@caupulican/pi-adaptative 0.81.32 → 0.81.37

package/dist/core/model-capability.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"model-capability.d.ts","sourceRoot":"","sources":["../../src/core/model-capability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,MAAM,MAAM,oBAAoB,GAAG,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,CAAC;AAExE,MAAM,MAAM,mBAAmB,GAAG,MAAM,GAAG,KAAK,GAAG,oBAAoB,CAAC;AAExE,MAAM,WAAW,sBAAsB;IACtC,KAAK,EAAE,oBAAoB,CAAC;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACrC,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACrC,0FAA0F;IAC1F,sBAAsB,EAAE,OAAO,CAAC;IAChC,4EAA4E;IAC5E,mBAAmB,EAAE,MAAM,CAAC;CAC5B;AAED,8DAA8D;AAC9D,eAAO,MAAM,iCAAiC,QAAS,CAAC;AACxD,oFAAoF;AACpF,eAAO,MAAM,iCAAiC,QAAS,CAAC;AACxD,+EAA+E;AAC/E,eAAO,MAAM,oCAAoC,OAAQ,CAAC;AAE1D,eAAO,MAAM,mCAAmC,EAAE,SAAS,MAAM,EAAkC,CAAC;AACpG,eAAO,MAAM,sCAAsC,EAAE,SAAS,MAAM,EAOnE,CAAC;AACF,eAAO,MAAM,mCAAmC,EAAE,SAAS,MAAM,EAAO,CAAC;AAEzE,eAAO,MAAM,8BAA8B,OAAO,CAAC;AA0CnD,wBAAgB,4BAA4B,CAAC,IAAI,EAAE;IAClD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,IAAI,CAAC,EAAE,mBAAmB,CAAC;CAC3B,GAAG,sBAAsB,CA2BzB;AAED,uFAAuF;AACvF,MAAM,WAAW,mBAAmB;IACnC,uDAAuD;IACvD,QAAQ,EAAE,MAAM,CAAC;IACjB,8EAA8E;IAC9E,mBAAmB,EAAE,MAAM,CAAC;CAC5B;AAED,6FAA6F;AAC7F,eAAO,MAAM,wCAAwC,IAAI,CAAC;AAC1D,eAAO,MAAM,qDAAqD,IAAI,CAAC;AAEvE;;;;;;;GAOG;AACH,wBAAgB,qCAAqC,CACpD,OAAO,EAAE,sBAAsB,EAC/B,OAAO,EAAE,mBAAmB,GAC1B,mBAAmB,CAMrB;AAED,6FAA6F;AAC7F,wBAAgB,4BAA4B,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,OAAO,EAAE,sBAAsB,GAAG,MAAM,EAAE,CAWpH","sourcesContent":["/**\n * Model capability auto-detection: derive what the harness may load onto a model FROM the model's\n * own metadata (`Model.contextWindow`), so small open models (4k/8k/16k windows, sub-1B params)\n * can still hold a usable chat instead of drowning in tool schemas and background-lane prompts.\n *\n * Derivation is metadata-first; defaults apply only when the metadata is missing (unknown/zero\n * window keeps today's full behavior rather than guessing). Detection can be disabled or forced\n * per class via the `modelCapability.mode` setting.\n */\n\nexport type ModelCapabilityClass = \"full\" | \"lean\" | \"minimal\" | \"chat\";\n\nexport type ModelCapabilityMode = \"auto\" | \"off\" | ModelCapabilityClass;\n\nexport interface ModelCapabilityProfile {\n\tclass: ModelCapabilityClass;\n\tcontextWindow?: number;\n\treasonCode: string;\n\t/** Allow-list; undefined = no allow-list restriction. */\n\tallowedToolNames?: readonly string[];\n\t/** Block-list applied after the allow-list; undefined = nothing blocked. */\n\tblockedToolNames?: readonly string[];\n\t/** Whether idle background lanes (goal auto-continue, research) may run on this model. */\n\tbackgroundLanesEnabled: boolean;\n\t/** Output-token cap for lane isolated completions, scaled to the window. */\n\tlaneMaxOutputTokens: number;\n}\n\n/** Windows at or above this keep the full harness surface. */\nexport const MODEL_CAPABILITY_FULL_MIN_CONTEXT = 32_768;\n/** Windows at or above this keep core tools but shed background-autonomy extras. */\nexport const MODEL_CAPABILITY_LEAN_MIN_CONTEXT = 16_384;\n/** Windows at or above this get the minimal coding set; below is chat-only. */\nexport const MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT = 8_192;\n\nexport const MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS: readonly string[] = [\"delegate\", \"context_audit\"];\nexport const MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS: readonly string[] = [\n\t\"read\",\n\t\"bash\",\n\t\"edit\",\n\t\"write\",\n\t// The executor tool: minimal-class models ARE the daily-ops executors, and its schema is tiny.\n\t\"run_toolkit_script\",\n];\nexport const MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS: readonly string[] = [];\n\nexport const DEFAULT_LANE_MAX_OUTPUT_TOKENS = 2048;\nconst MIN_LANE_MAX_OUTPUT_TOKENS = 256;\n\nfunction laneOutputTokensForWindow(contextWindow: number | undefined): number {\n\tif (contextWindow === undefined || contextWindow <= 0) return DEFAULT_LANE_MAX_OUTPUT_TOKENS;\n\t// A lane completion may use at most an eighth of the window for output, floored so tiny\n\t// windows still produce something parseable.\n\treturn Math.min(DEFAULT_LANE_MAX_OUTPUT_TOKENS, Math.max(MIN_LANE_MAX_OUTPUT_TOKENS, Math.floor(contextWindow / 8)));\n}\n\nfunction profileForClass(\n\tcapabilityClass: ModelCapabilityClass,\n\treasonCode: string,\n\tcontextWindow: number | undefined,\n): ModelCapabilityProfile {\n\tconst base = {\n\t\tclass: capabilityClass,\n\t\treasonCode,\n\t\tbackgroundLanesEnabled: true,\n\t\tlaneMaxOutputTokens: laneOutputTokensForWindow(contextWindow),\n\t\t...(contextWindow !== undefined && contextWindow > 0 ? { contextWindow } : {}),\n\t};\n\tswitch (capabilityClass) {\n\t\tcase \"full\":\n\t\t\treturn base;\n\t\tcase \"lean\":\n\t\t\treturn { ...base, blockedToolNames: MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS };\n\t\tcase \"minimal\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t\tcase \"chat\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t}\n}\n\nexport function deriveModelCapabilityProfile(args: {\n\tcontextWindow?: number;\n\tmode?: ModelCapabilityMode;\n}): ModelCapabilityProfile {\n\tconst mode = args.mode ?? \"auto\";\n\tconst contextWindow =\n\t\targs.contextWindow !== undefined && Number.isFinite(args.contextWindow) && args.contextWindow > 0\n\t\t\t? args.contextWindow\n\t\t\t: undefined;\n\tif (mode === \"off\") {\n\t\treturn profileForClass(\"full\", \"detection_disabled\", contextWindow);\n\t}\n\tif (mode !== \"auto\") {\n\t\treturn profileForClass(mode, \"forced_by_setting\", contextWindow);\n\t}\n\n\tif (contextWindow === undefined) {\n\t\t// Metadata missing: defaults, never guesses.\n\t\treturn profileForClass(\"full\", \"unknown_context_window_defaults\", undefined);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_FULL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"full\", \"large_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_LEAN_MIN_CONTEXT) {\n\t\treturn profileForClass(\"lean\", \"lean_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"minimal\", \"minimal_context_window\", contextWindow);\n\t}\n\treturn profileForClass(\"chat\", \"chat_only_context_window\", contextWindow);\n}\n\n/** Goal-continuation (autosteer) budgets, scaled to the session's capability class. */\nexport interface ContinuationBudgets {\n\t/** Maximum continuation prompts per idle goal loop. */\n\tmaxTurns: number;\n\t/** Wall-clock budget in minutes; 0 means \"disabled\" (upstream convention). */\n\tmaxWallClockMinutes: number;\n}\n\n/** Lean-class continuation caps: a 16-32k window cannot afford the full autosteer budget. */\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS = 2;\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES = 5;\n\n/**\n * Scale goal-continuation budgets to the model's capability class. Lean-window models (16-32k) keep\n * autonomy but at a reduced budget; every other class passes the configured budget through unchanged\n * (full stays full; minimal/chat never reach here — their background lanes are disabled upstream).\n *\n * Both dimensions are a straight `min(configured, cap)`: a disabled wall-clock budget (0) stays\n * disabled because `min(0, cap) === 0`, so the cap only ever tightens an already-positive budget.\n */\nexport function scaleContinuationBudgetsForCapability(\n\tprofile: ModelCapabilityProfile,\n\tbudgets: ContinuationBudgets,\n): ContinuationBudgets {\n\tif (profile.class !== \"lean\") return budgets;\n\treturn {\n\t\tmaxTurns: Math.min(budgets.maxTurns, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS),\n\t\tmaxWallClockMinutes: Math.min(budgets.maxWallClockMinutes, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES),\n\t};\n}\n\n/** Apply the profile's allow/block lists to a requested tool-name list, preserving order. */\nexport function filterToolNamesForCapability(toolNames: readonly string[], profile: ModelCapabilityProfile): string[] {\n\tlet filtered = [...toolNames];\n\tif (profile.allowedToolNames !== undefined) {\n\t\tconst allowed = new Set(profile.allowedToolNames);\n\t\tfiltered = filtered.filter((name) => allowed.has(name));\n\t}\n\tif (profile.blockedToolNames !== undefined) {\n\t\tconst blocked = new Set(profile.blockedToolNames);\n\t\tfiltered = filtered.filter((name) => !blocked.has(name));\n\t}\n\treturn filtered;\n}\n"]}
+{"version":3,"file":"model-capability.d.ts","sourceRoot":"","sources":["../../src/core/model-capability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,MAAM,MAAM,oBAAoB,GAAG,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,CAAC;AAExE,MAAM,MAAM,mBAAmB,GAAG,MAAM,GAAG,KAAK,GAAG,oBAAoB,CAAC;AAExE,MAAM,WAAW,sBAAsB;IACtC,KAAK,EAAE,oBAAoB,CAAC;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACrC,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACrC,0FAA0F;IAC1F,sBAAsB,EAAE,OAAO,CAAC;IAChC,4EAA4E;IAC5E,mBAAmB,EAAE,MAAM,CAAC;CAC5B;AAED,8DAA8D;AAC9D,eAAO,MAAM,iCAAiC,QAAS,CAAC;AACxD,oFAAoF;AACpF,eAAO,MAAM,iCAAiC,QAAS,CAAC;AACxD,+EAA+E;AAC/E,eAAO,MAAM,oCAAoC,OAAQ,CAAC;AAE1D,eAAO,MAAM,mCAAmC,EAAE,SAAS,MAAM,EAAkC,CAAC;AACpG,eAAO,MAAM,sCAAsC,EAAE,SAAS,MAAM,EASnE,CAAC;AACF,eAAO,MAAM,mCAAmC,EAAE,SAAS,MAAM,EAAO,CAAC;AAEzE,eAAO,MAAM,8BAA8B,OAAO,CAAC;AA0CnD,wBAAgB,4BAA4B,CAAC,IAAI,EAAE;IAClD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,IAAI,CAAC,EAAE,mBAAmB,CAAC;CAC3B,GAAG,sBAAsB,CA2BzB;AAED,uFAAuF;AACvF,MAAM,WAAW,mBAAmB;IACnC,uDAAuD;IACvD,QAAQ,EAAE,MAAM,CAAC;IACjB,8EAA8E;IAC9E,mBAAmB,EAAE,MAAM,CAAC;CAC5B;AAED,6FAA6F;AAC7F,eAAO,MAAM,wCAAwC,IAAI,CAAC;AAC1D,eAAO,MAAM,qDAAqD,IAAI,CAAC;AAEvE;;;;;;;GAOG;AACH,wBAAgB,qCAAqC,CACpD,OAAO,EAAE,sBAAsB,EAC/B,OAAO,EAAE,mBAAmB,GAC1B,mBAAmB,CAMrB;AAED,6FAA6F;AAC7F,wBAAgB,4BAA4B,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,OAAO,EAAE,sBAAsB,GAAG,MAAM,EAAE,CAWpH","sourcesContent":["/**\n * Model capability auto-detection: derive what the harness may load onto a model FROM the model's\n * own metadata (`Model.contextWindow`), so small open models (4k/8k/16k windows, sub-1B params)\n * can still hold a usable chat instead of drowning in tool schemas and background-lane prompts.\n *\n * Derivation is metadata-first; defaults apply only when the metadata is missing (unknown/zero\n * window keeps today's full behavior rather than guessing). Detection can be disabled or forced\n * per class via the `modelCapability.mode` setting.\n */\n\nexport type ModelCapabilityClass = \"full\" | \"lean\" | \"minimal\" | \"chat\";\n\nexport type ModelCapabilityMode = \"auto\" | \"off\" | ModelCapabilityClass;\n\nexport interface ModelCapabilityProfile {\n\tclass: ModelCapabilityClass;\n\tcontextWindow?: number;\n\treasonCode: string;\n\t/** Allow-list; undefined = no allow-list restriction. */\n\tallowedToolNames?: readonly string[];\n\t/** Block-list applied after the allow-list; undefined = nothing blocked. */\n\tblockedToolNames?: readonly string[];\n\t/** Whether idle background lanes (goal auto-continue, research) may run on this model. */\n\tbackgroundLanesEnabled: boolean;\n\t/** Output-token cap for lane isolated completions, scaled to the window. */\n\tlaneMaxOutputTokens: number;\n}\n\n/** Windows at or above this keep the full harness surface. */\nexport const MODEL_CAPABILITY_FULL_MIN_CONTEXT = 32_768;\n/** Windows at or above this keep core tools but shed background-autonomy extras. */\nexport const MODEL_CAPABILITY_LEAN_MIN_CONTEXT = 16_384;\n/** Windows at or above this get the minimal coding set; below is chat-only. */\nexport const MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT = 8_192;\n\nexport const MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS: readonly string[] = [\"delegate\", \"context_audit\"];\nexport const MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS: readonly string[] = [\n\t\"read\",\n\t\"bash\",\n\t\"python\",\n\t\"powershell\",\n\t\"edit\",\n\t\"write\",\n\t// The executor tool: minimal-class models ARE the daily-ops executors, and its schema is tiny.\n\t\"run_toolkit_script\",\n];\nexport const MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS: readonly string[] = [];\n\nexport const DEFAULT_LANE_MAX_OUTPUT_TOKENS = 2048;\nconst MIN_LANE_MAX_OUTPUT_TOKENS = 256;\n\nfunction laneOutputTokensForWindow(contextWindow: number | undefined): number {\n\tif (contextWindow === undefined || contextWindow <= 0) return DEFAULT_LANE_MAX_OUTPUT_TOKENS;\n\t// A lane completion may use at most an eighth of the window for output, floored so tiny\n\t// windows still produce something parseable.\n\treturn Math.min(DEFAULT_LANE_MAX_OUTPUT_TOKENS, Math.max(MIN_LANE_MAX_OUTPUT_TOKENS, Math.floor(contextWindow / 8)));\n}\n\nfunction profileForClass(\n\tcapabilityClass: ModelCapabilityClass,\n\treasonCode: string,\n\tcontextWindow: number | undefined,\n): ModelCapabilityProfile {\n\tconst base = {\n\t\tclass: capabilityClass,\n\t\treasonCode,\n\t\tbackgroundLanesEnabled: true,\n\t\tlaneMaxOutputTokens: laneOutputTokensForWindow(contextWindow),\n\t\t...(contextWindow !== undefined && contextWindow > 0 ? { contextWindow } : {}),\n\t};\n\tswitch (capabilityClass) {\n\t\tcase \"full\":\n\t\t\treturn base;\n\t\tcase \"lean\":\n\t\t\treturn { ...base, blockedToolNames: MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS };\n\t\tcase \"minimal\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t\tcase \"chat\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t}\n}\n\nexport function deriveModelCapabilityProfile(args: {\n\tcontextWindow?: number;\n\tmode?: ModelCapabilityMode;\n}): ModelCapabilityProfile {\n\tconst mode = args.mode ?? \"auto\";\n\tconst contextWindow =\n\t\targs.contextWindow !== undefined && Number.isFinite(args.contextWindow) && args.contextWindow > 0\n\t\t\t? args.contextWindow\n\t\t\t: undefined;\n\tif (mode === \"off\") {\n\t\treturn profileForClass(\"full\", \"detection_disabled\", contextWindow);\n\t}\n\tif (mode !== \"auto\") {\n\t\treturn profileForClass(mode, \"forced_by_setting\", contextWindow);\n\t}\n\n\tif (contextWindow === undefined) {\n\t\t// Metadata missing: defaults, never guesses.\n\t\treturn profileForClass(\"full\", \"unknown_context_window_defaults\", undefined);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_FULL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"full\", \"large_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_LEAN_MIN_CONTEXT) {\n\t\treturn profileForClass(\"lean\", \"lean_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"minimal\", \"minimal_context_window\", contextWindow);\n\t}\n\treturn profileForClass(\"chat\", \"chat_only_context_window\", contextWindow);\n}\n\n/** Goal-continuation (autosteer) budgets, scaled to the session's capability class. */\nexport interface ContinuationBudgets {\n\t/** Maximum continuation prompts per idle goal loop. */\n\tmaxTurns: number;\n\t/** Wall-clock budget in minutes; 0 means \"disabled\" (upstream convention). */\n\tmaxWallClockMinutes: number;\n}\n\n/** Lean-class continuation caps: a 16-32k window cannot afford the full autosteer budget. */\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS = 2;\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES = 5;\n\n/**\n * Scale goal-continuation budgets to the model's capability class. Lean-window models (16-32k) keep\n * autonomy but at a reduced budget; every other class passes the configured budget through unchanged\n * (full stays full; minimal/chat never reach here — their background lanes are disabled upstream).\n *\n * Both dimensions are a straight `min(configured, cap)`: a disabled wall-clock budget (0) stays\n * disabled because `min(0, cap) === 0`, so the cap only ever tightens an already-positive budget.\n */\nexport function scaleContinuationBudgetsForCapability(\n\tprofile: ModelCapabilityProfile,\n\tbudgets: ContinuationBudgets,\n): ContinuationBudgets {\n\tif (profile.class !== \"lean\") return budgets;\n\treturn {\n\t\tmaxTurns: Math.min(budgets.maxTurns, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS),\n\t\tmaxWallClockMinutes: Math.min(budgets.maxWallClockMinutes, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES),\n\t};\n}\n\n/** Apply the profile's allow/block lists to a requested tool-name list, preserving order. */\nexport function filterToolNamesForCapability(toolNames: readonly string[], profile: ModelCapabilityProfile): string[] {\n\tlet filtered = [...toolNames];\n\tif (profile.allowedToolNames !== undefined) {\n\t\tconst allowed = new Set(profile.allowedToolNames);\n\t\tfiltered = filtered.filter((name) => allowed.has(name));\n\t}\n\tif (profile.blockedToolNames !== undefined) {\n\t\tconst blocked = new Set(profile.blockedToolNames);\n\t\tfiltered = filtered.filter((name) => !blocked.has(name));\n\t}\n\treturn filtered;\n}\n"]}

package/dist/core/model-capability.js

@@ -17,6 +17,8 @@ export const MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS = ["delegate", "context_audit"]
 export const MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS = [
     "read",
     "bash",
+    "python",
+    "powershell",
     "edit",
     "write",
     // The executor tool: minimal-class models ARE the daily-ops executors, and its schema is tiny.

package/dist/core/model-capability.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-capability.js","sourceRoot":"","sources":["../../src/core/model-capability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAoBH,8DAA8D;AAC9D,MAAM,CAAC,MAAM,iCAAiC,GAAG,MAAM,CAAC;AACxD,oFAAoF;AACpF,MAAM,CAAC,MAAM,iCAAiC,GAAG,MAAM,CAAC;AACxD,+EAA+E;AAC/E,MAAM,CAAC,MAAM,oCAAoC,GAAG,KAAK,CAAC;AAE1D,MAAM,CAAC,MAAM,mCAAmC,GAAsB,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;AACpG,MAAM,CAAC,MAAM,sCAAsC,GAAsB;IACxE,MAAM;IACN,MAAM;IACN,MAAM;IACN,OAAO;IACP,+FAA+F;IAC/F,oBAAoB;CACpB,CAAC;AACF,MAAM,CAAC,MAAM,mCAAmC,GAAsB,EAAE,CAAC;AAEzE,MAAM,CAAC,MAAM,8BAA8B,GAAG,IAAI,CAAC;AACnD,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAEvC,SAAS,yBAAyB,CAAC,aAAiC,EAAU;IAC7E,IAAI,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC;QAAE,OAAO,8BAA8B,CAAC;IAC7F,wFAAwF;IACxF,6CAA6C;IAC7C,OAAO,IAAI,CAAC,GAAG,CAAC,8BAA8B,EAAE,IAAI,CAAC,GAAG,CAAC,0BAA0B,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;AAAA,CACrH;AAED,SAAS,eAAe,CACvB,eAAqC,EACrC,UAAkB,EAClB,aAAiC,EACR;IACzB,MAAM,IAAI,GAAG;QACZ,KAAK,EAAE,eAAe;QACtB,UAAU;QACV,sBAAsB,EAAE,IAAI;QAC5B,mBAAmB,EAAE,yBAAyB,CAAC,aAAa,CAAC;QAC7D,GAAG,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC9E,CAAC;IACF,QAAQ,eAAe,EAAE,CAAC;QACzB,KAAK,MAAM;YACV,OAAO,IAAI,CAAC;QACb,KAAK,MAAM;YACV,OAAO,EAAE,GAAG,IAAI,EAAE,gBAAgB,EAAE,mCAAmC,EAAE,CAAC;QAC3E,KAAK,SAAS;YACb,OAAO;gBACN,GAAG,IAAI;gBACP,gBAAgB,EAAE,sCAAsC;gBACxD,sBAAsB,EAAE,KAAK;aAC7B,CAAC;QACH,KAAK,MAAM;YACV,OAAO;gBACN,GAAG,IAAI;gBACP,gBAAgB,EAAE,mCAAmC;gBACrD,sBAAsB,EAAE,KAAK;aAC7B,CAAC;IACJ,CAAC;AAAA,CACD;AAED,MAAM,UAAU,4BAA4B,CAAC,IAG5C,EAA0B;IAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC;IACjC,MAAM,aAAa,GAClB,IAAI,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,IAAI,CAAC,aAAa,GAAG,CAAC;QAChG,CAAC,CAAC,IAAI,CAAC,aAAa;QACpB,CAAC,CAAC,SAAS,CAAC;IACd,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,OAAO,eAAe,CAAC,MAAM,EAAE,oBAAoB,EAAE,aAAa,CAAC,CAAC;IACrE,CAAC;IACD,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QACrB,OAAO,eAAe,CAAC,IAAI,EAAE,mBAAmB,EAAE,aAAa,CAAC,CAAC;IAClE,CAAC;IAED,IAAI,aAAa,KAAK,SAAS,EAAE,CAAC;QACjC,6CAA6C;QAC7C,OAAO,eAAe,CAAC,MAAM,EAAE,iCAAiC,EAAE,SAAS,CAAC,CAAC;IAC9E,CAAC;IACD,IAAI,aAAa,IAAI,iCAAiC,EAAE,CAAC;QACxD,OAAO,eAAe,CAAC,MAAM,EAAE,sBAAsB,EAAE,aAAa,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,aAAa,IAAI,iCAAiC,EAAE,CAAC;QACxD,OAAO,eAAe,CAAC,MAAM,EAAE,qBAAqB,EAAE,aAAa,CAAC,CAAC;IACtE,CAAC;IACD,IAAI,aAAa,IAAI,oCAAoC,EAAE,CAAC;QAC3D,OAAO,eAAe,CAAC,SAAS,EAAE,wBAAwB,EAAE,aAAa,CAAC,CAAC;IAC5E,CAAC;IACD,OAAO,eAAe,CAAC,MAAM,EAAE,0BAA0B,EAAE,aAAa,CAAC,CAAC;AAAA,CAC1E;AAUD,6FAA6F;AAC7F,MAAM,CAAC,MAAM,wCAAwC,GAAG,CAAC,CAAC;AAC1D,MAAM,CAAC,MAAM,qDAAqD,GAAG,CAAC,CAAC;AAEvE;;;;;;;GAOG;AACH,MAAM,UAAU,qCAAqC,CACpD,OAA+B,EAC/B,OAA4B,EACN;IACtB,IAAI,OAAO,CAAC,KAAK,KAAK,MAAM;QAAE,OAAO,OAAO,CAAC;IAC7C,OAAO;QACN,QAAQ,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,EAAE,wCAAwC,CAAC;QAC9E,mBAAmB,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,mBAAmB,EAAE,qDAAqD,CAAC;KACjH,CAAC;AAAA,CACF;AAED,6FAA6F;AAC7F,MAAM,UAAU,4BAA4B,CAAC,SAA4B,EAAE,OAA+B,EAAY;IACrH,IAAI,QAAQ,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IAC9B,IAAI,OAAO,CAAC,gBAAgB,KAAK,SAAS,EAAE,CAAC;QAC5C,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAClD,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IACzD,CAAC;IACD,IAAI,OAAO,CAAC,gBAAgB,KAAK,SAAS,EAAE,CAAC;QAC5C,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAClD,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO,QAAQ,CAAC;AAAA,CAChB","sourcesContent":["/**\n * Model capability auto-detection: derive what the harness may load onto a model FROM the model's\n * own metadata (`Model.contextWindow`), so small open models (4k/8k/16k windows, sub-1B params)\n * can still hold a usable chat instead of drowning in tool schemas and background-lane prompts.\n *\n * Derivation is metadata-first; defaults apply only when the metadata is missing (unknown/zero\n * window keeps today's full behavior rather than guessing). Detection can be disabled or forced\n * per class via the `modelCapability.mode` setting.\n */\n\nexport type ModelCapabilityClass = \"full\" | \"lean\" | \"minimal\" | \"chat\";\n\nexport type ModelCapabilityMode = \"auto\" | \"off\" | ModelCapabilityClass;\n\nexport interface ModelCapabilityProfile {\n\tclass: ModelCapabilityClass;\n\tcontextWindow?: number;\n\treasonCode: string;\n\t/** Allow-list; undefined = no allow-list restriction. */\n\tallowedToolNames?: readonly string[];\n\t/** Block-list applied after the allow-list; undefined = nothing blocked. */\n\tblockedToolNames?: readonly string[];\n\t/** Whether idle background lanes (goal auto-continue, research) may run on this model. */\n\tbackgroundLanesEnabled: boolean;\n\t/** Output-token cap for lane isolated completions, scaled to the window. */\n\tlaneMaxOutputTokens: number;\n}\n\n/** Windows at or above this keep the full harness surface. */\nexport const MODEL_CAPABILITY_FULL_MIN_CONTEXT = 32_768;\n/** Windows at or above this keep core tools but shed background-autonomy extras. */\nexport const MODEL_CAPABILITY_LEAN_MIN_CONTEXT = 16_384;\n/** Windows at or above this get the minimal coding set; below is chat-only. */\nexport const MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT = 8_192;\n\nexport const MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS: readonly string[] = [\"delegate\", \"context_audit\"];\nexport const MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS: readonly string[] = [\n\t\"read\",\n\t\"bash\",\n\t\"edit\",\n\t\"write\",\n\t// The executor tool: minimal-class models ARE the daily-ops executors, and its schema is tiny.\n\t\"run_toolkit_script\",\n];\nexport const MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS: readonly string[] = [];\n\nexport const DEFAULT_LANE_MAX_OUTPUT_TOKENS = 2048;\nconst MIN_LANE_MAX_OUTPUT_TOKENS = 256;\n\nfunction laneOutputTokensForWindow(contextWindow: number | undefined): number {\n\tif (contextWindow === undefined || contextWindow <= 0) return DEFAULT_LANE_MAX_OUTPUT_TOKENS;\n\t// A lane completion may use at most an eighth of the window for output, floored so tiny\n\t// windows still produce something parseable.\n\treturn Math.min(DEFAULT_LANE_MAX_OUTPUT_TOKENS, Math.max(MIN_LANE_MAX_OUTPUT_TOKENS, Math.floor(contextWindow / 8)));\n}\n\nfunction profileForClass(\n\tcapabilityClass: ModelCapabilityClass,\n\treasonCode: string,\n\tcontextWindow: number | undefined,\n): ModelCapabilityProfile {\n\tconst base = {\n\t\tclass: capabilityClass,\n\t\treasonCode,\n\t\tbackgroundLanesEnabled: true,\n\t\tlaneMaxOutputTokens: laneOutputTokensForWindow(contextWindow),\n\t\t...(contextWindow !== undefined && contextWindow > 0 ? { contextWindow } : {}),\n\t};\n\tswitch (capabilityClass) {\n\t\tcase \"full\":\n\t\t\treturn base;\n\t\tcase \"lean\":\n\t\t\treturn { ...base, blockedToolNames: MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS };\n\t\tcase \"minimal\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t\tcase \"chat\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t}\n}\n\nexport function deriveModelCapabilityProfile(args: {\n\tcontextWindow?: number;\n\tmode?: ModelCapabilityMode;\n}): ModelCapabilityProfile {\n\tconst mode = args.mode ?? \"auto\";\n\tconst contextWindow =\n\t\targs.contextWindow !== undefined && Number.isFinite(args.contextWindow) && args.contextWindow > 0\n\t\t\t? args.contextWindow\n\t\t\t: undefined;\n\tif (mode === \"off\") {\n\t\treturn profileForClass(\"full\", \"detection_disabled\", contextWindow);\n\t}\n\tif (mode !== \"auto\") {\n\t\treturn profileForClass(mode, \"forced_by_setting\", contextWindow);\n\t}\n\n\tif (contextWindow === undefined) {\n\t\t// Metadata missing: defaults, never guesses.\n\t\treturn profileForClass(\"full\", \"unknown_context_window_defaults\", undefined);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_FULL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"full\", \"large_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_LEAN_MIN_CONTEXT) {\n\t\treturn profileForClass(\"lean\", \"lean_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"minimal\", \"minimal_context_window\", contextWindow);\n\t}\n\treturn profileForClass(\"chat\", \"chat_only_context_window\", contextWindow);\n}\n\n/** Goal-continuation (autosteer) budgets, scaled to the session's capability class. */\nexport interface ContinuationBudgets {\n\t/** Maximum continuation prompts per idle goal loop. */\n\tmaxTurns: number;\n\t/** Wall-clock budget in minutes; 0 means \"disabled\" (upstream convention). */\n\tmaxWallClockMinutes: number;\n}\n\n/** Lean-class continuation caps: a 16-32k window cannot afford the full autosteer budget. */\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS = 2;\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES = 5;\n\n/**\n * Scale goal-continuation budgets to the model's capability class. Lean-window models (16-32k) keep\n * autonomy but at a reduced budget; every other class passes the configured budget through unchanged\n * (full stays full; minimal/chat never reach here — their background lanes are disabled upstream).\n *\n * Both dimensions are a straight `min(configured, cap)`: a disabled wall-clock budget (0) stays\n * disabled because `min(0, cap) === 0`, so the cap only ever tightens an already-positive budget.\n */\nexport function scaleContinuationBudgetsForCapability(\n\tprofile: ModelCapabilityProfile,\n\tbudgets: ContinuationBudgets,\n): ContinuationBudgets {\n\tif (profile.class !== \"lean\") return budgets;\n\treturn {\n\t\tmaxTurns: Math.min(budgets.maxTurns, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS),\n\t\tmaxWallClockMinutes: Math.min(budgets.maxWallClockMinutes, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES),\n\t};\n}\n\n/** Apply the profile's allow/block lists to a requested tool-name list, preserving order. */\nexport function filterToolNamesForCapability(toolNames: readonly string[], profile: ModelCapabilityProfile): string[] {\n\tlet filtered = [...toolNames];\n\tif (profile.allowedToolNames !== undefined) {\n\t\tconst allowed = new Set(profile.allowedToolNames);\n\t\tfiltered = filtered.filter((name) => allowed.has(name));\n\t}\n\tif (profile.blockedToolNames !== undefined) {\n\t\tconst blocked = new Set(profile.blockedToolNames);\n\t\tfiltered = filtered.filter((name) => !blocked.has(name));\n\t}\n\treturn filtered;\n}\n"]}
+{"version":3,"file":"model-capability.js","sourceRoot":"","sources":["../../src/core/model-capability.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAoBH,8DAA8D;AAC9D,MAAM,CAAC,MAAM,iCAAiC,GAAG,MAAM,CAAC;AACxD,oFAAoF;AACpF,MAAM,CAAC,MAAM,iCAAiC,GAAG,MAAM,CAAC;AACxD,+EAA+E;AAC/E,MAAM,CAAC,MAAM,oCAAoC,GAAG,KAAK,CAAC;AAE1D,MAAM,CAAC,MAAM,mCAAmC,GAAsB,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;AACpG,MAAM,CAAC,MAAM,sCAAsC,GAAsB;IACxE,MAAM;IACN,MAAM;IACN,QAAQ;IACR,YAAY;IACZ,MAAM;IACN,OAAO;IACP,+FAA+F;IAC/F,oBAAoB;CACpB,CAAC;AACF,MAAM,CAAC,MAAM,mCAAmC,GAAsB,EAAE,CAAC;AAEzE,MAAM,CAAC,MAAM,8BAA8B,GAAG,IAAI,CAAC;AACnD,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAEvC,SAAS,yBAAyB,CAAC,aAAiC,EAAU;IAC7E,IAAI,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC;QAAE,OAAO,8BAA8B,CAAC;IAC7F,wFAAwF;IACxF,6CAA6C;IAC7C,OAAO,IAAI,CAAC,GAAG,CAAC,8BAA8B,EAAE,IAAI,CAAC,GAAG,CAAC,0BAA0B,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;AAAA,CACrH;AAED,SAAS,eAAe,CACvB,eAAqC,EACrC,UAAkB,EAClB,aAAiC,EACR;IACzB,MAAM,IAAI,GAAG;QACZ,KAAK,EAAE,eAAe;QACtB,UAAU;QACV,sBAAsB,EAAE,IAAI;QAC5B,mBAAmB,EAAE,yBAAyB,CAAC,aAAa,CAAC;QAC7D,GAAG,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC9E,CAAC;IACF,QAAQ,eAAe,EAAE,CAAC;QACzB,KAAK,MAAM;YACV,OAAO,IAAI,CAAC;QACb,KAAK,MAAM;YACV,OAAO,EAAE,GAAG,IAAI,EAAE,gBAAgB,EAAE,mCAAmC,EAAE,CAAC;QAC3E,KAAK,SAAS;YACb,OAAO;gBACN,GAAG,IAAI;gBACP,gBAAgB,EAAE,sCAAsC;gBACxD,sBAAsB,EAAE,KAAK;aAC7B,CAAC;QACH,KAAK,MAAM;YACV,OAAO;gBACN,GAAG,IAAI;gBACP,gBAAgB,EAAE,mCAAmC;gBACrD,sBAAsB,EAAE,KAAK;aAC7B,CAAC;IACJ,CAAC;AAAA,CACD;AAED,MAAM,UAAU,4BAA4B,CAAC,IAG5C,EAA0B;IAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC;IACjC,MAAM,aAAa,GAClB,IAAI,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,IAAI,CAAC,aAAa,GAAG,CAAC;QAChG,CAAC,CAAC,IAAI,CAAC,aAAa;QACpB,CAAC,CAAC,SAAS,CAAC;IACd,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,OAAO,eAAe,CAAC,MAAM,EAAE,oBAAoB,EAAE,aAAa,CAAC,CAAC;IACrE,CAAC;IACD,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QACrB,OAAO,eAAe,CAAC,IAAI,EAAE,mBAAmB,EAAE,aAAa,CAAC,CAAC;IAClE,CAAC;IAED,IAAI,aAAa,KAAK,SAAS,EAAE,CAAC;QACjC,6CAA6C;QAC7C,OAAO,eAAe,CAAC,MAAM,EAAE,iCAAiC,EAAE,SAAS,CAAC,CAAC;IAC9E,CAAC;IACD,IAAI,aAAa,IAAI,iCAAiC,EAAE,CAAC;QACxD,OAAO,eAAe,CAAC,MAAM,EAAE,sBAAsB,EAAE,aAAa,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,aAAa,IAAI,iCAAiC,EAAE,CAAC;QACxD,OAAO,eAAe,CAAC,MAAM,EAAE,qBAAqB,EAAE,aAAa,CAAC,CAAC;IACtE,CAAC;IACD,IAAI,aAAa,IAAI,oCAAoC,EAAE,CAAC;QAC3D,OAAO,eAAe,CAAC,SAAS,EAAE,wBAAwB,EAAE,aAAa,CAAC,CAAC;IAC5E,CAAC;IACD,OAAO,eAAe,CAAC,MAAM,EAAE,0BAA0B,EAAE,aAAa,CAAC,CAAC;AAAA,CAC1E;AAUD,6FAA6F;AAC7F,MAAM,CAAC,MAAM,wCAAwC,GAAG,CAAC,CAAC;AAC1D,MAAM,CAAC,MAAM,qDAAqD,GAAG,CAAC,CAAC;AAEvE;;;;;;;GAOG;AACH,MAAM,UAAU,qCAAqC,CACpD,OAA+B,EAC/B,OAA4B,EACN;IACtB,IAAI,OAAO,CAAC,KAAK,KAAK,MAAM;QAAE,OAAO,OAAO,CAAC;IAC7C,OAAO;QACN,QAAQ,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,EAAE,wCAAwC,CAAC;QAC9E,mBAAmB,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,mBAAmB,EAAE,qDAAqD,CAAC;KACjH,CAAC;AAAA,CACF;AAED,6FAA6F;AAC7F,MAAM,UAAU,4BAA4B,CAAC,SAA4B,EAAE,OAA+B,EAAY;IACrH,IAAI,QAAQ,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IAC9B,IAAI,OAAO,CAAC,gBAAgB,KAAK,SAAS,EAAE,CAAC;QAC5C,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAClD,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IACzD,CAAC;IACD,IAAI,OAAO,CAAC,gBAAgB,KAAK,SAAS,EAAE,CAAC;QAC5C,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAClD,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO,QAAQ,CAAC;AAAA,CAChB","sourcesContent":["/**\n * Model capability auto-detection: derive what the harness may load onto a model FROM the model's\n * own metadata (`Model.contextWindow`), so small open models (4k/8k/16k windows, sub-1B params)\n * can still hold a usable chat instead of drowning in tool schemas and background-lane prompts.\n *\n * Derivation is metadata-first; defaults apply only when the metadata is missing (unknown/zero\n * window keeps today's full behavior rather than guessing). Detection can be disabled or forced\n * per class via the `modelCapability.mode` setting.\n */\n\nexport type ModelCapabilityClass = \"full\" | \"lean\" | \"minimal\" | \"chat\";\n\nexport type ModelCapabilityMode = \"auto\" | \"off\" | ModelCapabilityClass;\n\nexport interface ModelCapabilityProfile {\n\tclass: ModelCapabilityClass;\n\tcontextWindow?: number;\n\treasonCode: string;\n\t/** Allow-list; undefined = no allow-list restriction. */\n\tallowedToolNames?: readonly string[];\n\t/** Block-list applied after the allow-list; undefined = nothing blocked. */\n\tblockedToolNames?: readonly string[];\n\t/** Whether idle background lanes (goal auto-continue, research) may run on this model. */\n\tbackgroundLanesEnabled: boolean;\n\t/** Output-token cap for lane isolated completions, scaled to the window. */\n\tlaneMaxOutputTokens: number;\n}\n\n/** Windows at or above this keep the full harness surface. */\nexport const MODEL_CAPABILITY_FULL_MIN_CONTEXT = 32_768;\n/** Windows at or above this keep core tools but shed background-autonomy extras. */\nexport const MODEL_CAPABILITY_LEAN_MIN_CONTEXT = 16_384;\n/** Windows at or above this get the minimal coding set; below is chat-only. */\nexport const MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT = 8_192;\n\nexport const MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS: readonly string[] = [\"delegate\", \"context_audit\"];\nexport const MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS: readonly string[] = [\n\t\"read\",\n\t\"bash\",\n\t\"python\",\n\t\"powershell\",\n\t\"edit\",\n\t\"write\",\n\t// The executor tool: minimal-class models ARE the daily-ops executors, and its schema is tiny.\n\t\"run_toolkit_script\",\n];\nexport const MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS: readonly string[] = [];\n\nexport const DEFAULT_LANE_MAX_OUTPUT_TOKENS = 2048;\nconst MIN_LANE_MAX_OUTPUT_TOKENS = 256;\n\nfunction laneOutputTokensForWindow(contextWindow: number | undefined): number {\n\tif (contextWindow === undefined || contextWindow <= 0) return DEFAULT_LANE_MAX_OUTPUT_TOKENS;\n\t// A lane completion may use at most an eighth of the window for output, floored so tiny\n\t// windows still produce something parseable.\n\treturn Math.min(DEFAULT_LANE_MAX_OUTPUT_TOKENS, Math.max(MIN_LANE_MAX_OUTPUT_TOKENS, Math.floor(contextWindow / 8)));\n}\n\nfunction profileForClass(\n\tcapabilityClass: ModelCapabilityClass,\n\treasonCode: string,\n\tcontextWindow: number | undefined,\n): ModelCapabilityProfile {\n\tconst base = {\n\t\tclass: capabilityClass,\n\t\treasonCode,\n\t\tbackgroundLanesEnabled: true,\n\t\tlaneMaxOutputTokens: laneOutputTokensForWindow(contextWindow),\n\t\t...(contextWindow !== undefined && contextWindow > 0 ? { contextWindow } : {}),\n\t};\n\tswitch (capabilityClass) {\n\t\tcase \"full\":\n\t\t\treturn base;\n\t\tcase \"lean\":\n\t\t\treturn { ...base, blockedToolNames: MODEL_CAPABILITY_LEAN_BLOCKED_TOOLS };\n\t\tcase \"minimal\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_MINIMAL_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t\tcase \"chat\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tallowedToolNames: MODEL_CAPABILITY_CHAT_ALLOWED_TOOLS,\n\t\t\t\tbackgroundLanesEnabled: false,\n\t\t\t};\n\t}\n}\n\nexport function deriveModelCapabilityProfile(args: {\n\tcontextWindow?: number;\n\tmode?: ModelCapabilityMode;\n}): ModelCapabilityProfile {\n\tconst mode = args.mode ?? \"auto\";\n\tconst contextWindow =\n\t\targs.contextWindow !== undefined && Number.isFinite(args.contextWindow) && args.contextWindow > 0\n\t\t\t? args.contextWindow\n\t\t\t: undefined;\n\tif (mode === \"off\") {\n\t\treturn profileForClass(\"full\", \"detection_disabled\", contextWindow);\n\t}\n\tif (mode !== \"auto\") {\n\t\treturn profileForClass(mode, \"forced_by_setting\", contextWindow);\n\t}\n\n\tif (contextWindow === undefined) {\n\t\t// Metadata missing: defaults, never guesses.\n\t\treturn profileForClass(\"full\", \"unknown_context_window_defaults\", undefined);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_FULL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"full\", \"large_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_LEAN_MIN_CONTEXT) {\n\t\treturn profileForClass(\"lean\", \"lean_context_window\", contextWindow);\n\t}\n\tif (contextWindow >= MODEL_CAPABILITY_MINIMAL_MIN_CONTEXT) {\n\t\treturn profileForClass(\"minimal\", \"minimal_context_window\", contextWindow);\n\t}\n\treturn profileForClass(\"chat\", \"chat_only_context_window\", contextWindow);\n}\n\n/** Goal-continuation (autosteer) budgets, scaled to the session's capability class. */\nexport interface ContinuationBudgets {\n\t/** Maximum continuation prompts per idle goal loop. */\n\tmaxTurns: number;\n\t/** Wall-clock budget in minutes; 0 means \"disabled\" (upstream convention). */\n\tmaxWallClockMinutes: number;\n}\n\n/** Lean-class continuation caps: a 16-32k window cannot afford the full autosteer budget. */\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS = 2;\nexport const MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES = 5;\n\n/**\n * Scale goal-continuation budgets to the model's capability class. Lean-window models (16-32k) keep\n * autonomy but at a reduced budget; every other class passes the configured budget through unchanged\n * (full stays full; minimal/chat never reach here — their background lanes are disabled upstream).\n *\n * Both dimensions are a straight `min(configured, cap)`: a disabled wall-clock budget (0) stays\n * disabled because `min(0, cap) === 0`, so the cap only ever tightens an already-positive budget.\n */\nexport function scaleContinuationBudgetsForCapability(\n\tprofile: ModelCapabilityProfile,\n\tbudgets: ContinuationBudgets,\n): ContinuationBudgets {\n\tif (profile.class !== \"lean\") return budgets;\n\treturn {\n\t\tmaxTurns: Math.min(budgets.maxTurns, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_TURNS),\n\t\tmaxWallClockMinutes: Math.min(budgets.maxWallClockMinutes, MODEL_CAPABILITY_LEAN_MAX_CONTINUE_WALL_CLOCK_MINUTES),\n\t};\n}\n\n/** Apply the profile's allow/block lists to a requested tool-name list, preserving order. */\nexport function filterToolNamesForCapability(toolNames: readonly string[], profile: ModelCapabilityProfile): string[] {\n\tlet filtered = [...toolNames];\n\tif (profile.allowedToolNames !== undefined) {\n\t\tconst allowed = new Set(profile.allowedToolNames);\n\t\tfiltered = filtered.filter((name) => allowed.has(name));\n\t}\n\tif (profile.blockedToolNames !== undefined) {\n\t\tconst blocked = new Set(profile.blockedToolNames);\n\t\tfiltered = filtered.filter((name) => !blocked.has(name));\n\t}\n\treturn filtered;\n}\n"]}

package/dist/core/model-router/tool-escalation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA6F1D,wBAAgB,6BAA6B,CAAC,OAAO,EAAE;IACtD,IAAI,EAAE,SAAS,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,OAAO,CAeV","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA2G1D,wBAAgB,6BAA6B,CAAC,OAAO,EAAE;IACtD,IAAI,EAAE,SAAS,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,OAAO,CAeV","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"powershell\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"format-list\",\n\t\"format-table\",\n\t\"get-childitem\",\n\t\"get-command\",\n\t\"get-content\",\n\t\"get-item\",\n\t\"get-location\",\n\t\"get-process\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"resolve-path\",\n\t\"rg\",\n\t\"sed\",\n\t\"select-object\",\n\t\"select-string\",\n\t\"tail\",\n\t\"test\",\n\t\"test-path\",\n\t\"tsc\",\n\t\"wc\",\n\t\"where-object\",\n\t\"which\",\n\t\"write-output\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|remove-item\\b|move-item\\b|copy-item\\b|new-item\\b|rename-item\\b|set-content\\b|add-content\\b|out-file\\b|set-item\\b|start-process\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|powershell|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}

package/dist/core/model-router/tool-escalation.js

@@ -13,7 +13,7 @@ const READ_ONLY_TOOL_NAMES = new Set([
     "read_url_content",
     "read_browser_page",
 ]);
-const SHELL_TOOL_NAMES = new Set(["bash", "exec", "execute", "run", "run_command", "shell"]);
+const SHELL_TOOL_NAMES = new Set(["bash", "powershell", "exec", "execute", "run", "run_command", "shell"]);
 const READ_ONLY_COMMANDS = new Set([
     "awk",
     "cat",
@@ -21,6 +21,14 @@ const READ_ONLY_COMMANDS = new Set([
     "df",
     "du",
     "env",
+    "format-list",
+    "format-table",
+    "get-childitem",
+    "get-command",
+    "get-content",
+    "get-item",
+    "get-location",
+    "get-process",
     "git",
     "grep",
     "head",
@@ -30,20 +38,26 @@ const READ_ONLY_COMMANDS = new Set([
     "npm",
     "pnpm",
     "pwd",
+    "resolve-path",
     "rg",
     "sed",
+    "select-object",
+    "select-string",
     "tail",
     "test",
+    "test-path",
     "tsc",
     "wc",
+    "where-object",
     "which",
+    "write-output",
     "yarn",
 ]);
 const READ_ONLY_GIT_SUBCOMMANDS = new Set(["branch", "diff", "log", "rev-parse", "show", "status", "tag"]);
 const READ_ONLY_NPM_SUBCOMMANDS = new Set(["info", "list", "ls", "outdated", "view", "whoami"]);
-const MUTATING_SHELL_TOKEN_RE = /(^|\s)(>|>>|2>|&>|tee\b|rm\b|mv\b|cp\b|mkdir\b|touch\b|chmod\b|chown\b|install\b|commit\b|push\b|publish\b|deploy\b|apply\b|add\b|checkout\b|switch\b|reset\b|clean\b|stash\b|merge\b|rebase\b|npm\s+(?:i|install|ci|update|publish|run)\b|pnpm\s+(?:i|install|update|publish|run)\b|yarn\s+(?:add|install|upgrade|publish|run)\b)/i;
+const MUTATING_SHELL_TOKEN_RE = /(^|\s)(>|>>|2>|&>|tee\b|rm\b|mv\b|cp\b|mkdir\b|touch\b|chmod\b|chown\b|install\b|commit\b|push\b|publish\b|deploy\b|apply\b|add\b|checkout\b|switch\b|reset\b|clean\b|stash\b|merge\b|rebase\b|remove-item\b|move-item\b|copy-item\b|new-item\b|rename-item\b|set-content\b|add-content\b|out-file\b|set-item\b|start-process\b|npm\s+(?:i|install|ci|update|publish|run)\b|pnpm\s+(?:i|install|update|publish|run)\b|yarn\s+(?:add|install|upgrade|publish|run)\b)/i;
 const UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\$\(|\bfind\b[\s\S]*\s-exec(?:dir)?\b|\bxargs\b)/i;
-const MUTATING_TOOL_NAME_RE = /(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;
+const MUTATING_TOOL_NAME_RE = /(bash|powershell|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;
 function getShellCommand(args) {
     if (!args || typeof args !== "object")
         return undefined;

package/dist/core/model-router/tool-escalation.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,IAAI;IACJ,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,OAAO;IACP,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,qUAAqU,CAAC;AACvU,MAAM,gCAAgC,GAAG,sDAAsD,CAAC;AAChG,MAAM,qBAAqB,GAC1B,+IAA+I,CAAC;AAEjJ,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,gCAAgC,CAAC,IAAI,CAAC,OAAO,CAAC;QACtG,OAAO,KAAK,CAAC;IACd,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC9F,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,UAAU,6BAA6B,CAAC,OAM7C,EAAW;IACX,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,8FAA8F;IAC9F,8FAA8F;IAC9F,yFAAyF;IACzF,6CAA6C;IAC7C,IAAI,OAAO,CAAC,UAAU,KAAK,iBAAiB,IAAI,QAAQ,KAAK,oBAAoB;QAAE,OAAO,KAAK,CAAC;IAChG,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE3G,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,aAAa;IACb,cAAc;IACd,eAAe;IACf,aAAa;IACb,aAAa;IACb,UAAU;IACV,cAAc;IACd,aAAa;IACb,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,cAAc;IACd,IAAI;IACJ,KAAK;IACL,eAAe;IACf,eAAe;IACf,MAAM;IACN,MAAM;IACN,WAAW;IACX,KAAK;IACL,IAAI;IACJ,cAAc;IACd,OAAO;IACP,cAAc;IACd,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,scAAsc,CAAC;AACxc,MAAM,gCAAgC,GAAG,sDAAsD,CAAC;AAChG,MAAM,qBAAqB,GAC1B,0JAA0J,CAAC;AAE5J,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,gCAAgC,CAAC,IAAI,CAAC,OAAO,CAAC;QACtG,OAAO,KAAK,CAAC;IACd,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC9F,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC;AAAA,CACjH;AAED,MAAM,UAAU,6BAA6B,CAAC,OAM7C,EAAW;IACX,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,8FAA8F;IAC9F,8FAA8F;IAC9F,yFAAyF;IACzF,6CAA6C;IAC7C,IAAI,OAAO,CAAC,UAAU,KAAK,iBAAiB,IAAI,QAAQ,KAAK,oBAAoB;QAAE,OAAO,KAAK,CAAC;IAChG,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"powershell\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"format-list\",\n\t\"format-table\",\n\t\"get-childitem\",\n\t\"get-command\",\n\t\"get-content\",\n\t\"get-item\",\n\t\"get-location\",\n\t\"get-process\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"resolve-path\",\n\t\"rg\",\n\t\"sed\",\n\t\"select-object\",\n\t\"select-string\",\n\t\"tail\",\n\t\"test\",\n\t\"test-path\",\n\t\"tsc\",\n\t\"wc\",\n\t\"where-object\",\n\t\"which\",\n\t\"write-output\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|remove-item\\b|move-item\\b|copy-item\\b|new-item\\b|rename-item\\b|set-content\\b|add-content\\b|out-file\\b|set-item\\b|start-process\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst UNSAFE_NESTED_SHELL_EXECUTION_RE = /(`|\\$\\(|\\bfind\\b[\\s\\S]*\\s-exec(?:dir)?\\b|\\bxargs\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|powershell|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command) || UNSAFE_NESTED_SHELL_EXECUTION_RE.test(command))\n\t\treturn false;\n\tconst segments = command.split(/\\s*(?:&&|\\|\\||[;|\\r\\n])\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every((segment) => segment.length > 0 && isReadOnlyShellSegment(segment));\n}\n\nexport function shouldEscalateModelRouterTool(options: {\n\ttier: ModelTier;\n\ttoolName: string;\n\targs?: unknown;\n\t/** The route's reasonCode; executor-lane turns carry \"executor_direct\". */\n\treasonCode?: string;\n}): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\t// Executor-lane turns (G16) exist to run exactly one tool: run_toolkit_script, which enforces\n\t// its own safety (danger confirmation, structural exit-code contract). Escalating on it would\n\t// abort every executor turn at the moment it does its job. Any OTHER mutating tool still\n\t// escalates to the expensive model as usual.\n\tif (options.reasonCode === \"executor_direct\" && toolName === \"run_toolkit_script\") return false;\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}

package/dist/core/models/fitness-store.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"fitness-store.d.ts","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAEvE;;;;;;GAMG;AAEH,MAAM,WAAW,eAAe;IAC/B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,wBAAgB,sBAAsB,IAAI,eAAe,CAWxD;AAED,MAAM,WAAW,mBAAmB;IACnC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,kBAAkB,CAAC;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;CACtB;AAQD,qBAAa,YAAY;IACxB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,EAG9E;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,GAAG,YAAY,CAEpG;IAED,OAAO,CAAC,IAAI;IAaZ,iGAAiG;IACjG,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,kBAAkB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,mBAAmB,CAQhF;IAED,yFAAyF;IACzF,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAO1B;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,mBAAmB,EAAE,CAGjD;IAED,4EAA4E;IAC5E,MAAM,IAAI,mBAAmB,EAAE,CAG9B;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = cpuList.length;\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Drop a model's report for the CURRENT host (uninstall cleanup). No-op when absent. */\n\tremove(model: string): void {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tif (!file.hosts[host.id]?.[model]) return;\n\t\tdelete file.hosts[host.id][model];\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}
+{"version":3,"file":"fitness-store.d.ts","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAEvE;;;;;;GAMG;AAEH,MAAM,WAAW,eAAe;IAC/B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,wBAAgB,sBAAsB,IAAI,eAAe,CAWxD;AAED,MAAM,WAAW,mBAAmB;IACnC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,kBAAkB,CAAC;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;CACtB;AAQD,qBAAa,YAAY;IACxB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,EAG9E;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,GAAG,YAAY,CAEpG;IAED,OAAO,CAAC,IAAI;IAaZ,iGAAiG;IACjG,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,kBAAkB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,mBAAmB,CAQhF;IAED,yFAAyF;IACzF,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAO1B;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,mBAAmB,EAAE,CAGjD;IAED,4EAA4E;IAC5E,MAAM,IAAI,mBAAmB,EAAE,CAG9B;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = Math.max(1, cpuList.length);\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Drop a model's report for the CURRENT host (uninstall cleanup). No-op when absent. */\n\tremove(model: string): void {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tif (!file.hosts[host.id]?.[model]) return;\n\t\tdelete file.hosts[host.id][model];\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}

package/dist/core/models/fitness-store.js

@@ -4,7 +4,7 @@ import { dirname, join } from "node:path";
 export function currentHostFingerprint() {
     const cpuList = cpus();
     const cpu = (cpuList[0]?.model ?? "unknown-cpu").trim();
-    const cores = cpuList.length;
+    const cores = Math.max(1, cpuList.length);
     const totalMemGb = Math.round(totalmem() / 1024 ** 3);
     const id = `${cpu
         .toLowerCase()

package/dist/core/models/fitness-store.js.map

@@ -1 +1 @@
-{"version":3,"file":"fitness-store.js","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAmB1C,MAAM,UAAU,sBAAsB,GAAoB;IACzD,MAAM,OAAO,GAAG,IAAI,EAAE,CAAC;IACvB,MAAM,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,aAAa,CAAC,CAAC,IAAI,EAAE,CAAC;IACxD,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC;IACtD,MAAM,EAAE,GAAG,GAAG,GAAG;SACf,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,KAAK,UAAU,GAAG,CAAC;IAC1C,OAAO,EAAE,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AAAA,CACtC;AAeD,MAAM,OAAO,YAAY;IACP,QAAQ,CAAS;IACjB,WAAW,CAAwB;IAEpD,YAAY,QAAgB,EAAE,OAAiD,EAAE;QAChF,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,sBAAsB,CAAC;IAAA,CAClE;IAED,MAAM,CAAC,WAAW,CAAC,QAAgB,EAAE,OAAiD,EAAgB;QACrG,OAAO,IAAI,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,oBAAoB,CAAC,EAAE,OAAO,CAAC,CAAC;IAAA,CAChF;IAEO,IAAI,GAAqB;QAChC,IAAI,CAAC;YACJ,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;gBAAE,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;YACjE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAqB,CAAC;YACpF,IAAI,MAAM,IAAI,MAAM,CAAC,OAAO,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;gBACxF,OAAO,MAAM,CAAC;YACf,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,oFAAoF;QACrF,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;IAAA,CACjC;IAED,iGAAiG;IACjG,IAAI,CAAC,KAAa,EAAE,MAA0B,EAAE,EAAW,EAAuB;QACjF,MAAM,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAChC,MAAM,KAAK,GAAwB,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,IAAI,EAAE,CAAC;QAC/F,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,CAAC;QACzE,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACvD,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC/E,OAAO,KAAK,CAAC;IAAA,CACb;IAED,yFAAyF;IACzF,MAAM,CAAC,KAAa,EAAQ;QAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,KAAK,CAAC;YAAE,OAAO;QAC1C,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC;QAClC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACvD,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAAA,CAC/E;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAe,EAAyB;QAClD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IAAA,CACxE;IAED,4EAA4E;IAC5E,MAAM,GAA0B;QAC/B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;IAAA,CAC5E;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = cpuList.length;\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Drop a model's report for the CURRENT host (uninstall cleanup). No-op when absent. */\n\tremove(model: string): void {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tif (!file.hosts[host.id]?.[model]) return;\n\t\tdelete file.hosts[host.id][model];\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}
+{"version":3,"file":"fitness-store.js","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAmB1C,MAAM,UAAU,sBAAsB,GAAoB;IACzD,MAAM,OAAO,GAAG,IAAI,EAAE,CAAC;IACvB,MAAM,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,aAAa,CAAC,CAAC,IAAI,EAAE,CAAC;IACxD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC;IACtD,MAAM,EAAE,GAAG,GAAG,GAAG;SACf,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,KAAK,UAAU,GAAG,CAAC;IAC1C,OAAO,EAAE,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AAAA,CACtC;AAeD,MAAM,OAAO,YAAY;IACP,QAAQ,CAAS;IACjB,WAAW,CAAwB;IAEpD,YAAY,QAAgB,EAAE,OAAiD,EAAE;QAChF,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,sBAAsB,CAAC;IAAA,CAClE;IAED,MAAM,CAAC,WAAW,CAAC,QAAgB,EAAE,OAAiD,EAAgB;QACrG,OAAO,IAAI,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,oBAAoB,CAAC,EAAE,OAAO,CAAC,CAAC;IAAA,CAChF;IAEO,IAAI,GAAqB;QAChC,IAAI,CAAC;YACJ,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;gBAAE,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;YACjE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAqB,CAAC;YACpF,IAAI,MAAM,IAAI,MAAM,CAAC,OAAO,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;gBACxF,OAAO,MAAM,CAAC;YACf,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,oFAAoF;QACrF,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;IAAA,CACjC;IAED,iGAAiG;IACjG,IAAI,CAAC,KAAa,EAAE,MAA0B,EAAE,EAAW,EAAuB;QACjF,MAAM,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAChC,MAAM,KAAK,GAAwB,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,IAAI,EAAE,CAAC;QAC/F,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,CAAC;QACzE,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACvD,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC/E,OAAO,KAAK,CAAC;IAAA,CACb;IAED,yFAAyF;IACzF,MAAM,CAAC,KAAa,EAAQ;QAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,KAAK,CAAC;YAAE,OAAO;QAC1C,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC;QAClC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACvD,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAAA,CAC/E;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAe,EAAyB;QAClD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IAAA,CACxE;IAED,4EAA4E;IAC5E,MAAM,GAA0B;QAC/B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;IAAA,CAC5E;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = Math.max(1, cpuList.length);\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Drop a model's report for the CURRENT host (uninstall cleanup). No-op when absent. */\n\tremove(model: string): void {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tif (!file.hosts[host.id]?.[model]) return;\n\t\tdelete file.hosts[host.id][model];\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}

package/dist/core/models/local-runtime.d.ts

@@ -1,4 +1,4 @@
-import { type ChildProcess } from "node:child_process";
+import { type ChildProcess, type SpawnOptions } from "node:child_process";
 /**
  * Local model runtime manager (local-model-lifecycle-design.md): Ollama first, interface kept
  * runtime-agnostic. Pi SPAWNS the serve process itself with OWNED model storage
@@ -76,11 +76,12 @@ export type RuntimeCommandRunner = (command: string, args: string[], options?: {
     env?: NodeJS.ProcessEnv;
     onOutput?: (chunk: string) => void;
 }) => Promise<RuntimeCommandResult>;
+type LocalRuntimeSpawnOptions = Pick<SpawnOptions, "detached" | "stdio"> & {
+    env: NodeJS.ProcessEnv;
+};
 export interface LocalRuntimeDeps {
     fetchFn?: typeof fetch;
-    spawnFn?: (command: string, args: string[], options: {
-        env: NodeJS.ProcessEnv;
-    }) => Pick<ChildProcess, "pid" | "kill" | "unref" | "on">;
+    spawnFn?: (command: string, args: string[], options: LocalRuntimeSpawnOptions) => Pick<ChildProcess, "pid" | "kill" | "unref" | "on">;
     existsFn?: (path: string) => boolean;
     linkFile?: (source: string, target: string) => void;
     copyFile?: (source: string, target: string) => void;
@@ -305,4 +306,5 @@ export declare class TransformersRuntime {
     };
     installGuide(): string[];
 }
+export {};
 //# sourceMappingURL=local-runtime.d.ts.map