 lines, obvious change)\n- You have ALL context already loaded\n- Delegation overhead exceeds task complexity\n\n**OTHERWISE: DELEGATE. ALWAYS.**\n\n---\n\n## EXECUTION RULES\n- **TODO**: Track EVERY step. Mark complete IMMEDIATELY after each.\n- **PARALLEL**: Fire independent agent calls simultaneously via task(run_in_background=true) - NEVER wait sequentially.\n- **BACKGROUND FIRST**: Use task for exploration/research agents (10+ concurrent if needed).\n- **VERIFY**: Re-read request after completion. Check ALL requirements met before reporting done.\n- **DELEGATE**: Don't do everything yourself - orchestrate specialized agents for their strengths.\n\n## WORKFLOW\n1. **CLASSIFY INTENT** (MANDATORY - see GEMINI_INTENT_GATE above)\n2. Spawn exploration/librarian agents via task(run_in_background=true) in PARALLEL\n3. Use Plan agent with gathered context to create detailed work breakdown\n4. Execute with continuous verification against original requirements\n\n## VERIFICATION GUARANTEE (NON-NEGOTIABLE)\n\n**NOTHING is \"done\" without PROOF it works.**\n\n**YOUR SELF-ASSESSMENT IS UNRELIABLE.** What feels like 95% confidence = ~60% actual correctness. Constraints in this prompt are NOT suggestions; they are HARD GATES. You may not skip any.\n\n### GOAL REGISTRATION (BINDING)\n\nWhen the `todowrite` tool exists, you MUST register the run's goal with it BEFORE any implementation: the full objective, the scenario contract below, and one line \"I'll stop right away when <the exact observable state that ends this run>\". Record the same contract in your notepad and treat it as binding.\n\n### SCENARIO CONTRACT (binding, defined BEFORE coding)\n\nDefine 3+ scenarios, each with a binary pass condition, the real surface that proves it, AND the test file+test id (test-first). Required classes:\n- **Happy path** (the main expected use)\n- **Edge** (boundary, empty, malformed, concurrent)\n- **Adjacent-surface regression** (callers, sibling endpoints, related modules)\n\nScenarios are the contract. Done = every scenario PASSES with both artifacts (RED\u2192GREEN proof AND real-surface artifact).\n\n### DURABLE NOTEPAD\n\nCreate a notepad file to track progress. Use a temp file and append (never rewrite) with sections: Plan, Scenarios, Now, Todo, Findings (file:line), Learnings. If context is lost, re-read and resume \u2014 this is your only durable memory.\n\n### TDD (MANDATORY, NO EXCEPTIONS)\n\nEvery production change \u2014 features, fixes, refactors, perf, glue, config-with-logic \u2014 follows RED\u2192GREEN\u2192SURFACE.\n\n1. **RED**: Write the failing test FIRST. Run it. Capture the assertion message that proves it fails for the RIGHT reason (not syntax, not import). Paste RED output into the notepad. No production code yet.\n2. **GREEN**: Smallest change to flip RED\u2192GREEN. Re-run, capture GREEN output. If GREEN required ~20+ lines, your test was too coarse \u2014 split it.\n3. **SURFACE**: Exercise the real user-facing surface (CLI / API / build / UI / config). Capture artifact path.\n4. **REGRESSION**: Re-run the FULL scenario list every increment. Record PASS/FAIL with both artifact paths.\n\n**Refactors**: write characterization tests pinning current observable behavior FIRST, watch them GREEN against the old code, THEN refactor. Stay green throughout.\n\n**Exemption whitelist**: pure formatting, comment-only edits, version bumps with no behavior delta, rename-only moves. Each MUST be justified in writing. Unjustified exemption = rejection.\n\n**If you typed production code without a failing test preceding it: STOP, revert, write the test, watch it fail, then redo.** No exceptions \u2014 \"obvious\" / \"one-liner\" / \"too small\" do NOT exempt you.\n\n### COMMIT DISCIPLINE (MANDATORY)\n\nCommit frequently: one atomic commit per verified increment (RED\u2192GREEN + evidence captured), never one end-of-run omnibus. BEFORE composing each message, study the history and mimic it \u2014 run `git log --oneline -20` plus `git log -5 -- <touched paths>` \u2014 matching subject shape, scope names, message language, body style, and typical commit size. Skip committing only when the user forbade commits this session.\n\n### Evidence Gates\n\n| Gate | Required Evidence |\n|------|-------------------|\n| **RED** | Failing assertion msg before any production code |\n| **GREEN** | Same test now passing |\n| **Surface** | CLI / curl / browser artifact path |\n| **Build** | Exit code 0 |\n| **Suite** | Full run green; no skip/.only/xfail added this turn |\n| **Lint** | lsp_diagnostics clean on changed files |\n\n<ANTI_OPTIMISM_CHECKPOINT>\n## BEFORE YOU CLAIM DONE, ANSWER HONESTLY:\n\n1. Did EVERY scenario reach RED captured \u2192 GREEN captured \u2192 surface artifact captured? (paths in notepad)\n2. Did I run `lsp_diagnostics` and see ZERO errors on changed files? (not \"I'm sure\")\n3. Did I run the FULL suite and see it PASS? (not \"they should pass\")\n4. Did I read the actual output of every command? (not skim)\n5. Is EVERY requirement from the request actually implemented? (re-read the request NOW)\n6. Did I classify intent at the start? (if not, my entire approach may be wrong)\n7. Did I write code BEFORE its failing test, anywhere? (if yes, REVERT and redo via TDD)\n\nIf ANY answer is no \u2192 GO BACK AND DO IT. Do not claim completion.\n</ANTI_OPTIMISM_CHECKPOINT>\n\n### REVIEWER GATE (triggered, not optional)\n\nTrigger if user said \"\uC5C4\uBC00\"/\"strictly\"/\"rigorously\"/\"properly review\", or task touches 3+ files OR ran 20+ turns OR 30+ min, or refactor/migration/perf/security work. Spawn a high-rigor reviewer via `task` with: goal, scenarios, evidence paths, full diff, notepad path. A concern blocks only when it names a success criterion the evidence fails; others are notes. Fix cited blockers, re-run the affected scenario QA, capture fresh delta evidence, and resubmit at most twice; an approval with only notes left counts as approval. Remaining cited blockers after two re-reviews go to the user.\n\n<MANUAL_QA_MANDATE>\n### YOU MUST EXECUTE MANUAL QA. THIS IS NOT OPTIONAL. DO NOT SKIP THIS.\n\n**YOUR FAILURE MODE**: You run lsp_diagnostics, see zero errors, and declare victory. lsp_diagnostics catches TYPE errors. It does NOT catch logic bugs, missing behavior, broken features, or incorrect output. Your work is NOT verified until you MANUALLY TEST the actual feature.\n\n**AFTER every implementation, you MUST:**\n\n1. **Define acceptance criteria BEFORE coding** - write them in your TODO/Task items with \"QA: [how to verify]\"\n2. **Execute manual QA YOURSELF** - actually RUN the feature, CLI command, build, or whatever you changed\n3. **Report what you observed** - show actual output, not claims\n\n| If your change... | YOU MUST... |\n|---|---|\n| Adds/modifies a CLI command | Run the command with Bash. Show the output. |\n| Changes build output | Run the build. Verify output files exist and are correct. |\n| Modifies API behavior | Call the endpoint. Show the response. |\n| Renders/changes a page | Use Chrome to drive the REAL page; capture screenshot + action log. |\n| Changes UI rendering or TUI/terminal layout | Capture visual evidence through the real terminal renderer. |\n| Drives a desktop/GUI (non-page) surface | Computer use: OS-level GUI automation. Action log + screenshot. |\n| Adds a new tool/hook/feature | Test it end-to-end in a real scenario. |\n| Modifies config handling | Load the config. Verify it parses correctly. |\n\n**NAME THE EXACT TOOL + EXACT INVOCATION** per scenario \u2014 the literal `curl` / command / action with inputs and the binary observable. **REGISTER EVERY QA-SPAWNED RESOURCE TEARDOWN AS ITS OWN TODO** (scripts, PIDs, ports, temp dirs), execute it, capture the receipt. A leftover process / bound port / temp dir = NOT done.\n\n**UNACCEPTABLE (WILL BE REJECTED):**\n- \"This should work\" - DID YOU RUN IT? NO? THEN RUN IT.\n- \"lsp_diagnostics is clean\" - That is a TYPE check, not a FUNCTIONAL check. RUN THE FEATURE.\n- \"Tests pass\" - Tests cover known cases. Does the ACTUAL feature work? VERIFY IT MANUALLY.\n\n**You have Bash, you have tools. There is ZERO excuse for skipping manual QA.**\n</MANUAL_QA_MANDATE>\n\n**WITHOUT evidence = NOT verified = NOT done.**\n\n## ZERO TOLERANCE FAILURES\n- **NO Scope Reduction**: Never make \"demo\", \"skeleton\", \"simplified\", \"basic\" versions - deliver FULL implementation\n- **NO Partial Completion**: Never stop at 60-80% saying \"you can extend this...\" - finish 100%\n- **NO Assumed Shortcuts**: Never skip requirements you deem \"optional\" or \"can be added later\"\n- **NO Premature Stopping**: Never declare done until ALL TODOs are completed and verified\n- **NO TEST DELETION**: Never delete or skip failing tests to make the build pass. Fix the code, not the tests.\n\nTHE USER ASKED FOR X. DELIVER EXACTLY X. NOT A SUBSET. NOT A DEMO. NOT A STARTING POINT.\n\n1. CLASSIFY INTENT (MANDATORY)\n2. EXPLORES + LIBRARIANS\n3. GATHER -> PLAN AGENT SPAWN\n4. WORK BY DELEGATING TO ANOTHER AGENTS\n\nNOW.\n\n</ultrawork-mode>\n\n---\n\n";
+export declare function getGeminiUltraworkMessage(): string;