@caupulican/pi-adaptative 0.80.95 → 0.80.97

package/dist/core/toolkit/reflex-interpreter.d.ts

@@ -0,0 +1,28 @@
+import type { ToolkitScript } from "./script-registry.ts";
+/**
+ * Reflex interpreter (the BRAIN half of the user-ratified brain->muscle pipeline, statistically
+ * validated at 10/10 correct interpretations on the hard registry): a strict instruction
+ * interpreter that maps a fuzzy user request onto the toolkit registry. It NEVER executes
+ * anything — it only proposes `{script, args, danger, confidence}` for the deterministic
+ * executor path, which still enforces every safety rule (danger confirmation included).
+ */
+export declare const REFLEX_INTERPRETER_SYSTEM_PROMPT: string;
+export interface ReflexPlan {
+    script: string;
+    args: string[];
+    danger: boolean;
+    confidence: number;
+}
+export declare const REFLEX_MIN_CONFIDENCE = 0.75;
+export declare function buildReflexUserPrompt(request: string, scripts: readonly ToolkitScript[]): string;
+export declare function parseReflexPlan(text: string): ReflexPlan | undefined;
+/**
+ * Accept a plan only when it is confident AND names a real registry script ("none" and unknown
+ * names are honest refusals). The DANGER flag is never trusted from the plan — the registry's
+ * own flag governs confirmation, exactly as for a direct match.
+ */
+export declare function acceptReflexPlan(plan: ReflexPlan | undefined, scripts: readonly ToolkitScript[]): {
+    script: ToolkitScript;
+    args: string[];
+} | undefined;
+//# sourceMappingURL=reflex-interpreter.d.ts.map

package/dist/core/toolkit/reflex-interpreter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"reflex-interpreter.d.ts","sourceRoot":"","sources":["../../../src/core/toolkit/reflex-interpreter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAE1D;;;;;;GAMG;AAEH,eAAO,MAAM,gCAAgC,QAMjC,CAAC;AAEb,MAAM,WAAW,UAAU;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,MAAM,EAAE,OAAO,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,eAAO,MAAM,qBAAqB,OAAO,CAAC;AAE1C,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,aAAa,EAAE,GAAG,MAAM,CAKhG;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS,CA6BpE;AAED;;;;GAIG;AACH,wBAAgB,gBAAgB,CAC/B,IAAI,EAAE,UAAU,GAAG,SAAS,EAC5B,OAAO,EAAE,SAAS,aAAa,EAAE,GAC/B;IAAE,MAAM,EAAE,aAAa,CAAC;IAAC,IAAI,EAAE,MAAM,EAAE,CAAA;CAAE,GAAG,SAAS,CAKvD","sourcesContent":["import type { ToolkitScript } from \"./script-registry.ts\";\n\n/**\n * Reflex interpreter (the BRAIN half of the user-ratified brain->muscle pipeline, statistically\n * validated at 10/10 correct interpretations on the hard registry): a strict instruction\n * interpreter that maps a fuzzy user request onto the toolkit registry. It NEVER executes\n * anything — it only proposes `{script, args, danger, confidence}` for the deterministic\n * executor path, which still enforces every safety rule (danger confirmation included).\n */\n\nexport const REFLEX_INTERPRETER_SYSTEM_PROMPT = [\n\t\"You are a strict instruction interpreter for a script executor. You NEVER execute anything.\",\n\t\"Given a user request and the script registry, output STRICT JSON only:\",\n\t'{\"script\":\"<exact registry name>\",\"args\":[\"...\"],\"danger\":true|false,\"confidence\":<0..1>}',\n\t\"Pick the single best script. Extract arguments exactly as the script expects. Mark danger from the registry.\",\n\t'If no script fits, output {\"script\":\"none\",\"args\":[],\"danger\":false,\"confidence\":0}.',\n].join(\"\\n\");\n\nexport interface ReflexPlan {\n\tscript: string;\n\targs: string[];\n\tdanger: boolean;\n\tconfidence: number;\n}\n\nexport const REFLEX_MIN_CONFIDENCE = 0.75;\n\nexport function buildReflexUserPrompt(request: string, scripts: readonly ToolkitScript[]): string {\n\tconst registry = scripts\n\t\t.map((script) => `${script.name}: ${script.description}${script.danger ? \" [DANGEROUS]\" : \"\"}`)\n\t\t.join(\"\\n\");\n\treturn `Registry:\\n${registry}\\n\\nRequest: ${request.slice(0, 2000)}`;\n}\n\nexport function parseReflexPlan(text: string): ReflexPlan | undefined {\n\tconst trimmed = text.trim();\n\t// Strip a leading think block some local models emit even with thinking off.\n\tconst thinkEnd = trimmed.indexOf(\"</think>\");\n\tconst body = thinkEnd >= 0 ? trimmed.slice(thinkEnd + 8) : trimmed;\n\tconst start = body.indexOf(\"{\");\n\tconst end = body.lastIndexOf(\"}\");\n\tif (start < 0 || end <= start) return undefined;\n\ttry {\n\t\tconst parsed = JSON.parse(body.slice(start, end + 1)) as {\n\t\t\tscript?: unknown;\n\t\t\targs?: unknown;\n\t\t\tdanger?: unknown;\n\t\t\tconfidence?: unknown;\n\t\t};\n\t\tif (typeof parsed.script !== \"string\" || parsed.script.length === 0) return undefined;\n\t\tconst confidence =\n\t\t\ttypeof parsed.confidence === \"number\" && Number.isFinite(parsed.confidence)\n\t\t\t\t? Math.max(0, Math.min(1, parsed.confidence))\n\t\t\t\t: 0;\n\t\treturn {\n\t\t\tscript: parsed.script,\n\t\t\targs: Array.isArray(parsed.args) ? parsed.args.filter((arg): arg is string => typeof arg === \"string\") : [],\n\t\t\tdanger: parsed.danger === true,\n\t\t\tconfidence,\n\t\t};\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\n/**\n * Accept a plan only when it is confident AND names a real registry script (\"none\" and unknown\n * names are honest refusals). The DANGER flag is never trusted from the plan — the registry's\n * own flag governs confirmation, exactly as for a direct match.\n */\nexport function acceptReflexPlan(\n\tplan: ReflexPlan | undefined,\n\tscripts: readonly ToolkitScript[],\n): { script: ToolkitScript; args: string[] } | undefined {\n\tif (!plan || plan.confidence < REFLEX_MIN_CONFIDENCE) return undefined;\n\tconst script = scripts.find((entry) => entry.name === plan.script);\n\tif (!script) return undefined;\n\treturn { script, args: plan.args };\n}\n"]}

package/dist/core/toolkit/reflex-interpreter.js

@@ -0,0 +1,62 @@
+/**
+ * Reflex interpreter (the BRAIN half of the user-ratified brain->muscle pipeline, statistically
+ * validated at 10/10 correct interpretations on the hard registry): a strict instruction
+ * interpreter that maps a fuzzy user request onto the toolkit registry. It NEVER executes
+ * anything — it only proposes `{script, args, danger, confidence}` for the deterministic
+ * executor path, which still enforces every safety rule (danger confirmation included).
+ */
+export const REFLEX_INTERPRETER_SYSTEM_PROMPT = [
+    "You are a strict instruction interpreter for a script executor. You NEVER execute anything.",
+    "Given a user request and the script registry, output STRICT JSON only:",
+    '{"script":"<exact registry name>","args":["..."],"danger":true|false,"confidence":<0..1>}',
+    "Pick the single best script. Extract arguments exactly as the script expects. Mark danger from the registry.",
+    'If no script fits, output {"script":"none","args":[],"danger":false,"confidence":0}.',
+].join("\n");
+export const REFLEX_MIN_CONFIDENCE = 0.75;
+export function buildReflexUserPrompt(request, scripts) {
+    const registry = scripts
+        .map((script) => `${script.name}: ${script.description}${script.danger ? " [DANGEROUS]" : ""}`)
+        .join("\n");
+    return `Registry:\n${registry}\n\nRequest: ${request.slice(0, 2000)}`;
+}
+export function parseReflexPlan(text) {
+    const trimmed = text.trim();
+    // Strip a leading think block some local models emit even with thinking off.
+    const thinkEnd = trimmed.indexOf("</think>");
+    const body = thinkEnd >= 0 ? trimmed.slice(thinkEnd + 8) : trimmed;
+    const start = body.indexOf("{");
+    const end = body.lastIndexOf("}");
+    if (start < 0 || end <= start)
+        return undefined;
+    try {
+        const parsed = JSON.parse(body.slice(start, end + 1));
+        if (typeof parsed.script !== "string" || parsed.script.length === 0)
+            return undefined;
+        const confidence = typeof parsed.confidence === "number" && Number.isFinite(parsed.confidence)
+            ? Math.max(0, Math.min(1, parsed.confidence))
+            : 0;
+        return {
+            script: parsed.script,
+            args: Array.isArray(parsed.args) ? parsed.args.filter((arg) => typeof arg === "string") : [],
+            danger: parsed.danger === true,
+            confidence,
+        };
+    }
+    catch {
+        return undefined;
+    }
+}
+/**
+ * Accept a plan only when it is confident AND names a real registry script ("none" and unknown
+ * names are honest refusals). The DANGER flag is never trusted from the plan — the registry's
+ * own flag governs confirmation, exactly as for a direct match.
+ */
+export function acceptReflexPlan(plan, scripts) {
+    if (!plan || plan.confidence < REFLEX_MIN_CONFIDENCE)
+        return undefined;
+    const script = scripts.find((entry) => entry.name === plan.script);
+    if (!script)
+        return undefined;
+    return { script, args: plan.args };
+}
+//# sourceMappingURL=reflex-interpreter.js.map

package/dist/core/toolkit/reflex-interpreter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"reflex-interpreter.js","sourceRoot":"","sources":["../../../src/core/toolkit/reflex-interpreter.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AAEH,MAAM,CAAC,MAAM,gCAAgC,GAAG;IAC/C,6FAA6F;IAC7F,wEAAwE;IACxE,2FAA2F;IAC3F,8GAA8G;IAC9G,sFAAsF;CACtF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AASb,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC;AAE1C,MAAM,UAAU,qBAAqB,CAAC,OAAe,EAAE,OAAiC,EAAU;IACjG,MAAM,QAAQ,GAAG,OAAO;SACtB,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;SAC9F,IAAI,CAAC,IAAI,CAAC,CAAC;IACb,OAAO,cAAc,QAAQ,gBAAgB,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC;AAAA,CACtE;AAED,MAAM,UAAU,eAAe,CAAC,IAAY,EAA0B;IACrE,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,6EAA6E;IAC7E,MAAM,QAAQ,GAAG,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;IAC7C,MAAM,IAAI,GAAG,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IACnE,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAChC,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IAClC,IAAI,KAAK,GAAG,CAAC,IAAI,GAAG,IAAI,KAAK;QAAE,OAAO,SAAS,CAAC;IAChD,IAAI,CAAC;QACJ,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAKnD,CAAC;QACF,IAAI,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,SAAS,CAAC;QACtF,MAAM,UAAU,GACf,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,UAAU,CAAC;YAC1E,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;YAC7C,CAAC,CAAC,CAAC,CAAC;QACN,OAAO;YACN,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAiB,EAAE,CAAC,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE;YAC3G,MAAM,EAAE,MAAM,CAAC,MAAM,KAAK,IAAI;YAC9B,UAAU;SACV,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAC/B,IAA4B,EAC5B,OAAiC,EACuB;IACxD,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,UAAU,GAAG,qBAAqB;QAAE,OAAO,SAAS,CAAC;IACvE,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,CAAC;IACnE,IAAI,CAAC,MAAM;QAAE,OAAO,SAAS,CAAC;IAC9B,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC;AAAA,CACnC","sourcesContent":["import type { ToolkitScript } from \"./script-registry.ts\";\n\n/**\n * Reflex interpreter (the BRAIN half of the user-ratified brain->muscle pipeline, statistically\n * validated at 10/10 correct interpretations on the hard registry): a strict instruction\n * interpreter that maps a fuzzy user request onto the toolkit registry. It NEVER executes\n * anything — it only proposes `{script, args, danger, confidence}` for the deterministic\n * executor path, which still enforces every safety rule (danger confirmation included).\n */\n\nexport const REFLEX_INTERPRETER_SYSTEM_PROMPT = [\n\t\"You are a strict instruction interpreter for a script executor. You NEVER execute anything.\",\n\t\"Given a user request and the script registry, output STRICT JSON only:\",\n\t'{\"script\":\"<exact registry name>\",\"args\":[\"...\"],\"danger\":true|false,\"confidence\":<0..1>}',\n\t\"Pick the single best script. Extract arguments exactly as the script expects. Mark danger from the registry.\",\n\t'If no script fits, output {\"script\":\"none\",\"args\":[],\"danger\":false,\"confidence\":0}.',\n].join(\"\\n\");\n\nexport interface ReflexPlan {\n\tscript: string;\n\targs: string[];\n\tdanger: boolean;\n\tconfidence: number;\n}\n\nexport const REFLEX_MIN_CONFIDENCE = 0.75;\n\nexport function buildReflexUserPrompt(request: string, scripts: readonly ToolkitScript[]): string {\n\tconst registry = scripts\n\t\t.map((script) => `${script.name}: ${script.description}${script.danger ? \" [DANGEROUS]\" : \"\"}`)\n\t\t.join(\"\\n\");\n\treturn `Registry:\\n${registry}\\n\\nRequest: ${request.slice(0, 2000)}`;\n}\n\nexport function parseReflexPlan(text: string): ReflexPlan | undefined {\n\tconst trimmed = text.trim();\n\t// Strip a leading think block some local models emit even with thinking off.\n\tconst thinkEnd = trimmed.indexOf(\"</think>\");\n\tconst body = thinkEnd >= 0 ? trimmed.slice(thinkEnd + 8) : trimmed;\n\tconst start = body.indexOf(\"{\");\n\tconst end = body.lastIndexOf(\"}\");\n\tif (start < 0 || end <= start) return undefined;\n\ttry {\n\t\tconst parsed = JSON.parse(body.slice(start, end + 1)) as {\n\t\t\tscript?: unknown;\n\t\t\targs?: unknown;\n\t\t\tdanger?: unknown;\n\t\t\tconfidence?: unknown;\n\t\t};\n\t\tif (typeof parsed.script !== \"string\" || parsed.script.length === 0) return undefined;\n\t\tconst confidence =\n\t\t\ttypeof parsed.confidence === \"number\" && Number.isFinite(parsed.confidence)\n\t\t\t\t? Math.max(0, Math.min(1, parsed.confidence))\n\t\t\t\t: 0;\n\t\treturn {\n\t\t\tscript: parsed.script,\n\t\t\targs: Array.isArray(parsed.args) ? parsed.args.filter((arg): arg is string => typeof arg === \"string\") : [],\n\t\t\tdanger: parsed.danger === true,\n\t\t\tconfidence,\n\t\t};\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\n/**\n * Accept a plan only when it is confident AND names a real registry script (\"none\" and unknown\n * names are honest refusals). The DANGER flag is never trusted from the plan — the registry's\n * own flag governs confirmation, exactly as for a direct match.\n */\nexport function acceptReflexPlan(\n\tplan: ReflexPlan | undefined,\n\tscripts: readonly ToolkitScript[],\n): { script: ToolkitScript; args: string[] } | undefined {\n\tif (!plan || plan.confidence < REFLEX_MIN_CONFIDENCE) return undefined;\n\tconst script = scripts.find((entry) => entry.name === plan.script);\n\tif (!script) return undefined;\n\treturn { script, args: plan.args };\n}\n"]}

package/dist/core/tools/run-toolkit-script.d.ts

@@ -1,5 +1,6 @@
 import { type Static, Type } from "typebox";
 import type { ToolDefinition } from "../extensions/types.ts";
+import { type ReflexPlan } from "../toolkit/reflex-interpreter.ts";
 import { type ToolkitScript } from "../toolkit/script-registry.ts";
 import type { ScriptExecution } from "../toolkit/script-runner.ts";
 declare const runToolkitScriptSchema: Type.TObject<{
@@ -18,6 +19,13 @@ export interface RunToolkitScriptDetails {
 export interface RunToolkitScriptDependencies {
     getScripts: () => ToolkitScript[];
     execute: (script: ToolkitScript, args: readonly string[]) => Promise<ScriptExecution>;
+    /**
+     * Optional reflex interpreter (local brain): consulted ONLY when the deterministic Level-0
+     * matcher is ambiguous, to resolve fuzzy phrasing ("prepare db" vs "update db") into a
+     * registry pick. Its plan is advisory — the danger/confirm rules and structural execution
+     * contract apply identically to brain-selected scripts.
+     */
+    interpret?: (request: string, scripts: readonly ToolkitScript[]) => Promise<ReflexPlan | undefined>;
 }
 export declare function createRunToolkitScriptToolDefinition(deps: RunToolkitScriptDependencies): ToolDefinition;
 export {};

package/dist/core/tools/run-toolkit-script.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run-toolkit-script.d.ts","sourceRoot":"","sources":["../../../src/core/tools/run-toolkit-script.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,MAAM,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAC5C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAsB,KAAK,aAAa,EAAE,MAAM,+BAA+B,CAAC;AACvF,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,6BAA6B,CAAC;AAEnE,QAAA,MAAM,sBAAsB;;;;EAc3B,CAAC;AAEF,MAAM,MAAM,qBAAqB,GAAG,MAAM,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAE1E,MAAM,WAAW,uBAAuB;IACvC,OAAO,EAAE,UAAU,GAAG,QAAQ,GAAG,WAAW,GAAG,WAAW,GAAG,uBAAuB,GAAG,gBAAgB,CAAC;IACxG,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,4BAA4B;IAC5C,UAAU,EAAE,MAAM,aAAa,EAAE,CAAC;IAClC,OAAO,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,SAAS,MAAM,EAAE,KAAK,OAAO,CAAC,eAAe,CAAC,CAAC;CACtF;AAMD,wBAAgB,oCAAoC,CAAC,IAAI,EAAE,4BAA4B,GAAG,cAAc,CAmGvG","sourcesContent":["import { type Static, Type } from \"typebox\";\nimport type { ToolDefinition } from \"../extensions/types.ts\";\nimport { matchToolkitScript, type ToolkitScript } from \"../toolkit/script-registry.ts\";\nimport type { ScriptExecution } from \"../toolkit/script-runner.ts\";\n\nconst runToolkitScriptSchema = Type.Object(\n\t{\n\t\tscript: Type.String({\n\t\t\tdescription:\n\t\t\t\t\"The toolkit script to run: its registered name, a taught alias, or a natural request (e.g. 'restore-db' or 'restore the database'). Ambiguous requests return a shortlist instead of executing.\",\n\t\t}),\n\t\targs: Type.Optional(Type.Array(Type.String(), { description: \"Arguments passed to the script as-is.\" })),\n\t\tconfirm: Type.Optional(\n\t\t\tType.Boolean({\n\t\t\t\tdescription: \"Required true to run a script flagged danger: scripts marked dangerous never run without it.\",\n\t\t\t}),\n\t\t),\n\t},\n\t{ additionalProperties: false },\n);\n\nexport type RunToolkitScriptInput = Static<typeof runToolkitScriptSchema>;\n\nexport interface RunToolkitScriptDetails {\n\toutcome: \"executed\" | \"failed\" | \"ambiguous\" | \"not_found\" | \"confirmation_required\" | \"empty_registry\";\n\tscriptName?: string;\n\texitCode?: number | null;\n\tdurationMs?: number;\n\tshortlist?: string[];\n}\n\nexport interface RunToolkitScriptDependencies {\n\tgetScripts: () => ToolkitScript[];\n\texecute: (script: ToolkitScript, args: readonly string[]) => Promise<ScriptExecution>;\n}\n\nfunction boundedOutput(text: string, maxChars = 4000): string {\n\treturn text.length <= maxChars ? text : `${text.slice(0, maxChars)}\\n… (truncated)`;\n}\n\nexport function createRunToolkitScriptToolDefinition(deps: RunToolkitScriptDependencies): ToolDefinition {\n\treturn {\n\t\tname: \"run_toolkit_script\",\n\t\tlabel: \"run_toolkit_script\",\n\t\tdescription:\n\t\t\t\"Run one of the user's registered toolkit scripts (daily-ops allowlist). Finding is a registry lookup — ambiguous requests return a shortlist to disambiguate, never a guess. The harness executes (uv/powershell/bash) and ALWAYS returns exit code, stdout, and stderr; failures are reported as errors — a failed script can never look like success.\",\n\t\tpromptSnippet: \"Run a registered toolkit script by name/alias; output and errors come back structurally.\",\n\t\tpromptGuidelines: [\n\t\t\t\"Prefer the exact registered name; on a shortlist response, ask the user which one (or pick only when their request clearly names it).\",\n\t\t\t\"Scripts flagged dangerous require confirm: true — obtain explicit user confirmation first.\",\n\t\t\t\"Report the script's real output/exit code; never claim success when exitCode is non-zero.\",\n\t\t],\n\t\tparameters: runToolkitScriptSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\tinput: RunToolkitScriptInput,\n\t\t): Promise<{\n\t\t\tcontent: Array<{ type: \"text\"; text: string }>;\n\t\t\tdetails: RunToolkitScriptDetails;\n\t\t\tisError?: boolean;\n\t\t}> {\n\t\t\tconst scripts = deps.getScripts();\n\t\t\tif (scripts.length === 0) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: \"run_toolkit_script: no scripts registered (settings.toolkit.scripts).\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"empty_registry\" },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst match = matchToolkitScript(input.script, scripts);\n\t\t\tif (match.kind === \"none\") {\n\t\t\t\tconst closest = match.closest.map((script) => `${script.name} — ${script.description}`).join(\"\\n\");\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `script_not_found: nothing registered matches \"${input.script}\".${closest ? `\\nClosest:\\n${closest}` : \"\"}`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"not_found\" },\n\t\t\t\t\tisError: true,\n\t\t\t\t};\n\t\t\t}\n\t\t\tif (match.kind === \"ambiguous\") {\n\t\t\t\tconst shortlist = match.shortlist.map((script) => `${script.name} — ${script.description}`);\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `Ambiguous request — did you mean:\\n${shortlist.join(\"\\n\")}\\nCall again with the exact name.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"ambiguous\", shortlist: match.shortlist.map((script) => script.name) },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst script = match.script;\n\t\t\tif (script.danger && input.confirm !== true) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `\"${script.name}\" is flagged DANGEROUS (${script.description}). It was NOT run. Confirm with the user, then call again with confirm: true.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"confirmation_required\", scriptName: script.name },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst execution = await deps.execute(script, input.args ?? []);\n\t\t\tconst failed = execution.exitCode !== 0 || execution.timedOut;\n\t\t\tconst header = failed\n\t\t\t\t? `FAILED: ${script.name} exited ${execution.timedOut ? \"by timeout\" : execution.exitCode} after ${execution.durationMs}ms`\n\t\t\t\t: `${script.name} succeeded in ${execution.durationMs}ms`;\n\t\t\tconst body = [\n\t\t\t\theader,\n\t\t\t\texecution.stdout.trim() ? `stdout:\\n${boundedOutput(execution.stdout.trim())}` : \"stdout: (empty)\",\n\t\t\t\texecution.stderr.trim() ? `stderr:\\n${boundedOutput(execution.stderr.trim())}` : \"\",\n\t\t\t]\n\t\t\t\t.filter(Boolean)\n\t\t\t\t.join(\"\\n\");\n\n\t\t\treturn {\n\t\t\t\tcontent: [{ type: \"text\" as const, text: body }],\n\t\t\t\tdetails: {\n\t\t\t\t\toutcome: failed ? \"failed\" : \"executed\",\n\t\t\t\t\tscriptName: script.name,\n\t\t\t\t\texitCode: execution.exitCode,\n\t\t\t\t\tdurationMs: execution.durationMs,\n\t\t\t\t},\n\t\t\t\t...(failed ? { isError: true } : {}),\n\t\t\t};\n\t\t},\n\t};\n}\n"]}
+{"version":3,"file":"run-toolkit-script.d.ts","sourceRoot":"","sources":["../../../src/core/tools/run-toolkit-script.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,MAAM,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAC5C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAoB,KAAK,UAAU,EAAE,MAAM,kCAAkC,CAAC;AACrF,OAAO,EAAsB,KAAK,aAAa,EAAE,MAAM,+BAA+B,CAAC;AACvF,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,6BAA6B,CAAC;AAEnE,QAAA,MAAM,sBAAsB;;;;EAc3B,CAAC;AAEF,MAAM,MAAM,qBAAqB,GAAG,MAAM,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAE1E,MAAM,WAAW,uBAAuB;IACvC,OAAO,EAAE,UAAU,GAAG,QAAQ,GAAG,WAAW,GAAG,WAAW,GAAG,uBAAuB,GAAG,gBAAgB,CAAC;IACxG,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,4BAA4B;IAC5C,UAAU,EAAE,MAAM,aAAa,EAAE,CAAC;IAClC,OAAO,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,SAAS,MAAM,EAAE,KAAK,OAAO,CAAC,eAAe,CAAC,CAAC;IACtF;;;;;OAKG;IACH,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,aAAa,EAAE,KAAK,OAAO,CAAC,UAAU,GAAG,SAAS,CAAC,CAAC;CACpG;AAMD,wBAAgB,oCAAoC,CAAC,IAAI,EAAE,4BAA4B,GAAG,cAAc,CAmHvG","sourcesContent":["import { type Static, Type } from \"typebox\";\nimport type { ToolDefinition } from \"../extensions/types.ts\";\nimport { acceptReflexPlan, type ReflexPlan } from \"../toolkit/reflex-interpreter.ts\";\nimport { matchToolkitScript, type ToolkitScript } from \"../toolkit/script-registry.ts\";\nimport type { ScriptExecution } from \"../toolkit/script-runner.ts\";\n\nconst runToolkitScriptSchema = Type.Object(\n\t{\n\t\tscript: Type.String({\n\t\t\tdescription:\n\t\t\t\t\"The toolkit script to run: its registered name, a taught alias, or a natural request (e.g. 'restore-db' or 'restore the database'). Ambiguous requests return a shortlist instead of executing.\",\n\t\t}),\n\t\targs: Type.Optional(Type.Array(Type.String(), { description: \"Arguments passed to the script as-is.\" })),\n\t\tconfirm: Type.Optional(\n\t\t\tType.Boolean({\n\t\t\t\tdescription: \"Required true to run a script flagged danger: scripts marked dangerous never run without it.\",\n\t\t\t}),\n\t\t),\n\t},\n\t{ additionalProperties: false },\n);\n\nexport type RunToolkitScriptInput = Static<typeof runToolkitScriptSchema>;\n\nexport interface RunToolkitScriptDetails {\n\toutcome: \"executed\" | \"failed\" | \"ambiguous\" | \"not_found\" | \"confirmation_required\" | \"empty_registry\";\n\tscriptName?: string;\n\texitCode?: number | null;\n\tdurationMs?: number;\n\tshortlist?: string[];\n}\n\nexport interface RunToolkitScriptDependencies {\n\tgetScripts: () => ToolkitScript[];\n\texecute: (script: ToolkitScript, args: readonly string[]) => Promise<ScriptExecution>;\n\t/**\n\t * Optional reflex interpreter (local brain): consulted ONLY when the deterministic Level-0\n\t * matcher is ambiguous, to resolve fuzzy phrasing (\"prepare db\" vs \"update db\") into a\n\t * registry pick. Its plan is advisory — the danger/confirm rules and structural execution\n\t * contract apply identically to brain-selected scripts.\n\t */\n\tinterpret?: (request: string, scripts: readonly ToolkitScript[]) => Promise<ReflexPlan | undefined>;\n}\n\nfunction boundedOutput(text: string, maxChars = 4000): string {\n\treturn text.length <= maxChars ? text : `${text.slice(0, maxChars)}\\n… (truncated)`;\n}\n\nexport function createRunToolkitScriptToolDefinition(deps: RunToolkitScriptDependencies): ToolDefinition {\n\treturn {\n\t\tname: \"run_toolkit_script\",\n\t\tlabel: \"run_toolkit_script\",\n\t\tdescription:\n\t\t\t\"Run one of the user's registered toolkit scripts (daily-ops allowlist). Finding is a registry lookup — ambiguous requests return a shortlist to disambiguate, never a guess. The harness executes (uv/powershell/bash) and ALWAYS returns exit code, stdout, and stderr; failures are reported as errors — a failed script can never look like success.\",\n\t\tpromptSnippet: \"Run a registered toolkit script by name/alias; output and errors come back structurally.\",\n\t\tpromptGuidelines: [\n\t\t\t\"Prefer the exact registered name; on a shortlist response, ask the user which one (or pick only when their request clearly names it).\",\n\t\t\t\"Scripts flagged dangerous require confirm: true — obtain explicit user confirmation first.\",\n\t\t\t\"Report the script's real output/exit code; never claim success when exitCode is non-zero.\",\n\t\t],\n\t\tparameters: runToolkitScriptSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\tinput: RunToolkitScriptInput,\n\t\t): Promise<{\n\t\t\tcontent: Array<{ type: \"text\"; text: string }>;\n\t\t\tdetails: RunToolkitScriptDetails;\n\t\t\tisError?: boolean;\n\t\t}> {\n\t\t\tconst scripts = deps.getScripts();\n\t\t\tif (scripts.length === 0) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: \"run_toolkit_script: no scripts registered (settings.toolkit.scripts).\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"empty_registry\" },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst match = matchToolkitScript(input.script, scripts);\n\t\t\tif (match.kind === \"none\") {\n\t\t\t\tconst closest = match.closest.map((script) => `${script.name} — ${script.description}`).join(\"\\n\");\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `script_not_found: nothing registered matches \"${input.script}\".${closest ? `\\nClosest:\\n${closest}` : \"\"}`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"not_found\" },\n\t\t\t\t\tisError: true,\n\t\t\t\t};\n\t\t\t}\n\t\t\tlet interpreted: { script: ToolkitScript; args: string[]; confidence: number } | undefined;\n\t\t\tif (match.kind === \"ambiguous\" && deps.interpret) {\n\t\t\t\ttry {\n\t\t\t\t\tconst plan = await deps.interpret(input.script, scripts);\n\t\t\t\t\tconst accepted = acceptReflexPlan(plan, scripts);\n\t\t\t\t\tif (accepted && plan) {\n\t\t\t\t\t\tinterpreted = { ...accepted, confidence: plan.confidence };\n\t\t\t\t\t}\n\t\t\t\t} catch {\n\t\t\t\t\t// interpreter is best-effort; ambiguity handling below stays authoritative\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (match.kind === \"ambiguous\" && !interpreted) {\n\t\t\t\tconst shortlist = match.shortlist.map((script) => `${script.name} — ${script.description}`);\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `Ambiguous request — did you mean:\\n${shortlist.join(\"\\n\")}\\nCall again with the exact name.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"ambiguous\", shortlist: match.shortlist.map((script) => script.name) },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst script = interpreted?.script ?? (match as { script: ToolkitScript }).script;\n\t\t\tif (script.danger && input.confirm !== true) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `\"${script.name}\" is flagged DANGEROUS (${script.description}). It was NOT run. Confirm with the user, then call again with confirm: true.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"confirmation_required\", scriptName: script.name },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\t// Explicit args from the caller win; a brain-extracted arg list fills in for fuzzy requests.\n\t\t\tconst execution = await deps.execute(script, input.args ?? interpreted?.args ?? []);\n\t\t\tconst failed = execution.exitCode !== 0 || execution.timedOut;\n\t\t\tconst header = failed\n\t\t\t\t? `FAILED: ${script.name} exited ${execution.timedOut ? \"by timeout\" : execution.exitCode} after ${execution.durationMs}ms`\n\t\t\t\t: `${script.name} succeeded in ${execution.durationMs}ms`;\n\t\t\tconst body = [\n\t\t\t\theader,\n\t\t\t\texecution.stdout.trim() ? `stdout:\\n${boundedOutput(execution.stdout.trim())}` : \"stdout: (empty)\",\n\t\t\t\texecution.stderr.trim() ? `stderr:\\n${boundedOutput(execution.stderr.trim())}` : \"\",\n\t\t\t]\n\t\t\t\t.filter(Boolean)\n\t\t\t\t.join(\"\\n\");\n\n\t\t\treturn {\n\t\t\t\tcontent: [{ type: \"text\" as const, text: body }],\n\t\t\t\tdetails: {\n\t\t\t\t\t...(interpreted\n\t\t\t\t\t\t? { interpreter: { script: interpreted.script.name, confidence: interpreted.confidence } }\n\t\t\t\t\t\t: {}),\n\t\t\t\t\toutcome: failed ? \"failed\" : \"executed\",\n\t\t\t\t\tscriptName: script.name,\n\t\t\t\t\texitCode: execution.exitCode,\n\t\t\t\t\tdurationMs: execution.durationMs,\n\t\t\t\t},\n\t\t\t\t...(failed ? { isError: true } : {}),\n\t\t\t};\n\t\t},\n\t};\n}\n"]}

package/dist/core/tools/run-toolkit-script.js

@@ -1,4 +1,5 @@
 import { Type } from "typebox";
+import { acceptReflexPlan } from "../toolkit/reflex-interpreter.js";
 import { matchToolkitScript } from "../toolkit/script-registry.js";
 const runToolkitScriptSchema = Type.Object({
     script: Type.String({
@@ -51,7 +52,20 @@ export function createRunToolkitScriptToolDefinition(deps) {
                     isError: true,
                 };
             }
-            if (match.kind === "ambiguous") {
+            let interpreted;
+            if (match.kind === "ambiguous" && deps.interpret) {
+                try {
+                    const plan = await deps.interpret(input.script, scripts);
+                    const accepted = acceptReflexPlan(plan, scripts);
+                    if (accepted && plan) {
+                        interpreted = { ...accepted, confidence: plan.confidence };
+                    }
+                }
+                catch {
+                    // interpreter is best-effort; ambiguity handling below stays authoritative
+                }
+            }
+            if (match.kind === "ambiguous" && !interpreted) {
                 const shortlist = match.shortlist.map((script) => `${script.name} — ${script.description}`);
                 return {
                     content: [
@@ -63,7 +77,7 @@ export function createRunToolkitScriptToolDefinition(deps) {
                     details: { outcome: "ambiguous", shortlist: match.shortlist.map((script) => script.name) },
                 };
             }
-            const script = match.script;
+            const script = interpreted?.script ?? match.script;
             if (script.danger && input.confirm !== true) {
                 return {
                     content: [
@@ -75,7 +89,8 @@ export function createRunToolkitScriptToolDefinition(deps) {
                     details: { outcome: "confirmation_required", scriptName: script.name },
                 };
             }
-            const execution = await deps.execute(script, input.args ?? []);
+            // Explicit args from the caller win; a brain-extracted arg list fills in for fuzzy requests.
+            const execution = await deps.execute(script, input.args ?? interpreted?.args ?? []);
             const failed = execution.exitCode !== 0 || execution.timedOut;
             const header = failed
                 ? `FAILED: ${script.name} exited ${execution.timedOut ? "by timeout" : execution.exitCode} after ${execution.durationMs}ms`
@@ -90,6 +105,9 @@ export function createRunToolkitScriptToolDefinition(deps) {
             return {
                 content: [{ type: "text", text: body }],
                 details: {
+                    ...(interpreted
+                        ? { interpreter: { script: interpreted.script.name, confidence: interpreted.confidence } }
+                        : {}),
                     outcome: failed ? "failed" : "executed",
                     scriptName: script.name,
                     exitCode: execution.exitCode,

package/dist/core/tools/run-toolkit-script.js.map

@@ -1 +1 @@
-{"version":3,"file":"run-toolkit-script.js","sourceRoot":"","sources":["../../../src/core/tools/run-toolkit-script.ts"],"names":[],"mappings":"AAAA,OAAO,EAAe,IAAI,EAAE,MAAM,SAAS,CAAC;AAE5C,OAAO,EAAE,kBAAkB,EAAsB,MAAM,+BAA+B,CAAC;AAGvF,MAAM,sBAAsB,GAAG,IAAI,CAAC,MAAM,CACzC;IACC,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC;QACnB,WAAW,EACV,iMAAiM;KAClM,CAAC;IACF,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE,WAAW,EAAE,uCAAuC,EAAE,CAAC,CAAC;IACxG,OAAO,EAAE,IAAI,CAAC,QAAQ,CACrB,IAAI,CAAC,OAAO,CAAC;QACZ,WAAW,EAAE,8FAA8F;KAC3G,CAAC,CACF;CACD,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAC/B,CAAC;AAiBF,SAAS,aAAa,CAAC,IAAY,EAAE,QAAQ,GAAG,IAAI,EAAU;IAC7D,OAAO,IAAI,CAAC,MAAM,IAAI,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,mBAAiB,CAAC;AAAA,CACpF;AAED,MAAM,UAAU,oCAAoC,CAAC,IAAkC,EAAkB;IACxG,OAAO;QACN,IAAI,EAAE,oBAAoB;QAC1B,KAAK,EAAE,oBAAoB;QAC3B,WAAW,EACV,6VAAyV;QAC1V,aAAa,EAAE,0FAA0F;QACzG,gBAAgB,EAAE;YACjB,uIAAuI;YACvI,8FAA4F;YAC5F,2FAA2F;SAC3F;QACD,UAAU,EAAE,sBAAsB;QAClC,KAAK,CAAC,OAAO,CACZ,WAAW,EACX,KAA4B,EAK1B;YACF,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;YAClC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC1B,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,uEAAuE;yBAC7E;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,gBAAgB,EAAE;iBACtC,CAAC;YACH,CAAC;YAED,MAAM,KAAK,GAAG,kBAAkB,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;YACxD,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAC3B,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,QAAM,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACnG,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,iDAAiD,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,eAAe,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE;yBACjH;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,WAAW,EAAE;oBACjC,OAAO,EAAE,IAAI;iBACb,CAAC;YACH,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;gBAChC,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,QAAM,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC;gBAC5F,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,wCAAsC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,mCAAmC;yBACnG;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;iBAC1F,CAAC;YACH,CAAC;YAED,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;YAC5B,IAAI,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;gBAC7C,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,IAAI,MAAM,CAAC,IAAI,2BAA2B,MAAM,CAAC,WAAW,+EAA+E;yBACjJ;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,uBAAuB,EAAE,UAAU,EAAE,MAAM,CAAC,IAAI,EAAE;iBACtE,CAAC;YACH,CAAC;YAED,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;YAC/D,MAAM,MAAM,GAAG,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM;gBACpB,CAAC,CAAC,WAAW,MAAM,CAAC,IAAI,WAAW,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,QAAQ,UAAU,SAAS,CAAC,UAAU,IAAI;gBAC3H,CAAC,CAAC,GAAG,MAAM,CAAC,IAAI,iBAAiB,SAAS,CAAC,UAAU,IAAI,CAAC;YAC3D,MAAM,IAAI,GAAG;gBACZ,MAAM;gBACN,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY,aAAa,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,iBAAiB;gBAClG,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY,aAAa,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;aACnF;iBACC,MAAM,CAAC,OAAO,CAAC;iBACf,IAAI,CAAC,IAAI,CAAC,CAAC;YAEb,OAAO;gBACN,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;gBAChD,OAAO,EAAE;oBACR,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU;oBACvC,UAAU,EAAE,MAAM,CAAC,IAAI;oBACvB,QAAQ,EAAE,SAAS,CAAC,QAAQ;oBAC5B,UAAU,EAAE,SAAS,CAAC,UAAU;iBAChC;gBACD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACpC,CAAC;QAAA,CACF;KACD,CAAC;AAAA,CACF","sourcesContent":["import { type Static, Type } from \"typebox\";\nimport type { ToolDefinition } from \"../extensions/types.ts\";\nimport { matchToolkitScript, type ToolkitScript } from \"../toolkit/script-registry.ts\";\nimport type { ScriptExecution } from \"../toolkit/script-runner.ts\";\n\nconst runToolkitScriptSchema = Type.Object(\n\t{\n\t\tscript: Type.String({\n\t\t\tdescription:\n\t\t\t\t\"The toolkit script to run: its registered name, a taught alias, or a natural request (e.g. 'restore-db' or 'restore the database'). Ambiguous requests return a shortlist instead of executing.\",\n\t\t}),\n\t\targs: Type.Optional(Type.Array(Type.String(), { description: \"Arguments passed to the script as-is.\" })),\n\t\tconfirm: Type.Optional(\n\t\t\tType.Boolean({\n\t\t\t\tdescription: \"Required true to run a script flagged danger: scripts marked dangerous never run without it.\",\n\t\t\t}),\n\t\t),\n\t},\n\t{ additionalProperties: false },\n);\n\nexport type RunToolkitScriptInput = Static<typeof runToolkitScriptSchema>;\n\nexport interface RunToolkitScriptDetails {\n\toutcome: \"executed\" | \"failed\" | \"ambiguous\" | \"not_found\" | \"confirmation_required\" | \"empty_registry\";\n\tscriptName?: string;\n\texitCode?: number | null;\n\tdurationMs?: number;\n\tshortlist?: string[];\n}\n\nexport interface RunToolkitScriptDependencies {\n\tgetScripts: () => ToolkitScript[];\n\texecute: (script: ToolkitScript, args: readonly string[]) => Promise<ScriptExecution>;\n}\n\nfunction boundedOutput(text: string, maxChars = 4000): string {\n\treturn text.length <= maxChars ? text : `${text.slice(0, maxChars)}\\n… (truncated)`;\n}\n\nexport function createRunToolkitScriptToolDefinition(deps: RunToolkitScriptDependencies): ToolDefinition {\n\treturn {\n\t\tname: \"run_toolkit_script\",\n\t\tlabel: \"run_toolkit_script\",\n\t\tdescription:\n\t\t\t\"Run one of the user's registered toolkit scripts (daily-ops allowlist). Finding is a registry lookup — ambiguous requests return a shortlist to disambiguate, never a guess. The harness executes (uv/powershell/bash) and ALWAYS returns exit code, stdout, and stderr; failures are reported as errors — a failed script can never look like success.\",\n\t\tpromptSnippet: \"Run a registered toolkit script by name/alias; output and errors come back structurally.\",\n\t\tpromptGuidelines: [\n\t\t\t\"Prefer the exact registered name; on a shortlist response, ask the user which one (or pick only when their request clearly names it).\",\n\t\t\t\"Scripts flagged dangerous require confirm: true — obtain explicit user confirmation first.\",\n\t\t\t\"Report the script's real output/exit code; never claim success when exitCode is non-zero.\",\n\t\t],\n\t\tparameters: runToolkitScriptSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\tinput: RunToolkitScriptInput,\n\t\t): Promise<{\n\t\t\tcontent: Array<{ type: \"text\"; text: string }>;\n\t\t\tdetails: RunToolkitScriptDetails;\n\t\t\tisError?: boolean;\n\t\t}> {\n\t\t\tconst scripts = deps.getScripts();\n\t\t\tif (scripts.length === 0) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: \"run_toolkit_script: no scripts registered (settings.toolkit.scripts).\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"empty_registry\" },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst match = matchToolkitScript(input.script, scripts);\n\t\t\tif (match.kind === \"none\") {\n\t\t\t\tconst closest = match.closest.map((script) => `${script.name} — ${script.description}`).join(\"\\n\");\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `script_not_found: nothing registered matches \"${input.script}\".${closest ? `\\nClosest:\\n${closest}` : \"\"}`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"not_found\" },\n\t\t\t\t\tisError: true,\n\t\t\t\t};\n\t\t\t}\n\t\t\tif (match.kind === \"ambiguous\") {\n\t\t\t\tconst shortlist = match.shortlist.map((script) => `${script.name} — ${script.description}`);\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `Ambiguous request — did you mean:\\n${shortlist.join(\"\\n\")}\\nCall again with the exact name.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"ambiguous\", shortlist: match.shortlist.map((script) => script.name) },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst script = match.script;\n\t\t\tif (script.danger && input.confirm !== true) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `\"${script.name}\" is flagged DANGEROUS (${script.description}). It was NOT run. Confirm with the user, then call again with confirm: true.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"confirmation_required\", scriptName: script.name },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst execution = await deps.execute(script, input.args ?? []);\n\t\t\tconst failed = execution.exitCode !== 0 || execution.timedOut;\n\t\t\tconst header = failed\n\t\t\t\t? `FAILED: ${script.name} exited ${execution.timedOut ? \"by timeout\" : execution.exitCode} after ${execution.durationMs}ms`\n\t\t\t\t: `${script.name} succeeded in ${execution.durationMs}ms`;\n\t\t\tconst body = [\n\t\t\t\theader,\n\t\t\t\texecution.stdout.trim() ? `stdout:\\n${boundedOutput(execution.stdout.trim())}` : \"stdout: (empty)\",\n\t\t\t\texecution.stderr.trim() ? `stderr:\\n${boundedOutput(execution.stderr.trim())}` : \"\",\n\t\t\t]\n\t\t\t\t.filter(Boolean)\n\t\t\t\t.join(\"\\n\");\n\n\t\t\treturn {\n\t\t\t\tcontent: [{ type: \"text\" as const, text: body }],\n\t\t\t\tdetails: {\n\t\t\t\t\toutcome: failed ? \"failed\" : \"executed\",\n\t\t\t\t\tscriptName: script.name,\n\t\t\t\t\texitCode: execution.exitCode,\n\t\t\t\t\tdurationMs: execution.durationMs,\n\t\t\t\t},\n\t\t\t\t...(failed ? { isError: true } : {}),\n\t\t\t};\n\t\t},\n\t};\n}\n"]}
+{"version":3,"file":"run-toolkit-script.js","sourceRoot":"","sources":["../../../src/core/tools/run-toolkit-script.ts"],"names":[],"mappings":"AAAA,OAAO,EAAe,IAAI,EAAE,MAAM,SAAS,CAAC;AAE5C,OAAO,EAAE,gBAAgB,EAAmB,MAAM,kCAAkC,CAAC;AACrF,OAAO,EAAE,kBAAkB,EAAsB,MAAM,+BAA+B,CAAC;AAGvF,MAAM,sBAAsB,GAAG,IAAI,CAAC,MAAM,CACzC;IACC,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC;QACnB,WAAW,EACV,iMAAiM;KAClM,CAAC;IACF,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE,WAAW,EAAE,uCAAuC,EAAE,CAAC,CAAC;IACxG,OAAO,EAAE,IAAI,CAAC,QAAQ,CACrB,IAAI,CAAC,OAAO,CAAC;QACZ,WAAW,EAAE,8FAA8F;KAC3G,CAAC,CACF;CACD,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAC/B,CAAC;AAwBF,SAAS,aAAa,CAAC,IAAY,EAAE,QAAQ,GAAG,IAAI,EAAU;IAC7D,OAAO,IAAI,CAAC,MAAM,IAAI,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,mBAAiB,CAAC;AAAA,CACpF;AAED,MAAM,UAAU,oCAAoC,CAAC,IAAkC,EAAkB;IACxG,OAAO;QACN,IAAI,EAAE,oBAAoB;QAC1B,KAAK,EAAE,oBAAoB;QAC3B,WAAW,EACV,6VAAyV;QAC1V,aAAa,EAAE,0FAA0F;QACzG,gBAAgB,EAAE;YACjB,uIAAuI;YACvI,8FAA4F;YAC5F,2FAA2F;SAC3F;QACD,UAAU,EAAE,sBAAsB;QAClC,KAAK,CAAC,OAAO,CACZ,WAAW,EACX,KAA4B,EAK1B;YACF,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;YAClC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC1B,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,uEAAuE;yBAC7E;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,gBAAgB,EAAE;iBACtC,CAAC;YACH,CAAC;YAED,MAAM,KAAK,GAAG,kBAAkB,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;YACxD,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAC3B,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,QAAM,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACnG,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,iDAAiD,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,eAAe,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE;yBACjH;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,WAAW,EAAE;oBACjC,OAAO,EAAE,IAAI;iBACb,CAAC;YACH,CAAC;YACD,IAAI,WAAsF,CAAC;YAC3F,IAAI,KAAK,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClD,IAAI,CAAC;oBACJ,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;oBACzD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;oBACjD,IAAI,QAAQ,IAAI,IAAI,EAAE,CAAC;wBACtB,WAAW,GAAG,EAAE,GAAG,QAAQ,EAAE,UAAU,EAAE,IAAI,CAAC,UAAU,EAAE,CAAC;oBAC5D,CAAC;gBACF,CAAC;gBAAC,MAAM,CAAC;oBACR,2EAA2E;gBAC5E,CAAC;YACF,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,WAAW,EAAE,CAAC;gBAChD,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,IAAI,QAAM,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC;gBAC5F,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,wCAAsC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,mCAAmC;yBACnG;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;iBAC1F,CAAC;YACH,CAAC;YAED,MAAM,MAAM,GAAG,WAAW,EAAE,MAAM,IAAK,KAAmC,CAAC,MAAM,CAAC;YAClF,IAAI,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;gBAC7C,OAAO;oBACN,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,MAAe;4BACrB,IAAI,EAAE,IAAI,MAAM,CAAC,IAAI,2BAA2B,MAAM,CAAC,WAAW,+EAA+E;yBACjJ;qBACD;oBACD,OAAO,EAAE,EAAE,OAAO,EAAE,uBAAuB,EAAE,UAAU,EAAE,MAAM,CAAC,IAAI,EAAE;iBACtE,CAAC;YACH,CAAC;YAED,6FAA6F;YAC7F,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,IAAI,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,CAAC;YACpF,MAAM,MAAM,GAAG,SAAS,CAAC,QAAQ,KAAK,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM;gBACpB,CAAC,CAAC,WAAW,MAAM,CAAC,IAAI,WAAW,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,QAAQ,UAAU,SAAS,CAAC,UAAU,IAAI;gBAC3H,CAAC,CAAC,GAAG,MAAM,CAAC,IAAI,iBAAiB,SAAS,CAAC,UAAU,IAAI,CAAC;YAC3D,MAAM,IAAI,GAAG;gBACZ,MAAM;gBACN,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY,aAAa,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,iBAAiB;gBAClG,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,YAAY,aAAa,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;aACnF;iBACC,MAAM,CAAC,OAAO,CAAC;iBACf,IAAI,CAAC,IAAI,CAAC,CAAC;YAEb,OAAO;gBACN,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;gBAChD,OAAO,EAAE;oBACR,GAAG,CAAC,WAAW;wBACd,CAAC,CAAC,EAAE,WAAW,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,CAAC,IAAI,EAAE,UAAU,EAAE,WAAW,CAAC,UAAU,EAAE,EAAE;wBAC1F,CAAC,CAAC,EAAE,CAAC;oBACN,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU;oBACvC,UAAU,EAAE,MAAM,CAAC,IAAI;oBACvB,QAAQ,EAAE,SAAS,CAAC,QAAQ;oBAC5B,UAAU,EAAE,SAAS,CAAC,UAAU;iBAChC;gBACD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACpC,CAAC;QAAA,CACF;KACD,CAAC;AAAA,CACF","sourcesContent":["import { type Static, Type } from \"typebox\";\nimport type { ToolDefinition } from \"../extensions/types.ts\";\nimport { acceptReflexPlan, type ReflexPlan } from \"../toolkit/reflex-interpreter.ts\";\nimport { matchToolkitScript, type ToolkitScript } from \"../toolkit/script-registry.ts\";\nimport type { ScriptExecution } from \"../toolkit/script-runner.ts\";\n\nconst runToolkitScriptSchema = Type.Object(\n\t{\n\t\tscript: Type.String({\n\t\t\tdescription:\n\t\t\t\t\"The toolkit script to run: its registered name, a taught alias, or a natural request (e.g. 'restore-db' or 'restore the database'). Ambiguous requests return a shortlist instead of executing.\",\n\t\t}),\n\t\targs: Type.Optional(Type.Array(Type.String(), { description: \"Arguments passed to the script as-is.\" })),\n\t\tconfirm: Type.Optional(\n\t\t\tType.Boolean({\n\t\t\t\tdescription: \"Required true to run a script flagged danger: scripts marked dangerous never run without it.\",\n\t\t\t}),\n\t\t),\n\t},\n\t{ additionalProperties: false },\n);\n\nexport type RunToolkitScriptInput = Static<typeof runToolkitScriptSchema>;\n\nexport interface RunToolkitScriptDetails {\n\toutcome: \"executed\" | \"failed\" | \"ambiguous\" | \"not_found\" | \"confirmation_required\" | \"empty_registry\";\n\tscriptName?: string;\n\texitCode?: number | null;\n\tdurationMs?: number;\n\tshortlist?: string[];\n}\n\nexport interface RunToolkitScriptDependencies {\n\tgetScripts: () => ToolkitScript[];\n\texecute: (script: ToolkitScript, args: readonly string[]) => Promise<ScriptExecution>;\n\t/**\n\t * Optional reflex interpreter (local brain): consulted ONLY when the deterministic Level-0\n\t * matcher is ambiguous, to resolve fuzzy phrasing (\"prepare db\" vs \"update db\") into a\n\t * registry pick. Its plan is advisory — the danger/confirm rules and structural execution\n\t * contract apply identically to brain-selected scripts.\n\t */\n\tinterpret?: (request: string, scripts: readonly ToolkitScript[]) => Promise<ReflexPlan | undefined>;\n}\n\nfunction boundedOutput(text: string, maxChars = 4000): string {\n\treturn text.length <= maxChars ? text : `${text.slice(0, maxChars)}\\n… (truncated)`;\n}\n\nexport function createRunToolkitScriptToolDefinition(deps: RunToolkitScriptDependencies): ToolDefinition {\n\treturn {\n\t\tname: \"run_toolkit_script\",\n\t\tlabel: \"run_toolkit_script\",\n\t\tdescription:\n\t\t\t\"Run one of the user's registered toolkit scripts (daily-ops allowlist). Finding is a registry lookup — ambiguous requests return a shortlist to disambiguate, never a guess. The harness executes (uv/powershell/bash) and ALWAYS returns exit code, stdout, and stderr; failures are reported as errors — a failed script can never look like success.\",\n\t\tpromptSnippet: \"Run a registered toolkit script by name/alias; output and errors come back structurally.\",\n\t\tpromptGuidelines: [\n\t\t\t\"Prefer the exact registered name; on a shortlist response, ask the user which one (or pick only when their request clearly names it).\",\n\t\t\t\"Scripts flagged dangerous require confirm: true — obtain explicit user confirmation first.\",\n\t\t\t\"Report the script's real output/exit code; never claim success when exitCode is non-zero.\",\n\t\t],\n\t\tparameters: runToolkitScriptSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\tinput: RunToolkitScriptInput,\n\t\t): Promise<{\n\t\t\tcontent: Array<{ type: \"text\"; text: string }>;\n\t\t\tdetails: RunToolkitScriptDetails;\n\t\t\tisError?: boolean;\n\t\t}> {\n\t\t\tconst scripts = deps.getScripts();\n\t\t\tif (scripts.length === 0) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: \"run_toolkit_script: no scripts registered (settings.toolkit.scripts).\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"empty_registry\" },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst match = matchToolkitScript(input.script, scripts);\n\t\t\tif (match.kind === \"none\") {\n\t\t\t\tconst closest = match.closest.map((script) => `${script.name} — ${script.description}`).join(\"\\n\");\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `script_not_found: nothing registered matches \"${input.script}\".${closest ? `\\nClosest:\\n${closest}` : \"\"}`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"not_found\" },\n\t\t\t\t\tisError: true,\n\t\t\t\t};\n\t\t\t}\n\t\t\tlet interpreted: { script: ToolkitScript; args: string[]; confidence: number } | undefined;\n\t\t\tif (match.kind === \"ambiguous\" && deps.interpret) {\n\t\t\t\ttry {\n\t\t\t\t\tconst plan = await deps.interpret(input.script, scripts);\n\t\t\t\t\tconst accepted = acceptReflexPlan(plan, scripts);\n\t\t\t\t\tif (accepted && plan) {\n\t\t\t\t\t\tinterpreted = { ...accepted, confidence: plan.confidence };\n\t\t\t\t\t}\n\t\t\t\t} catch {\n\t\t\t\t\t// interpreter is best-effort; ambiguity handling below stays authoritative\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (match.kind === \"ambiguous\" && !interpreted) {\n\t\t\t\tconst shortlist = match.shortlist.map((script) => `${script.name} — ${script.description}`);\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `Ambiguous request — did you mean:\\n${shortlist.join(\"\\n\")}\\nCall again with the exact name.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"ambiguous\", shortlist: match.shortlist.map((script) => script.name) },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst script = interpreted?.script ?? (match as { script: ToolkitScript }).script;\n\t\t\tif (script.danger && input.confirm !== true) {\n\t\t\t\treturn {\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\" as const,\n\t\t\t\t\t\t\ttext: `\"${script.name}\" is flagged DANGEROUS (${script.description}). It was NOT run. Confirm with the user, then call again with confirm: true.`,\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t\tdetails: { outcome: \"confirmation_required\", scriptName: script.name },\n\t\t\t\t};\n\t\t\t}\n\n\t\t\t// Explicit args from the caller win; a brain-extracted arg list fills in for fuzzy requests.\n\t\t\tconst execution = await deps.execute(script, input.args ?? interpreted?.args ?? []);\n\t\t\tconst failed = execution.exitCode !== 0 || execution.timedOut;\n\t\t\tconst header = failed\n\t\t\t\t? `FAILED: ${script.name} exited ${execution.timedOut ? \"by timeout\" : execution.exitCode} after ${execution.durationMs}ms`\n\t\t\t\t: `${script.name} succeeded in ${execution.durationMs}ms`;\n\t\t\tconst body = [\n\t\t\t\theader,\n\t\t\t\texecution.stdout.trim() ? `stdout:\\n${boundedOutput(execution.stdout.trim())}` : \"stdout: (empty)\",\n\t\t\t\texecution.stderr.trim() ? `stderr:\\n${boundedOutput(execution.stderr.trim())}` : \"\",\n\t\t\t]\n\t\t\t\t.filter(Boolean)\n\t\t\t\t.join(\"\\n\");\n\n\t\t\treturn {\n\t\t\t\tcontent: [{ type: \"text\" as const, text: body }],\n\t\t\t\tdetails: {\n\t\t\t\t\t...(interpreted\n\t\t\t\t\t\t? { interpreter: { script: interpreted.script.name, confidence: interpreted.confidence } }\n\t\t\t\t\t\t: {}),\n\t\t\t\t\toutcome: failed ? \"failed\" : \"executed\",\n\t\t\t\t\tscriptName: script.name,\n\t\t\t\t\texitCode: execution.exitCode,\n\t\t\t\t\tdurationMs: execution.durationMs,\n\t\t\t\t},\n\t\t\t\t...(failed ? { isError: true } : {}),\n\t\t\t};\n\t\t},\n\t};\n}\n"]}

package/dist/modes/interactive/components/settings-selector.d.ts

@@ -1,7 +1,7 @@
 import type { ThinkingLevel } from "@caupulican/pi-agent-core";
 import type { Transport } from "@caupulican/pi-ai";
 import { Container, type SelectItem, SelectList, SettingsList } from "@caupulican/pi-tui";
-import type { AutoLearnSettings, AutonomySettings, ContextCurationSettings, ContextPromptEnforcementSettings, MemoryRetrievalSettings, ModelRouterSettings, ResearchLaneSettings, SelfModificationSettings, SettingsScope, WarningSettings, WorkerDelegationSettings } from "../../../core/settings-manager.ts";
+import type { AutoLearnSettings, AutonomySettings, ContextCurationSettings, ContextPromptEnforcementSettings, LearningPolicySettings, MemoryRetrievalSettings, ModelCapabilitySettings, ModelRouterSettings, ResearchLaneSettings, SelfModificationSettings, SettingsScope, WarningSettings, WorkerDelegationSettings } from "../../../core/settings-manager.ts";
 export interface SettingsConfig {
     autoCompact: boolean;
     showImages: boolean;
@@ -42,6 +42,10 @@ export interface SettingsConfig {
     workerDelegation: WorkerDelegationSettings;
     contextCuration: ContextCurationSettings;
     contextCurationScope?: SettingsScope;
+    learningPolicy: LearningPolicySettings;
+    learningPolicyScope?: SettingsScope;
+    modelCapability: ModelCapabilitySettings;
+    modelCapabilityScope?: SettingsScope;
     workerDelegationScope?: SettingsScope;
     modelRouter: ModelRouterSettings;
     modelRouterScope?: SettingsScope;
@@ -89,6 +93,8 @@ export interface SettingsCallbacks {
     onResearchLaneChange: (settings: ResearchLaneSettings, scope: SettingsScope) => void;
     onWorkerDelegationChange: (settings: WorkerDelegationSettings, scope: SettingsScope) => void;
     onContextCurationChange: (settings: ContextCurationSettings, scope: SettingsScope) => void;
+    onLearningPolicyChange: (settings: LearningPolicySettings, scope: SettingsScope) => void;
+    onModelCapabilityChange: (settings: ModelCapabilitySettings, scope: SettingsScope) => void;
     onModelRouterChange: (settings: ModelRouterSettings, scope: SettingsScope) => void;
     onAutoLearnChange: (settings: AutoLearnSettings, scope: SettingsScope) => void;
     onContextPolicyEnforcementChange: (settings: ContextPromptEnforcementSettings, scope: SettingsScope) => void;