@mikode13/harness 5.4.0 → 6.0.0

package/dist/engines/domain/model/stepContext.js.map

@@ -1 +1 @@
-{"version":3,"file":"stepContext.js","sourceRoot":"","sources":["../../../../src/engines/domain/model/stepContext.ts"],"names":[],"mappings":"AAIA,kFAAkF;AAClF,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,EAC9B,YAAY,EACZ,kBAAkB,EAClB,KAAK,EACL,OAAO,EACP,SAAS,GAQT;IACA,MAAM,QAAQ,GAAgB;QAC7B,YAAY,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;QACvC,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,MAAM;QACnC,QAAQ,EAAE,CAAC;QACX,SAAS,EAAE,CAAC;QACZ,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,EAAE;QACf,cAAc,EAAE,EAAE;KAClB,CAAC;IACF,MAAM,KAAK,GAAG,IAAI,GAAG,EAAwB,CAAC;IAC9C,MAAM,MAAM,GAAG,IAAI,GAAG,EAAiD,CAAC;IACxE,MAAM,OAAO,GAAkC,EAAE,CAAC;IAElD,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAChE,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;YACnB,KAAK,MAAM;gBACV,QAAQ,CAAC,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC;gBACtC,MAAM;YACP,mFAAmF;YACnF,wCAAwC;YACxC,KAAK,WAAW;gBACf,MAAM;YACP,KAAK,cAAc;gBAClB,IAAI,kBAAkB,KAAK,SAAS,IAAI,IAAI,CAAC,MAAM,KAAK,kBAAkB,EAAE,CAAC;oBAC5E,QAAQ,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;gBACxD,CAAC;gBACD,MAAM;YACP,KAAK,UAAU;gBACd,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;gBACzB,QAAQ,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC;gBACxD,MAAM;YACP,KAAK,YAAY,CAAC,CAAC,CAAC;gBACnB,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC;gBACtC,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;gBAClE,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;gBAC9B,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;gBAChB,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;gBAC5B,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACpC,MAAM,OAAO,GAAG,IAAI,IAAI,SAAS,CAAC,IAAI,CAAC,CAAC;gBACxC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,UAAU,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;gBAC/E,MAAM;YACP,CAAC;QACF,CAAC;IACF,CAAC;IAED,QAAQ,CAAC,WAAW,GAAG,CAAC,GAAG,MAAM,CAAC;SAChC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAC;SACrE,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC;IAC9C,QAAQ,CAAC,cAAc,GAAG,OAAO;SAC/B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC;SAC3C,KAAK,CAAC,CAAC,EAAE,cAAc,CAAC,CAAC;IAC3B,OAAO,QAAQ,CAAC;AACjB,CAAC","sourcesContent":["import type { StepContext } from '#src/agent/domain/agent';\nimport type { Message, ToolCallPart } from '#src/llm/domain/message';\nimport type { ToolDefinition } from '#src/llm/domain/tool';\n\n// Enough to name what fills a context without listing every result of a long run.\nconst largestResults = 5;\n\n/**\n * What one call to the model sends, in characters, by where it comes from. Characters, not\n * tokens: the harness does not tokenize, and the provider's count arrives with the answer.\n * `summarize` names a tool call as its tool does, or gives nothing.\n */\nexport function measureContext({\n\tinstructions,\n\tproviderDataSource,\n\ttools,\n\tcontext,\n\tsummarize,\n}: {\n\tinstructions: string | undefined;\n\t/** The only `providerData` the client sends back; every other is dropped. */\n\tproviderDataSource: string | undefined;\n\ttools: ToolDefinition[];\n\tcontext: Message[];\n\tsummarize: (call: ToolCallPart) => string | undefined;\n}): StepContext {\n\tconst measured: StepContext = {\n\t\tinstructions: instructions?.length ?? 0,\n\t\ttools: JSON.stringify(tools).length,\n\t\tmessages: 0,\n\t\treasoning: 0,\n\t\ttoolCalls: 0,\n\t\ttoolResults: [],\n\t\tlargestResults: [],\n\t};\n\tconst calls = new Map<string, ToolCallPart>();\n\tconst byTool = new Map<string, { characters: number; calls: number }>();\n\tconst results: StepContext['largestResults'] = [];\n\n\tfor (const part of context.flatMap(message => message.content)) {\n\t\tswitch (part.type) {\n\t\t\tcase 'text':\n\t\t\t\tmeasured.messages += part.text.length;\n\t\t\t\tbreak;\n\t\t\t// Readable reasoning is never sent back: only the provider's own blocks are, which\n\t\t\t// already hold it, signed or encrypted.\n\t\t\tcase 'reasoning':\n\t\t\t\tbreak;\n\t\t\tcase 'providerData':\n\t\t\t\tif (providerDataSource === undefined || part.source === providerDataSource) {\n\t\t\t\t\tmeasured.reasoning += JSON.stringify(part.data).length;\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\tcase 'toolCall':\n\t\t\t\tcalls.set(part.id, part);\n\t\t\t\tmeasured.toolCalls += JSON.stringify(part.input).length;\n\t\t\t\tbreak;\n\t\t\tcase 'toolResult': {\n\t\t\t\tconst characters = part.output.length;\n\t\t\t\tconst tool = byTool.get(part.name) ?? { characters: 0, calls: 0 };\n\t\t\t\ttool.characters += characters;\n\t\t\t\ttool.calls += 1;\n\t\t\t\tbyTool.set(part.name, tool);\n\t\t\t\tconst call = calls.get(part.callId);\n\t\t\t\tconst summary = call && summarize(call);\n\t\t\t\tresults.push({ tool: part.name, characters, ...(summary ? { summary } : {}) });\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t}\n\n\tmeasured.toolResults = [...byTool]\n\t\t.map(([tool, { characters, calls }]) => ({ tool, characters, calls }))\n\t\t.sort((a, b) => b.characters - a.characters);\n\tmeasured.largestResults = results\n\t\t.sort((a, b) => b.characters - a.characters)\n\t\t.slice(0, largestResults);\n\treturn measured;\n}\n"]}
+{"version":3,"file":"stepContext.js","sourceRoot":"","sources":["../../../../src/engines/domain/model/stepContext.ts"],"names":[],"mappings":"AACA,OAAO,EAAgB,WAAW,EAAqB,MAAM,yBAAyB,CAAC;AAGvF,kFAAkF;AAClF,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,EAC9B,YAAY,EACZ,kBAAkB,EAClB,KAAK,EACL,OAAO,EACP,SAAS,GAQT;IACA,MAAM,QAAQ,GAAgB;QAC7B,YAAY,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;QACvC,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,MAAM;QACnC,QAAQ,EAAE,CAAC;QACX,SAAS,EAAE,CAAC;QACZ,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,EAAE;QACf,cAAc,EAAE,EAAE;KAClB,CAAC;IACF,MAAM,KAAK,GAAG,IAAI,GAAG,EAAwB,CAAC;IAC9C,MAAM,MAAM,GAAG,IAAI,GAAG,EAAiD,CAAC;IACxE,MAAM,OAAO,GAAkC,EAAE,CAAC;IAElD,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,OAAO,EAAE,kBAAkB,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACjG,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;YACnB,KAAK,MAAM;gBACV,QAAQ,CAAC,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC;gBACtC,MAAM;YACP,8EAA8E;YAC9E,KAAK,YAAY;gBAChB,IAAI,IAAI,CAAC,MAAM,KAAK,kBAAkB,EAAE,CAAC;oBACxC,QAAQ,CAAC,QAAQ,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;gBACvD,CAAC;gBACD,MAAM;YACP,mFAAmF;YACnF,wCAAwC;YACxC,KAAK,WAAW;gBACf,MAAM;YACP,KAAK,cAAc;gBAClB,IAAI,kBAAkB,KAAK,SAAS,IAAI,IAAI,CAAC,MAAM,KAAK,kBAAkB,EAAE,CAAC;oBAC5E,QAAQ,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;gBACxD,CAAC;gBACD,MAAM;YACP,KAAK,UAAU;gBACd,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;gBACzB,QAAQ,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC;gBACxD,MAAM;YACP,KAAK,YAAY,CAAC,CAAC,CAAC;gBACnB,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC;gBACtC,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;gBAClE,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;gBAC9B,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;gBAChB,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;gBAC5B,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACpC,MAAM,OAAO,GAAG,IAAI,IAAI,SAAS,CAAC,IAAI,CAAC,CAAC;gBACxC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,UAAU,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;gBAC/E,MAAM;YACP,CAAC;QACF,CAAC;IACF,CAAC;IAED,QAAQ,CAAC,WAAW,GAAG,CAAC,GAAG,MAAM,CAAC;SAChC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAC;SACrE,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC;IAC9C,QAAQ,CAAC,cAAc,GAAG,OAAO;SAC/B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC;SAC3C,KAAK,CAAC,CAAC,EAAE,cAAc,CAAC,CAAC;IAC3B,OAAO,QAAQ,CAAC;AACjB,CAAC","sourcesContent":["import type { StepContext } from '#src/agent/domain/agent';\nimport { type Message, sentContext, type ToolCallPart } from '#src/llm/domain/message';\nimport type { ToolDefinition } from '#src/llm/domain/tool';\n\n// Enough to name what fills a context without listing every result of a long run.\nconst largestResults = 5;\n\n/**\n * What one call to the model sends, in characters, by where it comes from. Characters, not\n * tokens: the harness does not tokenize, and the provider's count arrives with the answer.\n * `summarize` names a tool call as its tool does, or gives nothing.\n */\nexport function measureContext({\n\tinstructions,\n\tproviderDataSource,\n\ttools,\n\tcontext,\n\tsummarize,\n}: {\n\tinstructions: string | undefined;\n\t/** The only `providerData` the client sends back; every other is dropped. */\n\tproviderDataSource: string | undefined;\n\ttools: ToolDefinition[];\n\tcontext: Message[];\n\tsummarize: (call: ToolCallPart) => string | undefined;\n}): StepContext {\n\tconst measured: StepContext = {\n\t\tinstructions: instructions?.length ?? 0,\n\t\ttools: JSON.stringify(tools).length,\n\t\tmessages: 0,\n\t\treasoning: 0,\n\t\ttoolCalls: 0,\n\t\ttoolResults: [],\n\t\tlargestResults: [],\n\t};\n\tconst calls = new Map<string, ToolCallPart>();\n\tconst byTool = new Map<string, { characters: number; calls: number }>();\n\tconst results: StepContext['largestResults'] = [];\n\n\tfor (const part of sentContext(context, providerDataSource).flatMap(message => message.content)) {\n\t\tswitch (part.type) {\n\t\t\tcase 'text':\n\t\t\t\tmeasured.messages += part.text.length;\n\t\t\t\tbreak;\n\t\t\t// The provider's summary of the turns before it stands in for their messages.\n\t\t\tcase 'compaction':\n\t\t\t\tif (part.source === providerDataSource) {\n\t\t\t\t\tmeasured.messages += JSON.stringify(part.data).length;\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\t// Readable reasoning is never sent back: only the provider's own blocks are, which\n\t\t\t// already hold it, signed or encrypted.\n\t\t\tcase 'reasoning':\n\t\t\t\tbreak;\n\t\t\tcase 'providerData':\n\t\t\t\tif (providerDataSource === undefined || part.source === providerDataSource) {\n\t\t\t\t\tmeasured.reasoning += JSON.stringify(part.data).length;\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\tcase 'toolCall':\n\t\t\t\tcalls.set(part.id, part);\n\t\t\t\tmeasured.toolCalls += JSON.stringify(part.input).length;\n\t\t\t\tbreak;\n\t\t\tcase 'toolResult': {\n\t\t\t\tconst characters = part.output.length;\n\t\t\t\tconst tool = byTool.get(part.name) ?? { characters: 0, calls: 0 };\n\t\t\t\ttool.characters += characters;\n\t\t\t\ttool.calls += 1;\n\t\t\t\tbyTool.set(part.name, tool);\n\t\t\t\tconst call = calls.get(part.callId);\n\t\t\t\tconst summary = call && summarize(call);\n\t\t\t\tresults.push({ tool: part.name, characters, ...(summary ? { summary } : {}) });\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t}\n\n\tmeasured.toolResults = [...byTool]\n\t\t.map(([tool, { characters, calls }]) => ({ tool, characters, calls }))\n\t\t.sort((a, b) => b.characters - a.characters);\n\tmeasured.largestResults = results\n\t\t.sort((a, b) => b.characters - a.characters)\n\t\t.slice(0, largestResults);\n\treturn measured;\n}\n"]}

package/dist/factory/infrastructure/agentLLMFactory.d.ts

@@ -48,6 +48,13 @@ export interface CreateLLMAgentOptions {
      * bound. A run that asks for exactly the same tool calls 5 times in a row fails before that.
      */
     maxSteps?: number;
+    /**
+     * The input tokens at which the provider compacts the conversation on its side: it replaces
+     * every turn before that call with its own summary, which only its model reads back. The
+     * conversation keeps every original turn. Without it, the conversation is never compacted.
+     * At least 50,000 on Anthropic, where Claude Haiku cannot compact.
+     */
+    compactAt?: number;
     /**
      * Runs every tool call without asking, `destructive` ones included, for CI where no one can
      * answer. Without it, a `destructive` call goes to the run's `approve`, or is denied.
@@ -58,6 +65,8 @@ export interface CreateLLMAgentOptions {
 }
 /** A limit per role of `createLLMOrchestrator`; a role left out keeps its default. */
 export type RoleSteps = Partial<Record<OrchestratorRole, number>>;
+/** A number of tokens per role of `createLLMOrchestrator`; a role left out keeps its default. */
+export type RoleTokens = Partial<Record<OrchestratorRole, number>>;
 type OrchestratorRole = CommandRole;
 export interface CreateLLMOrchestratorOptions extends CreateOrchestratorOptions {
     /**
@@ -73,6 +82,12 @@ export interface CreateLLMOrchestratorOptions extends CreateOrchestratorOptions
      * before its limit.
      */
     maxSteps?: number | RoleSteps;
+    /**
+     * The input tokens at which each role's provider compacts its conversation, as
+     * `CreateLLMAgentOptions.compactAt`: one number for every role, or a number per role. Without
+     * it, no role compacts.
+     */
+    compactAt?: number | RoleTokens;
     /**
      * Commands the roles may run, such as the project's tests, through `runCommand`. Each goes to
      * the roles its `roles` names, the executor by default. See `createFileTools`. Each needs a
@@ -97,7 +112,7 @@ export interface CreateLLMOrchestratorOptions extends CreateOrchestratorOptions
  * `OPENAI_API_KEY` is missing. The Anthropic SDK resolves credentials lazily, so a missing
  * `ANTHROPIC_API_KEY` fails the first run instead, with an `UnrecoverableError`.
  */
-export declare function createLLMAgent(provider: AgentProvider, { model, reasoningEffort, systemPrompt, tools, workspace, summarizeUndone, maxSteps, autoApprove, logger, }: CreateLLMAgentOptions): Agent;
+export declare function createLLMAgent(provider: AgentProvider, { model, reasoningEffort, systemPrompt, tools, workspace, summarizeUndone, maxSteps, compactAt, autoApprove, logger, }: CreateLLMAgentOptions): Agent;
 /**
  * The file tools for an agent from `createLLMAgent` on `provider`, over `workspace`, in the
  * provider's own format:
@@ -152,5 +167,5 @@ export declare function createFileTools(provider: AgentProvider, workspace: Work
  * @throws {UnrecoverableError} when a role cannot be set up, for example because
  * `OPENAI_API_KEY` is missing, or when no program can read the repository.
  */
-export declare function createLLMOrchestrator({ provider, autoApprove, systemPrompts, workspace, maxSteps, commands, summarizeUndone, logger, }?: CreateLLMOrchestratorOptions): Promise<Agent>;
+export declare function createLLMOrchestrator({ provider, autoApprove, systemPrompts, workspace, maxSteps, compactAt, commands, summarizeUndone, logger, }?: CreateLLMOrchestratorOptions): Promise<Agent>;
 export {};

package/dist/factory/infrastructure/agentLLMFactory.js

@@ -39,29 +39,35 @@ const defaultRoleSteps = {
     executor: 80,
     reviewer: 65,
 };
+// No role compacts unless asked to.
+const defaultRoleCompaction = {
+    planner: undefined,
+    executor: undefined,
+    reviewer: undefined,
+};
 /**
- * Each role's step limit, from one number for all of them or a number per role.
+ * Each role's value of the option `name`, from one number for all of them or a number per role.
+ * Each number is checked by whatever takes it.
  *
  * @throws {InvalidAgentConfigError} for a value that is neither, and for a role that does not
  * exist, which is a typo that would otherwise leave that role on its default without a word.
  */
-function stepsPerRole(maxSteps) {
-    if (maxSteps === undefined)
-        return defaultRoleSteps;
-    if (typeof maxSteps === 'number') {
-        return { planner: maxSteps, executor: maxSteps, reviewer: maxSteps };
-    }
+function perRole(name, value, defaults, { unit, example }) {
+    if (value === undefined)
+        return defaults;
+    if (typeof value === 'number')
+        return { planner: value, executor: value, reviewer: value };
     // Untyped input, such as the CLI's configuration file, can hold `null` or a list here.
-    if (typeof maxSteps !== 'object' || maxSteps === null || Array.isArray(maxSteps)) {
-        throw new InvalidAgentConfigError('maxSteps must be a number or an object of steps per role, such as { "executor": 120 }');
+    if (typeof value !== 'object' || value === null || Array.isArray(value)) {
+        throw new InvalidAgentConfigError(`${name} must be a number or an object of ${unit} per role, such as ${example}`);
     }
-    const unknown = Object.keys(maxSteps).filter(role => !(role in defaultRoleSteps));
+    const unknown = Object.keys(value).filter(role => !(role in defaults));
     if (unknown.length > 0) {
-        throw new InvalidAgentConfigError(`maxSteps names roles that do not exist: ${unknown.join(', ')}; expected planner, executor or reviewer`);
+        throw new InvalidAgentConfigError(`${name} names roles that do not exist: ${unknown.join(', ')}; expected planner, executor or reviewer`);
     }
     // A role given as `undefined`, which untyped input can do, keeps its default.
-    const given = Object.entries(maxSteps).filter(([, steps]) => steps !== undefined);
-    return { ...defaultRoleSteps, ...Object.fromEntries(given) };
+    const given = Object.entries(value).filter(([, each]) => each !== undefined);
+    return { ...defaults, ...Object.fromEntries(given) };
 }
 const defaultLLMAgentModels = {
     anthropic: 'sonnet',
@@ -70,12 +76,13 @@ const defaultLLMAgentModels = {
 function unknownProvider(provider) {
     return new InvalidAgentConfigError(`"${String(provider)}" is not an agent provider; expected one of: ${agentProviders.join(', ')}`);
 }
-function llmClientFor(provider, { model, reasoningEffort, systemPrompt, logger, }) {
+function llmClientFor(provider, { model, reasoningEffort, compactAt, systemPrompt, logger, }) {
     switch (provider) {
         case 'anthropic':
             return new ClaudeLLMClient({
                 model: model ?? defaultLLMAgentModels.anthropic,
                 reasoningEffort,
+                compactAt,
                 systemPrompt,
                 logger,
             });
@@ -83,6 +90,7 @@ function llmClientFor(provider, { model, reasoningEffort, systemPrompt, logger,
             return new OpenAILLMClient({
                 model: model ?? defaultLLMAgentModels.openai,
                 reasoningEffort,
+                compactAt,
                 systemPrompt,
                 logger,
             });
@@ -148,8 +156,14 @@ function withGuidance(systemPrompt, guidance) {
     return guidance === '' ? systemPrompt : `${systemPrompt}\n\n${guidance}`;
 }
 /** An `LLMAgent` on `provider`, wrapped in the harness's retry policy. */
-function buildLLMAgent(provider, { model, reasoningEffort, systemPrompt, tools, history, summarizer, maxSteps, autoApprove, logger, }) {
-    const llmClient = llmClientFor(provider, { model, reasoningEffort, systemPrompt, logger });
+function buildLLMAgent(provider, { model, reasoningEffort, systemPrompt, tools, history, summarizer, maxSteps, compactAt, autoApprove, logger, }) {
+    const llmClient = llmClientFor(provider, {
+        model,
+        reasoningEffort,
+        compactAt,
+        systemPrompt,
+        logger,
+    });
     // LLMAgent records nothing from a failed call, so the original prompt is the whole story.
     return new RetryingAgent({
         inner: new LLMAgent({
@@ -177,7 +191,7 @@ function buildLLMAgent(provider, { model, reasoningEffort, systemPrompt, tools,
  * `OPENAI_API_KEY` is missing. The Anthropic SDK resolves credentials lazily, so a missing
  * `ANTHROPIC_API_KEY` fails the first run instead, with an `UnrecoverableError`.
  */
-export function createLLMAgent(provider, { model, reasoningEffort, systemPrompt, tools = [], workspace, summarizeUndone = true, maxSteps, autoApprove, logger = new Logger(), }) {
+export function createLLMAgent(provider, { model, reasoningEffort, systemPrompt, tools = [], workspace, summarizeUndone = true, maxSteps, compactAt, autoApprove, logger = new Logger(), }) {
     return buildLLMAgent(provider, {
         model,
         reasoningEffort,
@@ -189,6 +203,7 @@ export function createLLMAgent(provider, { model, reasoningEffort, systemPrompt,
         ...(workspace ? { history: historyOf(workspace) } : {}),
         ...(workspace && summarizeUndone ? { summarizer: summarizerFor(provider, logger) } : {}),
         maxSteps,
+        compactAt,
         autoApprove,
         logger,
     });
@@ -252,9 +267,15 @@ async function readingTools(workspace, role, commands) {
         return tools;
     return [...tools, await commandToolFor(workspace, own)];
 }
-// Sent after the workspace's roots, so it stays short: it only says how to find one's way
-// around the repository without reading all of it.
-const repositoryGuidance = 'Start with AGENTS.md and the architecture document it links to learn where things live, then read only the files your task needs. When you know you need several files, or several searches, ask for all of them in one answer rather than one at a time: each answer is a step, and steps are limited.';
+const batching = 'When you know you need several files, or several searches, ask for all of them in one answer rather than one at a time: each answer is a step, and steps are limited.';
+// How each role finds its way, sent after the workspace's roots, so it stays short. Only the
+// planner learns the repository; the executor and the reviewer work from its plan, so a plan
+// that leaves them guessing is the planner's to fix, not theirs to make up for.
+const repositoryGuidance = {
+    planner: `Start with AGENTS.md, which says where the architecture and the decisions are, and read only the parts your task needs. Name in the plan every file to change and what to change in it, so the executor needs nothing else. ${batching}`,
+    executor: `Work from the plan: read the files it names, and only what you need to change them. Do not study the repository's architecture or decisions; the plan already applied them. ${batching}`,
+    reviewer: `Work from the plan and the changes: read what they touch, and only what you need to judge them. ${batching}`,
+};
 /**
  * Builds the planner → executor → reviewer workflow with every role on the model APIs, each
  * holding its role's instructions, or `systemPrompts`, as its system prompt, followed by the
@@ -280,17 +301,24 @@ const repositoryGuidance = 'Start with AGENTS.md and the architecture document i
  * @throws {UnrecoverableError} when a role cannot be set up, for example because
  * `OPENAI_API_KEY` is missing, or when no program can read the repository.
  */
-export async function createLLMOrchestrator({ provider, autoApprove, systemPrompts = {}, workspace = { roots: [{ path: process.cwd(), access: 'read' }] }, maxSteps, commands, summarizeUndone = true, logger = new Logger(), } = {}) {
+export async function createLLMOrchestrator({ provider, autoApprove, systemPrompts = {}, workspace = { roots: [{ path: process.cwd(), access: 'read' }] }, maxSteps, compactAt, commands, summarizeUndone = true, logger = new Logger(), } = {}) {
     const roles = orchestratorRolesFor(provider);
-    const steps = stepsPerRole(maxSteps);
+    const steps = perRole('maxSteps', maxSteps, defaultRoleSteps, {
+        unit: 'steps',
+        example: '{ "executor": 120 }',
+    });
+    const compaction = perRole('compactAt', compactAt, defaultRoleCompaction, {
+        unit: 'tokens',
+        example: '{ "executor": 200000 }',
+    });
     const opened = await openWorkspace(workspace, logger);
     const guidance = workspaceGuidance(workspace.roots);
     const summarizer = summarizeUndone ? summarizerFor(roles.planner.provider, logger) : undefined;
     const shared = { history: opened.store, ...(summarizer ? { summarizer } : {}), logger };
     // A caller's own prompt replaces our instructions and guidance, not the workspace: the model
     // needs the roots to name a file, as `createLLMAgent` gives them.
-    const prompt = (own, instructions) => own === undefined
-        ? withGuidance(instructions, `${guidance}\n${repositoryGuidance}`)
+    const prompt = (role, own, instructions) => own === undefined
+        ? withGuidance(instructions, `${guidance}\n${repositoryGuidance[role]}`)
         : withGuidance(own, guidance);
     // Every command, checked once, before any role takes its share.
     if (commands !== undefined)
@@ -301,6 +329,7 @@ export async function createLLMOrchestrator({ provider, autoApprove, systemPromp
         systemPrompt,
         tools: await readingTools(opened, role, commands),
         maxSteps: steps[role],
+        compactAt: compaction[role],
         ...shared,
     });
     const { executor } = roles;
@@ -309,17 +338,18 @@ export async function createLLMOrchestrator({ provider, autoApprove, systemPromp
     // with a command can do.
     const executorRuns = executorToolset.some(tool => tool.name === 'runCommand');
     return new OrchestratorAgent({
-        plannerAgent: await reader(roles.planner, 'planner', prompt(systemPrompts.planner, plannerInstructions)),
+        plannerAgent: await reader(roles.planner, 'planner', prompt('planner', systemPrompts.planner, plannerInstructions)),
         executorAgent: buildLLMAgent(executor.provider, {
             model: executor.model,
             reasoningEffort: executor.reasoningEffort,
-            systemPrompt: prompt(systemPrompts.executor, executorRuns ? executorInstructions : executorInstructionsWithoutCommands),
+            systemPrompt: prompt('executor', systemPrompts.executor, executorRuns ? executorInstructions : executorInstructionsWithoutCommands),
             tools: executorToolset,
             maxSteps: steps.executor,
+            compactAt: compaction.executor,
             autoApprove,
             ...shared,
         }),
-        reviewerAgent: await reader(roles.reviewer, 'reviewer', prompt(systemPrompts.reviewer, executorRuns ? reviewerInstructions : reviewerInstructionsWithoutCommands)),
+        reviewerAgent: await reader(roles.reviewer, 'reviewer', prompt('reviewer', systemPrompts.reviewer, executorRuns ? reviewerInstructions : reviewerInstructionsWithoutCommands)),
         reviewerDecisionValidator: new ReviewerDecisionValidator(),
         logger,
     });

package/dist/factory/infrastructure/agentLLMFactory.js.map

@@ -1 +1 @@
-{"version":3,"file":"agentLLMFactory.js","sourceRoot":"","sources":["../../../src/factory/infrastructure/agentLLMFactory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGpC,OAAO,EAAE,QAAQ,EAAE,MAAM,oCAAoC,CAAC;AAE9D,OAAO,EACN,eAAe,EACf,eAAe,EACf,yBAAyB,GACzB,MAAM,yCAAyC,CAAC;AACjD,OAAO,EACN,eAAe,EACf,YAAY,EACZ,sBAAsB,GACtB,MAAM,yCAAyC,CAAC;AACjD,OAAO,EACN,oBAAoB,EACpB,mCAAmC,EACnC,iBAAiB,EACjB,mBAAmB,EACnB,oBAAoB,EACpB,mCAAmC,GACnC,MAAM,mDAAmD,CAAC;AAC3D,OAAO,EAAE,yBAAyB,EAAE,MAAM,mEAAmE,CAAC;AAE9G,OAAO,EAAE,aAAa,EAAE,MAAM,uCAAuC,CAAC;AACtE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AACpE,OAAO,EAAE,OAAO,EAAE,MAAM,4BAA4B,CAAC;AAErD,OAAO,EAAE,MAAM,EAAE,MAAM,mCAAmC,CAAC;AAE3D,OAAO,EAGN,WAAW,EACX,gBAAgB,GAChB,MAAM,4BAA4B,CAAC;AAGpC,OAAO,EACN,cAAc,EAEd,YAAY,EAEZ,SAAS,EACT,aAAa,EACb,gBAAgB,GAEhB,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EAAE,qBAAqB,EAAE,MAAM,2CAA2C,CAAC;AAClF,OAAO,EAAE,qBAAqB,EAAE,MAAM,2CAA2C,CAAC;AAClF,OAAO,EAAE,oBAAoB,EAAE,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAE,eAAe,EAAE,oBAAoB,EAAa,MAAM,mBAAmB,CAAC;AACrF,OAAO,EACN,cAAc,GAKd,MAAM,YAAY,CAAC;AAEpB,oDAAoD;AACpD,MAAM,CAAC,MAAM,SAAS,GAAG;IACxB,SAAS,EAAE,eAAe;IAC1B,MAAM,EAAE,YAAY;CACsD,CAAC;AAE5E,+DAA+D;AAC/D,MAAM,CAAC,MAAM,mBAAmB,GAAG;IAClC,SAAS,EAAE,yBAAyB;IACpC,MAAM,EAAE,sBAAsB;CACiD,CAAC;AAEjF,4EAA4E;AAC5E,MAAM,UAAU,YAAY,CAC3B,QAAkB,EAClB,KAAa;IAEb,OAAO,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC,CAAC;AAC5C,CAAC;AAED,uFAAuF;AACvF,MAAM,UAAU,sBAAsB,CACrC,QAAkB,EAClB,KAAa;IAEb,OAAO,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC,CAAC;AACtD,CAAC;AA+CD,8FAA8F;AAC9F,MAAM,gBAAgB,GAAqC;IAC1D,OAAO,EAAE,EAAE;IACX,QAAQ,EAAE,EAAE;IACZ,QAAQ,EAAE,EAAE;CACZ,CAAC;AAEF;;;;;GAKG;AACH,SAAS,YAAY,CAAC,QAAiB;IACtC,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,gBAAgB,CAAC;IACpD,IAAI,OAAO,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAClC,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC;IACtE,CAAC;IACD,uFAAuF;IACvF,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QAClF,MAAM,IAAI,uBAAuB,CAChC,uFAAuF,CACvF,CAAC;IACH,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,gBAAgB,CAAC,CAAC,CAAC;IAClF,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,MAAM,IAAI,uBAAuB,CAChC,2CAA2C,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,0CAA0C,CACvG,CAAC;IACH,CAAC;IACD,8EAA8E;IAC9E,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,QAAmC,CAAC,CAAC,MAAM,CACvE,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,KAAK,KAAK,SAAS,CAClC,CAAC;IACF,OAAO,EAAE,GAAG,gBAAgB,EAAE,GAAG,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC;AAC9D,CAAC;AA6BD,MAAM,qBAAqB,GAAG;IAC7B,SAAS,EAAE,QAAQ;IACnB,MAAM,EAAE,cAAc;CAC+B,CAAC;AAEvD,SAAS,eAAe,CAAC,QAAe;IACvC,OAAO,IAAI,uBAAuB,CACjC,IAAI,MAAM,CAAC,QAAQ,CAAC,gDAAgD,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAC/F,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CACpB,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,YAAY,EACZ,MAAM,GAMN;IAED,QAAQ,QAAQ,EAAE,CAAC;QAClB,KAAK,WAAW;YACf,OAAO,IAAI,eAAe,CAAC;gBAC1B,KAAK,EAAE,KAAK,IAAI,qBAAqB,CAAC,SAAS;gBAC/C,eAAe;gBACf,YAAY;gBACZ,MAAM;aACN,CAAC,CAAC;QACJ,KAAK,QAAQ;YACZ,OAAO,IAAI,eAAe,CAAC;gBAC1B,KAAK,EAAE,KAAK,IAAI,qBAAqB,CAAC,MAAM;gBAC5C,eAAe;gBACf,YAAY;gBACZ,MAAM;aACN,CAAC,CAAC;QACJ;YACC,gCAAgC;YAChC,MAAM,eAAe,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;AACF,CAAC;AAED,sEAAsE;AACtE,SAAS,aAAa,CAAC,QAAuB,EAAE,MAAe;IAC9D,OAAO,QAAQ,KAAK,WAAW;QAC9B,CAAC,CAAC,IAAI,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC;QACnE,CAAC,CAAC,IAAI,eAAe,CAAC;YACpB,KAAK,EAAE,cAAc;YACrB,eAAe,EAAE,MAAM;YACvB,YAAY,EAAE,EAAE;YAChB,MAAM;SACN,CAAC,CAAC;AACN,CAAC;AAED;;;GAGG;AACH,SAAS,SAAS,CAAC,SAA2B;IAC7C,sFAAsF;IACtF,IAAI,SAAS,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClC,MAAM,IAAI,uBAAuB,CAAC,qCAAqC,CAAC,CAAC;IAC1E,CAAC;IACD,IAAI,KAAyC,CAAC;IAC9C,MAAM,IAAI,GAAG,GAAG,EAAE;QACjB,KAAK,KAAK,gBAAgB,CAAC,SAAS,CAAC;aACnC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,SAAS,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;aAClD,KAAK,CAAC,CAAC,KAAc,EAAE,EAAE;YACzB,+EAA+E;YAC/E,KAAK,GAAG,SAAS,CAAC;YAClB,MAAM,KAAK,CAAC;QACb,CAAC,CAAC,CAAC;QACJ,OAAO,KAAK,CAAC;IACd,CAAC,CAAC;IACF,OAAO;QACN,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,QAAQ,EAAE;QAC/C,aAAa,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,aAAa,EAAE;QACzD,OAAO,EAAE,KAAK,EAAC,KAAK,EAAC,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC;KACrD,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAsB;IAChD,MAAM,CAAC,KAAK,EAAE,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IACtF,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,MAAM,SAAS,GAAG,CAAC,KAAK,EAAE,GAAG,MAAM,CAAC,CAAC,GAAG,CACvC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,KAAK,IAAI,GAAG,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,cAAc,EAAE,CAC5E,CAAC;IACF,OAAO;QACN,gBAAgB;QAChB,GAAG,SAAS;QACZ,uCAAuC,KAAK,CAAC,IAAI,2HAA2H;KAC5K,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACd,CAAC;AAED,SAAS,YAAY,CAAC,YAAoB,EAAE,QAAgB;IAC3D,OAAO,QAAQ,KAAK,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,YAAY,OAAO,QAAQ,EAAE,CAAC;AAC1E,CAAC;AAED,0EAA0E;AAC1E,SAAS,aAAa,CACrB,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,YAAY,EACZ,KAAK,EACL,OAAO,EACP,UAAU,EACV,QAAQ,EACR,WAAW,EACX,MAAM,GAWN;IAED,MAAM,SAAS,GAAG,YAAY,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,EAAE,CAAC,CAAC;IAC3F,0FAA0F;IAC1F,OAAO,IAAI,aAAa,CAAC;QACxB,KAAK,EAAE,IAAI,QAAQ,CAAC;YACnB,SAAS;YACT,KAAK;YACL,WAAW;YACX,MAAM;YACN,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/B,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;SAC/C,CAAC;QACF,MAAM;QACN,YAAY,EAAE,KAAK;KACnB,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,cAAc,CAC7B,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,YAAY,EACZ,KAAK,GAAG,EAAE,EACV,SAAS,EACT,eAAe,GAAG,IAAI,EACtB,QAAQ,EACR,WAAW,EACX,MAAM,GAAG,IAAI,MAAM,EAAE,GACE;IAExB,OAAO,aAAa,CAAC,QAAQ,EAAE;QAC9B,KAAK;QACL,eAAe;QACf,YAAY,EAAE,SAAS;YACtB,CAAC,CAAC,YAAY,CAAC,YAAY,EAAE,iBAAiB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YAChE,CAAC,CAAC,YAAY;QACf,uDAAuD;QACvD,KAAK,EAAE,KAAoB;QAC3B,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvD,GAAG,CAAC,SAAS,IAAI,eAAe,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,aAAa,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACxF,QAAQ;QACR,WAAW;QACX,MAAM;KACN,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACpC,QAAuB,EACvB,SAA2B,EAC3B,EAAE,MAAM,GAAG,IAAI,MAAM,EAAE,EAAE,QAAQ,KAAgD,EAAE;IAEnF,gCAAgC;IAChC,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC;QAAE,MAAM,eAAe,CAAC,QAAQ,CAAC,CAAC;IAChE,IAAI,QAAQ,KAAK,SAAS;QAAE,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IACvD,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACtD,OAAO,CAAC,MAAM,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC,CAA0B,CAAC;AACnF,CAAC;AAED,2EAA2E;AAC3E,KAAK,UAAU,aAAa,CAC3B,QAAuB,EACvB,SAA0B,EAC1B,QAA8B;IAE9B,MAAM,KAAK,GAAG,YAAY,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAChD,sEAAsE;IACtE,IAAI,QAAQ,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC/E,OAAO,CAAC,GAAG,KAAK,EAAE,MAAM,cAAc,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,YAAY,CAC1B,SAA0B,EAC1B,IAAiB,EACjB,QAA8B;IAE9B,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,SAAS,CAAC;IAC1C,MAAM,OAAO,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACpF,MAAM,KAAK,GAAG;QACb,GAAG,OAAO;QACV,qBAAqB,CAAC,MAAM,CAAC;QAC7B,qBAAqB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;KACxC,CAAC;IACF,MAAM,GAAG,GAAG,QAAQ,IAAI,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IACpD,IAAI,CAAC,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACxD,OAAO,CAAC,GAAG,KAAK,EAAE,MAAM,cAAc,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,CAAC;AACzD,CAAC;AAED,0FAA0F;AAC1F,mDAAmD;AACnD,MAAM,kBAAkB,GACvB,ySAAyS,CAAC;AAE3S;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,EAC3C,QAAQ,EACR,WAAW,EACX,aAAa,GAAG,EAAE,EAClB,SAAS,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,EAAE,EAChE,QAAQ,EACR,QAAQ,EACR,eAAe,GAAG,IAAI,EACtB,MAAM,GAAG,IAAI,MAAM,EAAE,MACY,EAAE;IACnC,MAAM,KAAK,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;IACrC,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,iBAAiB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACpD,MAAM,UAAU,GAAG,eAAe,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/F,MAAM,MAAM,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;IACxF,6FAA6F;IAC7F,kEAAkE;IAClE,MAAM,MAAM,GAAG,CAAC,GAAuB,EAAE,YAAoB,EAAE,EAAE,CAChE,GAAG,KAAK,SAAS;QAChB,CAAC,CAAC,YAAY,CAAC,YAAY,EAAE,GAAG,QAAQ,KAAK,kBAAkB,EAAE,CAAC;QAClE,CAAC,CAAC,YAAY,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;IAChC,gEAAgE;IAChE,IAAI,QAAQ,KAAK,SAAS;QAAE,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IACvD,MAAM,MAAM,GAAG,KAAK,EACnB,EAAE,QAAQ,EAAE,KAAK,EAAE,eAAe,EAAQ,EAC1C,IAAiB,EACjB,YAAoB,EACnB,EAAE,CACH,aAAa,CAAC,QAAQ,EAAE;QACvB,KAAK;QACL,eAAe;QACf,YAAY;QACZ,KAAK,EAAE,MAAM,YAAY,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,CAAC;QACjD,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC;QACrB,GAAG,MAAM;KACT,CAAC,CAAC;IACJ,MAAM,EAAE,QAAQ,EAAE,GAAG,KAAK,CAAC;IAC3B,MAAM,eAAe,GAAG,MAAM,aAAa,CAC1C,QAAQ,CAAC,QAAQ,EACjB,MAAM,EACN,QAAQ,IAAI,WAAW,CAAC,QAAQ,EAAE,UAAU,CAAC,CAC7C,CAAC;IACF,4FAA4F;IAC5F,yBAAyB;IACzB,MAAM,YAAY,GAAG,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC;IAE9E,OAAO,IAAI,iBAAiB,CAAC;QAC5B,YAAY,EAAE,MAAM,MAAM,CACzB,KAAK,CAAC,OAAO,EACb,SAAS,EACT,MAAM,CAAC,aAAa,CAAC,OAAO,EAAE,mBAAmB,CAAC,CAClD;QACD,aAAa,EAAE,aAAa,CAAC,QAAQ,CAAC,QAAQ,EAAE;YAC/C,KAAK,EAAE,QAAQ,CAAC,KAAK;YACrB,eAAe,EAAE,QAAQ,CAAC,eAAe;YACzC,YAAY,EAAE,MAAM,CACnB,aAAa,CAAC,QAAQ,EACtB,YAAY,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,mCAAmC,CACzE;YACD,KAAK,EAAE,eAAe;YACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,WAAW;YACX,GAAG,MAAM;SACT,CAAC;QACF,aAAa,EAAE,MAAM,MAAM,CAC1B,KAAK,CAAC,QAAQ,EACd,UAAU,EACV,MAAM,CACL,aAAa,CAAC,QAAQ,EACtB,YAAY,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,mCAAmC,CACzE,CACD;QACD,yBAAyB,EAAE,IAAI,yBAAyB,EAAE;QAC1D,MAAM;KACN,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { resolve } from 'node:path';\nimport type { Agent } from '#src/agent/domain/agent';\nimport type { HistoryReader } from '#src/engines/domain/model/historyFollower';\nimport { LLMAgent } from '#src/engines/domain/model/llmAgent';\nimport type { LLMClient } from '#src/llm/domain/llm';\nimport {\n\tClaudeLLMClient,\n\tclaudeLLMModels,\n\tclaudeLLMReasoningEfforts,\n} from '#src/llm/infrastructure/claudeLLMClient';\nimport {\n\tOpenAILLMClient,\n\topenAIModels,\n\topenAIReasoningEfforts,\n} from '#src/llm/infrastructure/openAILLMClient';\nimport {\n\texecutorInstructions,\n\texecutorInstructionsWithoutCommands,\n\tOrchestratorAgent,\n\tplannerInstructions,\n\treviewerInstructions,\n\treviewerInstructionsWithoutCommands,\n} from '#src/orchestration/domain/model/orchestratorAgent';\nimport { ReviewerDecisionValidator } from '#src/orchestration/infrastructure/model/reviewerDecisionValidator';\nimport type { RecoveryStore } from '#src/recovery/domain/recoveryStore';\nimport { RetryingAgent } from '#src/retry/domain/model/retryingAgent';\nimport { InvalidAgentConfigError } from '#src/shared/domain/errors';\nimport { isOneOf } from '#src/shared/domain/isOneOf';\nimport type { ILogger } from '#src/shared/domain/logger';\nimport { Logger } from '#src/shared/infrastructure/logger';\nimport type { WorkspaceRoot } from '#src/tools/domain/accessPolicy';\nimport {\n\ttype CommandRole,\n\ttype Commands,\n\tcommandsFor,\n\tvalidateCommands,\n} from '#src/tools/domain/commands';\nimport type { AgentTool } from '#src/tools/domain/preparedCall';\nimport type { Tool } from '#src/tools/domain/tool';\nimport {\n\tcommandToolFor,\n\ttype FileTool,\n\tfileToolsFor,\n\ttype OpenedWorkspace,\n\topenStore,\n\topenWorkspace,\n\tstateDirectoryOf,\n\ttype WorkspaceOptions,\n} from '#src/tools/infrastructure/fileTools';\nimport { createShowChangesTool } from '#src/tools/infrastructure/showChangesTool';\nimport { createTrackedReadFile } from '#src/tools/infrastructure/trackedReadFile';\nimport { createWorkspaceTools } from '#src/tools/infrastructure/workspaceTools';\nimport { isAgentProvider, orchestratorRolesFor, type Role } from './agentFactory.ts';\nimport {\n\tagentProviders,\n\ttype AgentModel,\n\ttype AgentProvider,\n\ttype CreateOrchestratorOptions,\n\ttype ReasoningEffort,\n} from './types.ts';\n\n/** The models each provider's model API accepts. */\nexport const llmModels = {\n\tanthropic: claudeLLMModels,\n\topenai: openAIModels,\n} as const satisfies Readonly<Record<AgentProvider, readonly AgentModel[]>>;\n\n/** The reasoning efforts each provider's model API accepts. */\nexport const llmReasoningEfforts = {\n\tanthropic: claudeLLMReasoningEfforts,\n\topenai: openAIReasoningEfforts,\n} as const satisfies Readonly<Record<AgentProvider, readonly ReasoningEffort[]>>;\n\n/** Narrows a model to one accepted by the selected provider's model API. */\nexport function isLLMModelOf<Provider extends AgentProvider>(\n\tprovider: Provider,\n\tvalue: string,\n): value is (typeof llmModels)[Provider][number] {\n\treturn isOneOf(llmModels[provider], value);\n}\n\n/** Narrows a reasoning effort to one accepted by the selected provider's model API. */\nexport function isLLMReasoningEffortOf<Provider extends AgentProvider>(\n\tprovider: Provider,\n\tvalue: string,\n): value is (typeof llmReasoningEfforts)[Provider][number] {\n\treturn isOneOf(llmReasoningEfforts[provider], value);\n}\n\nexport interface CreateLLMAgentOptions {\n\t/**\n\t * Defaults to `'sonnet'` on Anthropic and `'gpt-5.6-luna'` on OpenAI. The provider's client\n\t * rejects a model it does not support.\n\t */\n\tmodel?: AgentModel;\n\t/** Defaults to `'high'`. The provider's client rejects an effort it does not support. */\n\treasoningEffort?: ReasoningEffort;\n\t/** Sent as the model's instructions on every call. */\n\tsystemPrompt: string;\n\t/** The tools the model may call: the consumer's own, and those of `createFileTools`. Defaults to none. */\n\ttools?: (Tool | FileTool)[];\n\t/**\n\t * The folders the agent works on, the same as given to `createFileTools`. With it, the agent\n\t * follows the workspace's history: after the user undoes runs, its next run starts from the\n\t * conversation as it was, and is told what was undone. The system prompt is followed by the\n\t * roots and how to name a file in them.\n\t */\n\tworkspace?: WorkspaceOptions;\n\t/**\n\t * With `workspace`, a summary of what was undone, from the provider's cheap model, in place\n\t * of the undone prompts. That is Claude Haiku without thinking, or `gpt-5.6-luna` at high\n\t * effort, and the call is billed with the run's tokens. Defaults to `true`.\n\t */\n\tsummarizeUndone?: boolean;\n\t/**\n\t * How many calls to the model one run may make before it fails. Defaults to 65, with no upper\n\t * bound. A run that asks for exactly the same tool calls 5 times in a row fails before that.\n\t */\n\tmaxSteps?: number;\n\t/**\n\t * Runs every tool call without asking, `destructive` ones included, for CI where no one can\n\t * answer. Without it, a `destructive` call goes to the run's `approve`, or is denied.\n\t */\n\tautoApprove?: boolean;\n\t/** Defaults to warnings on stderr. */\n\tlogger?: ILogger;\n}\n\n/** A limit per role of `createLLMOrchestrator`; a role left out keeps its default. */\nexport type RoleSteps = Partial<Record<OrchestratorRole, number>>;\n\n// The same three roles a command can be given to.\ntype OrchestratorRole = CommandRole;\n\n// The executor reads, edits and checks, so it needs more calls than the roles that only read.\nconst defaultRoleSteps: Record<OrchestratorRole, number> = {\n\tplanner: 65,\n\texecutor: 80,\n\treviewer: 65,\n};\n\n/**\n * Each role's step limit, from one number for all of them or a number per role.\n *\n * @throws {InvalidAgentConfigError} for a value that is neither, and for a role that does not\n * exist, which is a typo that would otherwise leave that role on its default without a word.\n */\nfunction stepsPerRole(maxSteps: unknown): Record<OrchestratorRole, number> {\n\tif (maxSteps === undefined) return defaultRoleSteps;\n\tif (typeof maxSteps === 'number') {\n\t\treturn { planner: maxSteps, executor: maxSteps, reviewer: maxSteps };\n\t}\n\t// Untyped input, such as the CLI's configuration file, can hold `null` or a list here.\n\tif (typeof maxSteps !== 'object' || maxSteps === null || Array.isArray(maxSteps)) {\n\t\tthrow new InvalidAgentConfigError(\n\t\t\t'maxSteps must be a number or an object of steps per role, such as { \"executor\": 120 }',\n\t\t);\n\t}\n\tconst unknown = Object.keys(maxSteps).filter(role => !(role in defaultRoleSteps));\n\tif (unknown.length > 0) {\n\t\tthrow new InvalidAgentConfigError(\n\t\t\t`maxSteps names roles that do not exist: ${unknown.join(', ')}; expected planner, executor or reviewer`,\n\t\t);\n\t}\n\t// A role given as `undefined`, which untyped input can do, keeps its default.\n\tconst given = Object.entries(maxSteps as Record<string, unknown>).filter(\n\t\t([, steps]) => steps !== undefined,\n\t);\n\treturn { ...defaultRoleSteps, ...Object.fromEntries(given) };\n}\n\nexport interface CreateLLMOrchestratorOptions extends CreateOrchestratorOptions {\n\t/**\n\t * The folders the agents work on. The executor changes files only in a `write` root, every\n\t * change recorded in the workspace's history. Defaults to the current working directory,\n\t * read only.\n\t */\n\tworkspace?: WorkspaceOptions;\n\t/**\n\t * How many calls to the model one run of each role may make: one number for every role, or a\n\t * number per role. Defaults to 80 for the executor and 65 for the planner and the reviewer,\n\t * with no upper bound. A run that asks for exactly the same tool calls 5 times in a row fails\n\t * before its limit.\n\t */\n\tmaxSteps?: number | RoleSteps;\n\t/**\n\t * Commands the roles may run, such as the project's tests, through `runCommand`. Each goes to\n\t * the roles its `roles` names, the executor by default. See `createFileTools`. Each needs a\n\t * `write` root to run in.\n\t */\n\tcommands?: Commands;\n\t/**\n\t * A summary of what the user undid, from the planner's provider's cheap model, for every\n\t * role. See `CreateLLMAgentOptions.summarizeUndone`. Defaults to `true`.\n\t */\n\tsummarizeUndone?: boolean;\n}\n\nconst defaultLLMAgentModels = {\n\tanthropic: 'sonnet',\n\topenai: 'gpt-5.6-luna',\n} as const satisfies Record<AgentProvider, AgentModel>;\n\nfunction unknownProvider(provider: never): InvalidAgentConfigError {\n\treturn new InvalidAgentConfigError(\n\t\t`\"${String(provider)}\" is not an agent provider; expected one of: ${agentProviders.join(', ')}`,\n\t);\n}\n\nfunction llmClientFor(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt,\n\t\tlogger,\n\t}: {\n\t\tmodel?: AgentModel;\n\t\treasoningEffort?: ReasoningEffort;\n\t\tsystemPrompt: string;\n\t\tlogger: ILogger;\n\t},\n): LLMClient {\n\tswitch (provider) {\n\t\tcase 'anthropic':\n\t\t\treturn new ClaudeLLMClient({\n\t\t\t\tmodel: model ?? defaultLLMAgentModels.anthropic,\n\t\t\t\treasoningEffort,\n\t\t\t\tsystemPrompt,\n\t\t\t\tlogger,\n\t\t\t});\n\t\tcase 'openai':\n\t\t\treturn new OpenAILLMClient({\n\t\t\t\tmodel: model ?? defaultLLMAgentModels.openai,\n\t\t\t\treasoningEffort,\n\t\t\t\tsystemPrompt,\n\t\t\t\tlogger,\n\t\t\t});\n\t\tdefault:\n\t\t\t// Reachable from untyped input.\n\t\t\tthrow unknownProvider(provider);\n\t}\n}\n\n/** The provider's cheap model, for the summary of what was undone. */\nfunction summarizerFor(provider: AgentProvider, logger: ILogger): LLMClient {\n\treturn provider === 'anthropic'\n\t\t? new ClaudeLLMClient({ model: 'haiku', systemPrompt: '', logger })\n\t\t: new OpenAILLMClient({\n\t\t\t\tmodel: 'gpt-5.6-luna',\n\t\t\t\treasoningEffort: 'high',\n\t\t\t\tsystemPrompt: '',\n\t\t\t\tlogger,\n\t\t\t});\n}\n\n/**\n * The workspace's history, opened on first use, so a factory that builds an agent stays\n * synchronous. Every agent and every file tool of one workspace reads the same files on disk.\n */\nfunction historyOf(workspace: WorkspaceOptions): HistoryReader {\n\t// Checked now: an agent with no root to follow is a setup error, not a first-run one.\n\tif (workspace.roots.length === 0) {\n\t\tthrow new InvalidAgentConfigError('A workspace needs at least one root');\n\t}\n\tlet store: Promise<RecoveryStore> | undefined;\n\tconst open = () => {\n\t\tstore ??= stateDirectoryOf(workspace)\n\t\t\t.then(directory => openStore(workspace, directory))\n\t\t\t.catch((error: unknown) => {\n\t\t\t\t// Not kept: the next run tries again, as one whose disk was full for a moment.\n\t\t\t\tstore = undefined;\n\t\t\t\tthrow error;\n\t\t\t});\n\t\treturn store;\n\t};\n\treturn {\n\t\tlistRuns: async () => (await open()).listRuns(),\n\t\tlistRevisions: async () => (await open()).listRevisions(),\n\t\treadRun: async runId => (await open()).readRun(runId),\n\t};\n}\n\n/**\n * What the model is told about the workspace after its instructions: the roots, which it may\n * change, and how to name a file. A host path in an error sends the model looking for it, so\n * the real root is given here once.\n */\nfunction workspaceGuidance(roots: WorkspaceRoot[]): string {\n\tconst [first, ...others] = roots.map(root => ({ ...root, path: resolve(root.path) }));\n\tif (!first) return '';\n\tconst described = [first, ...others].map(\n\t\t({ path, access }) => `- ${path}${access === 'write' ? '' : ' (read only)'}`,\n\t);\n\treturn [\n\t\t'The workspace:',\n\t\t...described,\n\t\t`Name a file by its path relative to ${first.path}, such as src/index.ts, and a file in another folder of the workspace by its full path. Read a file before you change it.`,\n\t].join('\\n');\n}\n\nfunction withGuidance(systemPrompt: string, guidance: string): string {\n\treturn guidance === '' ? systemPrompt : `${systemPrompt}\\n\\n${guidance}`;\n}\n\n/** An `LLMAgent` on `provider`, wrapped in the harness's retry policy. */\nfunction buildLLMAgent(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt,\n\t\ttools,\n\t\thistory,\n\t\tsummarizer,\n\t\tmaxSteps,\n\t\tautoApprove,\n\t\tlogger,\n\t}: {\n\t\tmodel?: AgentModel;\n\t\treasoningEffort?: ReasoningEffort;\n\t\tsystemPrompt: string;\n\t\ttools: AgentTool[];\n\t\thistory?: HistoryReader;\n\t\tsummarizer?: LLMClient;\n\t\tmaxSteps?: number;\n\t\tautoApprove?: boolean;\n\t\tlogger: ILogger;\n\t},\n): Agent {\n\tconst llmClient = llmClientFor(provider, { model, reasoningEffort, systemPrompt, logger });\n\t// LLMAgent records nothing from a failed call, so the original prompt is the whole story.\n\treturn new RetryingAgent({\n\t\tinner: new LLMAgent({\n\t\t\tllmClient,\n\t\t\ttools,\n\t\t\tautoApprove,\n\t\t\tlogger,\n\t\t\t...(history ? { history } : {}),\n\t\t\t...(summarizer ? { summarizer } : {}),\n\t\t\t...(maxSteps === undefined ? {} : { maxSteps }),\n\t\t}),\n\t\tlogger,\n\t\tnoteFailures: false,\n\t});\n}\n\n/**\n * Builds an agent whose conversation MiKode owns, on a provider's model API, already wrapped in\n * the harness's retry policy. It needs the provider's API key and bills per token, where\n * `createAgent` uses the Agent SDK's login. Each tool judges the risk of each call; a\n * `destructive` one runs only when the run's `approve` allows it, or with `autoApprove`.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, a model or reasoning effort its\n * client does not support, or two tools that share a name.\n * @throws {UnrecoverableError} when the client cannot be set up, for example because\n * `OPENAI_API_KEY` is missing. The Anthropic SDK resolves credentials lazily, so a missing\n * `ANTHROPIC_API_KEY` fails the first run instead, with an `UnrecoverableError`.\n */\nexport function createLLMAgent(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt,\n\t\ttools = [],\n\t\tworkspace,\n\t\tsummarizeUndone = true,\n\t\tmaxSteps,\n\t\tautoApprove,\n\t\tlogger = new Logger(),\n\t}: CreateLLMAgentOptions,\n): Agent {\n\treturn buildLLMAgent(provider, {\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt: workspace\n\t\t\t? withGuidance(systemPrompt, workspaceGuidance(workspace.roots))\n\t\t\t: systemPrompt,\n\t\t// A file tool is a harness tool under its public name.\n\t\ttools: tools as AgentTool[],\n\t\t...(workspace ? { history: historyOf(workspace) } : {}),\n\t\t...(workspace && summarizeUndone ? { summarizer: summarizerFor(provider, logger) } : {}),\n\t\tmaxSteps,\n\t\tautoApprove,\n\t\tlogger,\n\t});\n}\n\n/**\n * The file tools for an agent from `createLLMAgent` on `provider`, over `workspace`, in the\n * provider's own format:\n *\n * - OpenAI: `listFiles`, `searchText`, `readFile` and its native `apply_patch`;\n * - Anthropic: `listFiles`, `searchText`, its native text editor, whose `view` reads, and\n *   `delete_file`.\n *\n * Every one goes through the workspace's access policy: a path outside the roots, a secret,\n * a file `.gitignore` excludes or the harness protects is refused. A change is made only in a\n * `write` root, only from a version the agent read, and is recorded in the workspace's history,\n * so it can be undone. Without a `write` root the tools only read. Give the agent the same\n * `workspace`, so it follows what the history undoes.\n *\n * With `commands`, they include `runCommand`, which runs one of them by name, with only the\n * parameters it declares and never through a shell. Each run asks the approver, showing the\n * program exactly as it will run, and gets a small environment without the provider keys. The\n * files of the `write` roots are captured before it and compared after, so what it changed is\n * in the history too. It is trusted execution, not a sandbox: a command can reach whatever\n * this user can. Not supported on Windows yet.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, a workspace with no root, or an\n * invalid command, including one whose `cwd` is not in a `write` root, or any on Windows.\n * @throws {UnrecoverableError} when no program can read the first root.\n */\nexport async function createFileTools(\n\tprovider: AgentProvider,\n\tworkspace: WorkspaceOptions,\n\t{ logger = new Logger(), commands }: { logger?: ILogger; commands?: Commands } = {},\n): Promise<FileTool[]> {\n\t// Reachable from untyped input.\n\tif (!isAgentProvider(provider)) throw unknownProvider(provider);\n\tif (commands !== undefined) validateCommands(commands);\n\tconst opened = await openWorkspace(workspace, logger);\n\treturn (await executorTools(provider, opened, commands)) as unknown as FileTool[];\n}\n\n/** The executor's file tools, and `runCommand` when there are commands. */\nasync function executorTools(\n\tprovider: AgentProvider,\n\tworkspace: OpenedWorkspace,\n\tcommands: Commands | undefined,\n): Promise<AgentTool[]> {\n\tconst tools = fileToolsFor(provider, workspace);\n\t// Every command may belong to other roles, leaving none for this one.\n\tif (commands === undefined || Object.keys(commands).length === 0) return tools;\n\treturn [...tools, await commandToolFor(workspace, commands)];\n}\n\n/**\n * What the planner and the reviewer read with: no edits, the current task's changes, and the\n * commands given to their role.\n */\nasync function readingTools(\n\tworkspace: OpenedWorkspace,\n\trole: CommandRole,\n\tcommands: Commands | undefined,\n): Promise<AgentTool[]> {\n\tconst { read, policy, store } = workspace;\n\tconst finding = createWorkspaceTools(read).filter(tool => tool.name !== 'readFile');\n\tconst tools = [\n\t\t...finding,\n\t\tcreateTrackedReadFile(policy),\n\t\tcreateShowChangesTool({ store, policy }),\n\t];\n\tconst own = commands && commandsFor(commands, role);\n\tif (!own || Object.keys(own).length === 0) return tools;\n\treturn [...tools, await commandToolFor(workspace, own)];\n}\n\n// Sent after the workspace's roots, so it stays short: it only says how to find one's way\n// around the repository without reading all of it.\nconst repositoryGuidance =\n\t'Start with AGENTS.md and the architecture document it links to learn where things live, then read only the files your task needs. When you know you need several files, or several searches, ask for all of them in one answer rather than one at a time: each answer is a step, and steps are limited.';\n\n/**\n * Builds the planner → executor → reviewer workflow with every role on the model APIs, each\n * holding its role's instructions, or `systemPrompts`, as its system prompt, followed by the\n * workspace's roots.\n *\n * - The planner and the reviewer read: `listFiles`, `searchText`, `readFile`, and\n *   `showChanges`, the current task's changes as far as each may read them.\n * - The executor gets its provider's file tools (see `createFileTools`), and changes files only\n *   in a `write` root; `autoApprove` reaches only it.\n * - Each role has its own step limit: `maxSteps`, 80 for the executor and 65 for the others by\n *   default.\n * - Each role gets `runCommand` with the `commands` whose `roles` name it, the executor by\n *   default. A command asks the approver in any role, and its files are captured around it.\n * - The executor is told to finish only once the tests and checks pass, and the reviewer that\n *   they passed, when the executor has a command; without one, both are told the tests were\n *   not run, and the reviewer judges from the code whether they would pass.\n *\n * One run of the orchestrator is one run of the workspace's history, every round and retry\n * included, and its `AgentResponse` names it. Every role follows what the user undoes.\n * Asynchronous because opening the workspace is.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, or a workspace with no root.\n * @throws {UnrecoverableError} when a role cannot be set up, for example because\n * `OPENAI_API_KEY` is missing, or when no program can read the repository.\n */\nexport async function createLLMOrchestrator({\n\tprovider,\n\tautoApprove,\n\tsystemPrompts = {},\n\tworkspace = { roots: [{ path: process.cwd(), access: 'read' }] },\n\tmaxSteps,\n\tcommands,\n\tsummarizeUndone = true,\n\tlogger = new Logger(),\n}: CreateLLMOrchestratorOptions = {}): Promise<Agent> {\n\tconst roles = orchestratorRolesFor(provider);\n\tconst steps = stepsPerRole(maxSteps);\n\tconst opened = await openWorkspace(workspace, logger);\n\tconst guidance = workspaceGuidance(workspace.roots);\n\tconst summarizer = summarizeUndone ? summarizerFor(roles.planner.provider, logger) : undefined;\n\tconst shared = { history: opened.store, ...(summarizer ? { summarizer } : {}), logger };\n\t// A caller's own prompt replaces our instructions and guidance, not the workspace: the model\n\t// needs the roots to name a file, as `createLLMAgent` gives them.\n\tconst prompt = (own: string | undefined, instructions: string) =>\n\t\town === undefined\n\t\t\t? withGuidance(instructions, `${guidance}\\n${repositoryGuidance}`)\n\t\t\t: withGuidance(own, guidance);\n\t// Every command, checked once, before any role takes its share.\n\tif (commands !== undefined) validateCommands(commands);\n\tconst reader = async (\n\t\t{ provider, model, reasoningEffort }: Role,\n\t\trole: CommandRole,\n\t\tsystemPrompt: string,\n\t) =>\n\t\tbuildLLMAgent(provider, {\n\t\t\tmodel,\n\t\t\treasoningEffort,\n\t\t\tsystemPrompt,\n\t\t\ttools: await readingTools(opened, role, commands),\n\t\t\tmaxSteps: steps[role],\n\t\t\t...shared,\n\t\t});\n\tconst { executor } = roles;\n\tconst executorToolset = await executorTools(\n\t\texecutor.provider,\n\t\topened,\n\t\tcommands && commandsFor(commands, 'executor'),\n\t);\n\t// The executor and the reviewer are told whether the tests were run, which only an executor\n\t// with a command can do.\n\tconst executorRuns = executorToolset.some(tool => tool.name === 'runCommand');\n\n\treturn new OrchestratorAgent({\n\t\tplannerAgent: await reader(\n\t\t\troles.planner,\n\t\t\t'planner',\n\t\t\tprompt(systemPrompts.planner, plannerInstructions),\n\t\t),\n\t\texecutorAgent: buildLLMAgent(executor.provider, {\n\t\t\tmodel: executor.model,\n\t\t\treasoningEffort: executor.reasoningEffort,\n\t\t\tsystemPrompt: prompt(\n\t\t\t\tsystemPrompts.executor,\n\t\t\t\texecutorRuns ? executorInstructions : executorInstructionsWithoutCommands,\n\t\t\t),\n\t\t\ttools: executorToolset,\n\t\t\tmaxSteps: steps.executor,\n\t\t\tautoApprove,\n\t\t\t...shared,\n\t\t}),\n\t\treviewerAgent: await reader(\n\t\t\troles.reviewer,\n\t\t\t'reviewer',\n\t\t\tprompt(\n\t\t\t\tsystemPrompts.reviewer,\n\t\t\t\texecutorRuns ? reviewerInstructions : reviewerInstructionsWithoutCommands,\n\t\t\t),\n\t\t),\n\t\treviewerDecisionValidator: new ReviewerDecisionValidator(),\n\t\tlogger,\n\t});\n}\n"]}
+{"version":3,"file":"agentLLMFactory.js","sourceRoot":"","sources":["../../../src/factory/infrastructure/agentLLMFactory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGpC,OAAO,EAAE,QAAQ,EAAE,MAAM,oCAAoC,CAAC;AAE9D,OAAO,EACN,eAAe,EACf,eAAe,EACf,yBAAyB,GACzB,MAAM,yCAAyC,CAAC;AACjD,OAAO,EACN,eAAe,EACf,YAAY,EACZ,sBAAsB,GACtB,MAAM,yCAAyC,CAAC;AACjD,OAAO,EACN,oBAAoB,EACpB,mCAAmC,EACnC,iBAAiB,EACjB,mBAAmB,EACnB,oBAAoB,EACpB,mCAAmC,GACnC,MAAM,mDAAmD,CAAC;AAC3D,OAAO,EAAE,yBAAyB,EAAE,MAAM,mEAAmE,CAAC;AAE9G,OAAO,EAAE,aAAa,EAAE,MAAM,uCAAuC,CAAC;AACtE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AACpE,OAAO,EAAE,OAAO,EAAE,MAAM,4BAA4B,CAAC;AAErD,OAAO,EAAE,MAAM,EAAE,MAAM,mCAAmC,CAAC;AAE3D,OAAO,EAGN,WAAW,EACX,gBAAgB,GAChB,MAAM,4BAA4B,CAAC;AAGpC,OAAO,EACN,cAAc,EAEd,YAAY,EAEZ,SAAS,EACT,aAAa,EACb,gBAAgB,GAEhB,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EAAE,qBAAqB,EAAE,MAAM,2CAA2C,CAAC;AAClF,OAAO,EAAE,qBAAqB,EAAE,MAAM,2CAA2C,CAAC;AAClF,OAAO,EAAE,oBAAoB,EAAE,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAE,eAAe,EAAE,oBAAoB,EAAa,MAAM,mBAAmB,CAAC;AACrF,OAAO,EACN,cAAc,GAKd,MAAM,YAAY,CAAC;AAEpB,oDAAoD;AACpD,MAAM,CAAC,MAAM,SAAS,GAAG;IACxB,SAAS,EAAE,eAAe;IAC1B,MAAM,EAAE,YAAY;CACsD,CAAC;AAE5E,+DAA+D;AAC/D,MAAM,CAAC,MAAM,mBAAmB,GAAG;IAClC,SAAS,EAAE,yBAAyB;IACpC,MAAM,EAAE,sBAAsB;CACiD,CAAC;AAEjF,4EAA4E;AAC5E,MAAM,UAAU,YAAY,CAC3B,QAAkB,EAClB,KAAa;IAEb,OAAO,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC,CAAC;AAC5C,CAAC;AAED,uFAAuF;AACvF,MAAM,UAAU,sBAAsB,CACrC,QAAkB,EAClB,KAAa;IAEb,OAAO,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC,CAAC;AACtD,CAAC;AAyDD,8FAA8F;AAC9F,MAAM,gBAAgB,GAAqC;IAC1D,OAAO,EAAE,EAAE;IACX,QAAQ,EAAE,EAAE;IACZ,QAAQ,EAAE,EAAE;CACZ,CAAC;AAEF,oCAAoC;AACpC,MAAM,qBAAqB,GAAiD;IAC3E,OAAO,EAAE,SAAS;IAClB,QAAQ,EAAE,SAAS;IACnB,QAAQ,EAAE,SAAS;CACnB,CAAC;AAEF;;;;;;GAMG;AACH,SAAS,OAAO,CACf,IAAY,EACZ,KAAc,EACd,QAAqC,EACrC,EAAE,IAAI,EAAE,OAAO,EAAqC;IAEpD,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IACzC,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;IAC3F,uFAAuF;IACvF,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzE,MAAM,IAAI,uBAAuB,CAChC,GAAG,IAAI,qCAAqC,IAAI,sBAAsB,OAAO,EAAE,CAC/E,CAAC;IACH,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC;IACvE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,MAAM,IAAI,uBAAuB,CAChC,GAAG,IAAI,mCAAmC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,0CAA0C,CACtG,CAAC;IACH,CAAC;IACD,8EAA8E;IAC9E,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,KAAgC,CAAC,CAAC,MAAM,CACpE,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,SAAS,CAChC,CAAC;IACF,OAAO,EAAE,GAAG,QAAQ,EAAE,GAAG,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC;AACtD,CAAC;AAmCD,MAAM,qBAAqB,GAAG;IAC7B,SAAS,EAAE,QAAQ;IACnB,MAAM,EAAE,cAAc;CAC+B,CAAC;AAEvD,SAAS,eAAe,CAAC,QAAe;IACvC,OAAO,IAAI,uBAAuB,CACjC,IAAI,MAAM,CAAC,QAAQ,CAAC,gDAAgD,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAC/F,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CACpB,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,SAAS,EACT,YAAY,EACZ,MAAM,GAON;IAED,QAAQ,QAAQ,EAAE,CAAC;QAClB,KAAK,WAAW;YACf,OAAO,IAAI,eAAe,CAAC;gBAC1B,KAAK,EAAE,KAAK,IAAI,qBAAqB,CAAC,SAAS;gBAC/C,eAAe;gBACf,SAAS;gBACT,YAAY;gBACZ,MAAM;aACN,CAAC,CAAC;QACJ,KAAK,QAAQ;YACZ,OAAO,IAAI,eAAe,CAAC;gBAC1B,KAAK,EAAE,KAAK,IAAI,qBAAqB,CAAC,MAAM;gBAC5C,eAAe;gBACf,SAAS;gBACT,YAAY;gBACZ,MAAM;aACN,CAAC,CAAC;QACJ;YACC,gCAAgC;YAChC,MAAM,eAAe,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;AACF,CAAC;AAED,sEAAsE;AACtE,SAAS,aAAa,CAAC,QAAuB,EAAE,MAAe;IAC9D,OAAO,QAAQ,KAAK,WAAW;QAC9B,CAAC,CAAC,IAAI,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC;QACnE,CAAC,CAAC,IAAI,eAAe,CAAC;YACpB,KAAK,EAAE,cAAc;YACrB,eAAe,EAAE,MAAM;YACvB,YAAY,EAAE,EAAE;YAChB,MAAM;SACN,CAAC,CAAC;AACN,CAAC;AAED;;;GAGG;AACH,SAAS,SAAS,CAAC,SAA2B;IAC7C,sFAAsF;IACtF,IAAI,SAAS,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClC,MAAM,IAAI,uBAAuB,CAAC,qCAAqC,CAAC,CAAC;IAC1E,CAAC;IACD,IAAI,KAAyC,CAAC;IAC9C,MAAM,IAAI,GAAG,GAAG,EAAE;QACjB,KAAK,KAAK,gBAAgB,CAAC,SAAS,CAAC;aACnC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,SAAS,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;aAClD,KAAK,CAAC,CAAC,KAAc,EAAE,EAAE;YACzB,+EAA+E;YAC/E,KAAK,GAAG,SAAS,CAAC;YAClB,MAAM,KAAK,CAAC;QACb,CAAC,CAAC,CAAC;QACJ,OAAO,KAAK,CAAC;IACd,CAAC,CAAC;IACF,OAAO;QACN,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,QAAQ,EAAE;QAC/C,aAAa,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,aAAa,EAAE;QACzD,OAAO,EAAE,KAAK,EAAC,KAAK,EAAC,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC;KACrD,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAsB;IAChD,MAAM,CAAC,KAAK,EAAE,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IACtF,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,MAAM,SAAS,GAAG,CAAC,KAAK,EAAE,GAAG,MAAM,CAAC,CAAC,GAAG,CACvC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,KAAK,IAAI,GAAG,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,cAAc,EAAE,CAC5E,CAAC;IACF,OAAO;QACN,gBAAgB;QAChB,GAAG,SAAS;QACZ,uCAAuC,KAAK,CAAC,IAAI,2HAA2H;KAC5K,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACd,CAAC;AAED,SAAS,YAAY,CAAC,YAAoB,EAAE,QAAgB;IAC3D,OAAO,QAAQ,KAAK,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,YAAY,OAAO,QAAQ,EAAE,CAAC;AAC1E,CAAC;AAED,0EAA0E;AAC1E,SAAS,aAAa,CACrB,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,YAAY,EACZ,KAAK,EACL,OAAO,EACP,UAAU,EACV,QAAQ,EACR,SAAS,EACT,WAAW,EACX,MAAM,GAYN;IAED,MAAM,SAAS,GAAG,YAAY,CAAC,QAAQ,EAAE;QACxC,KAAK;QACL,eAAe;QACf,SAAS;QACT,YAAY;QACZ,MAAM;KACN,CAAC,CAAC;IACH,0FAA0F;IAC1F,OAAO,IAAI,aAAa,CAAC;QACxB,KAAK,EAAE,IAAI,QAAQ,CAAC;YACnB,SAAS;YACT,KAAK;YACL,WAAW;YACX,MAAM;YACN,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/B,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;SAC/C,CAAC;QACF,MAAM;QACN,YAAY,EAAE,KAAK;KACnB,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,cAAc,CAC7B,QAAuB,EACvB,EACC,KAAK,EACL,eAAe,EACf,YAAY,EACZ,KAAK,GAAG,EAAE,EACV,SAAS,EACT,eAAe,GAAG,IAAI,EACtB,QAAQ,EACR,SAAS,EACT,WAAW,EACX,MAAM,GAAG,IAAI,MAAM,EAAE,GACE;IAExB,OAAO,aAAa,CAAC,QAAQ,EAAE;QAC9B,KAAK;QACL,eAAe;QACf,YAAY,EAAE,SAAS;YACtB,CAAC,CAAC,YAAY,CAAC,YAAY,EAAE,iBAAiB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YAChE,CAAC,CAAC,YAAY;QACf,uDAAuD;QACvD,KAAK,EAAE,KAAoB;QAC3B,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvD,GAAG,CAAC,SAAS,IAAI,eAAe,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,aAAa,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACxF,QAAQ;QACR,SAAS;QACT,WAAW;QACX,MAAM;KACN,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACpC,QAAuB,EACvB,SAA2B,EAC3B,EAAE,MAAM,GAAG,IAAI,MAAM,EAAE,EAAE,QAAQ,KAAgD,EAAE;IAEnF,gCAAgC;IAChC,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC;QAAE,MAAM,eAAe,CAAC,QAAQ,CAAC,CAAC;IAChE,IAAI,QAAQ,KAAK,SAAS;QAAE,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IACvD,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACtD,OAAO,CAAC,MAAM,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC,CAA0B,CAAC;AACnF,CAAC;AAED,2EAA2E;AAC3E,KAAK,UAAU,aAAa,CAC3B,QAAuB,EACvB,SAA0B,EAC1B,QAA8B;IAE9B,MAAM,KAAK,GAAG,YAAY,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAChD,sEAAsE;IACtE,IAAI,QAAQ,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC/E,OAAO,CAAC,GAAG,KAAK,EAAE,MAAM,cAAc,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,YAAY,CAC1B,SAA0B,EAC1B,IAAiB,EACjB,QAA8B;IAE9B,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,SAAS,CAAC;IAC1C,MAAM,OAAO,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACpF,MAAM,KAAK,GAAG;QACb,GAAG,OAAO;QACV,qBAAqB,CAAC,MAAM,CAAC;QAC7B,qBAAqB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;KACxC,CAAC;IACF,MAAM,GAAG,GAAG,QAAQ,IAAI,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IACpD,IAAI,CAAC,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACxD,OAAO,CAAC,GAAG,KAAK,EAAE,MAAM,cAAc,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,QAAQ,GACb,uKAAuK,CAAC;AAEzK,6FAA6F;AAC7F,6FAA6F;AAC7F,gFAAgF;AAChF,MAAM,kBAAkB,GAAgC;IACvD,OAAO,EAAE,+NAA+N,QAAQ,EAAE;IAClP,QAAQ,EAAE,+KAA+K,QAAQ,EAAE;IACnM,QAAQ,EAAE,mGAAmG,QAAQ,EAAE;CACvH,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,EAC3C,QAAQ,EACR,WAAW,EACX,aAAa,GAAG,EAAE,EAClB,SAAS,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,EAAE,EAChE,QAAQ,EACR,SAAS,EACT,QAAQ,EACR,eAAe,GAAG,IAAI,EACtB,MAAM,GAAG,IAAI,MAAM,EAAE,MACY,EAAE;IACnC,MAAM,KAAK,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IAC7C,MAAM,KAAK,GAAG,OAAO,CAAC,UAAU,EAAE,QAAQ,EAAE,gBAAgB,EAAE;QAC7D,IAAI,EAAE,OAAO;QACb,OAAO,EAAE,qBAAqB;KAC9B,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,OAAO,CAAC,WAAW,EAAE,SAAS,EAAE,qBAAqB,EAAE;QACzE,IAAI,EAAE,QAAQ;QACd,OAAO,EAAE,wBAAwB;KACjC,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,QAAQ,GAAG,iBAAiB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACpD,MAAM,UAAU,GAAG,eAAe,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/F,MAAM,MAAM,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;IACxF,6FAA6F;IAC7F,kEAAkE;IAClE,MAAM,MAAM,GAAG,CAAC,IAAiB,EAAE,GAAuB,EAAE,YAAoB,EAAE,EAAE,CACnF,GAAG,KAAK,SAAS;QAChB,CAAC,CAAC,YAAY,CAAC,YAAY,EAAE,GAAG,QAAQ,KAAK,kBAAkB,CAAC,IAAI,CAAC,EAAE,CAAC;QACxE,CAAC,CAAC,YAAY,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;IAChC,gEAAgE;IAChE,IAAI,QAAQ,KAAK,SAAS;QAAE,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IACvD,MAAM,MAAM,GAAG,KAAK,EACnB,EAAE,QAAQ,EAAE,KAAK,EAAE,eAAe,EAAQ,EAC1C,IAAiB,EACjB,YAAoB,EACnB,EAAE,CACH,aAAa,CAAC,QAAQ,EAAE;QACvB,KAAK;QACL,eAAe;QACf,YAAY;QACZ,KAAK,EAAE,MAAM,YAAY,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,CAAC;QACjD,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC;QACrB,SAAS,EAAE,UAAU,CAAC,IAAI,CAAC;QAC3B,GAAG,MAAM;KACT,CAAC,CAAC;IACJ,MAAM,EAAE,QAAQ,EAAE,GAAG,KAAK,CAAC;IAC3B,MAAM,eAAe,GAAG,MAAM,aAAa,CAC1C,QAAQ,CAAC,QAAQ,EACjB,MAAM,EACN,QAAQ,IAAI,WAAW,CAAC,QAAQ,EAAE,UAAU,CAAC,CAC7C,CAAC;IACF,4FAA4F;IAC5F,yBAAyB;IACzB,MAAM,YAAY,GAAG,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC;IAE9E,OAAO,IAAI,iBAAiB,CAAC;QAC5B,YAAY,EAAE,MAAM,MAAM,CACzB,KAAK,CAAC,OAAO,EACb,SAAS,EACT,MAAM,CAAC,SAAS,EAAE,aAAa,CAAC,OAAO,EAAE,mBAAmB,CAAC,CAC7D;QACD,aAAa,EAAE,aAAa,CAAC,QAAQ,CAAC,QAAQ,EAAE;YAC/C,KAAK,EAAE,QAAQ,CAAC,KAAK;YACrB,eAAe,EAAE,QAAQ,CAAC,eAAe;YACzC,YAAY,EAAE,MAAM,CACnB,UAAU,EACV,aAAa,CAAC,QAAQ,EACtB,YAAY,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,mCAAmC,CACzE;YACD,KAAK,EAAE,eAAe;YACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,SAAS,EAAE,UAAU,CAAC,QAAQ;YAC9B,WAAW;YACX,GAAG,MAAM;SACT,CAAC;QACF,aAAa,EAAE,MAAM,MAAM,CAC1B,KAAK,CAAC,QAAQ,EACd,UAAU,EACV,MAAM,CACL,UAAU,EACV,aAAa,CAAC,QAAQ,EACtB,YAAY,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,mCAAmC,CACzE,CACD;QACD,yBAAyB,EAAE,IAAI,yBAAyB,EAAE;QAC1D,MAAM;KACN,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { resolve } from 'node:path';\nimport type { Agent } from '#src/agent/domain/agent';\nimport type { HistoryReader } from '#src/engines/domain/model/historyFollower';\nimport { LLMAgent } from '#src/engines/domain/model/llmAgent';\nimport type { LLMClient } from '#src/llm/domain/llm';\nimport {\n\tClaudeLLMClient,\n\tclaudeLLMModels,\n\tclaudeLLMReasoningEfforts,\n} from '#src/llm/infrastructure/claudeLLMClient';\nimport {\n\tOpenAILLMClient,\n\topenAIModels,\n\topenAIReasoningEfforts,\n} from '#src/llm/infrastructure/openAILLMClient';\nimport {\n\texecutorInstructions,\n\texecutorInstructionsWithoutCommands,\n\tOrchestratorAgent,\n\tplannerInstructions,\n\treviewerInstructions,\n\treviewerInstructionsWithoutCommands,\n} from '#src/orchestration/domain/model/orchestratorAgent';\nimport { ReviewerDecisionValidator } from '#src/orchestration/infrastructure/model/reviewerDecisionValidator';\nimport type { RecoveryStore } from '#src/recovery/domain/recoveryStore';\nimport { RetryingAgent } from '#src/retry/domain/model/retryingAgent';\nimport { InvalidAgentConfigError } from '#src/shared/domain/errors';\nimport { isOneOf } from '#src/shared/domain/isOneOf';\nimport type { ILogger } from '#src/shared/domain/logger';\nimport { Logger } from '#src/shared/infrastructure/logger';\nimport type { WorkspaceRoot } from '#src/tools/domain/accessPolicy';\nimport {\n\ttype CommandRole,\n\ttype Commands,\n\tcommandsFor,\n\tvalidateCommands,\n} from '#src/tools/domain/commands';\nimport type { AgentTool } from '#src/tools/domain/preparedCall';\nimport type { Tool } from '#src/tools/domain/tool';\nimport {\n\tcommandToolFor,\n\ttype FileTool,\n\tfileToolsFor,\n\ttype OpenedWorkspace,\n\topenStore,\n\topenWorkspace,\n\tstateDirectoryOf,\n\ttype WorkspaceOptions,\n} from '#src/tools/infrastructure/fileTools';\nimport { createShowChangesTool } from '#src/tools/infrastructure/showChangesTool';\nimport { createTrackedReadFile } from '#src/tools/infrastructure/trackedReadFile';\nimport { createWorkspaceTools } from '#src/tools/infrastructure/workspaceTools';\nimport { isAgentProvider, orchestratorRolesFor, type Role } from './agentFactory.ts';\nimport {\n\tagentProviders,\n\ttype AgentModel,\n\ttype AgentProvider,\n\ttype CreateOrchestratorOptions,\n\ttype ReasoningEffort,\n} from './types.ts';\n\n/** The models each provider's model API accepts. */\nexport const llmModels = {\n\tanthropic: claudeLLMModels,\n\topenai: openAIModels,\n} as const satisfies Readonly<Record<AgentProvider, readonly AgentModel[]>>;\n\n/** The reasoning efforts each provider's model API accepts. */\nexport const llmReasoningEfforts = {\n\tanthropic: claudeLLMReasoningEfforts,\n\topenai: openAIReasoningEfforts,\n} as const satisfies Readonly<Record<AgentProvider, readonly ReasoningEffort[]>>;\n\n/** Narrows a model to one accepted by the selected provider's model API. */\nexport function isLLMModelOf<Provider extends AgentProvider>(\n\tprovider: Provider,\n\tvalue: string,\n): value is (typeof llmModels)[Provider][number] {\n\treturn isOneOf(llmModels[provider], value);\n}\n\n/** Narrows a reasoning effort to one accepted by the selected provider's model API. */\nexport function isLLMReasoningEffortOf<Provider extends AgentProvider>(\n\tprovider: Provider,\n\tvalue: string,\n): value is (typeof llmReasoningEfforts)[Provider][number] {\n\treturn isOneOf(llmReasoningEfforts[provider], value);\n}\n\nexport interface CreateLLMAgentOptions {\n\t/**\n\t * Defaults to `'sonnet'` on Anthropic and `'gpt-5.6-luna'` on OpenAI. The provider's client\n\t * rejects a model it does not support.\n\t */\n\tmodel?: AgentModel;\n\t/** Defaults to `'high'`. The provider's client rejects an effort it does not support. */\n\treasoningEffort?: ReasoningEffort;\n\t/** Sent as the model's instructions on every call. */\n\tsystemPrompt: string;\n\t/** The tools the model may call: the consumer's own, and those of `createFileTools`. Defaults to none. */\n\ttools?: (Tool | FileTool)[];\n\t/**\n\t * The folders the agent works on, the same as given to `createFileTools`. With it, the agent\n\t * follows the workspace's history: after the user undoes runs, its next run starts from the\n\t * conversation as it was, and is told what was undone. The system prompt is followed by the\n\t * roots and how to name a file in them.\n\t */\n\tworkspace?: WorkspaceOptions;\n\t/**\n\t * With `workspace`, a summary of what was undone, from the provider's cheap model, in place\n\t * of the undone prompts. That is Claude Haiku without thinking, or `gpt-5.6-luna` at high\n\t * effort, and the call is billed with the run's tokens. Defaults to `true`.\n\t */\n\tsummarizeUndone?: boolean;\n\t/**\n\t * How many calls to the model one run may make before it fails. Defaults to 65, with no upper\n\t * bound. A run that asks for exactly the same tool calls 5 times in a row fails before that.\n\t */\n\tmaxSteps?: number;\n\t/**\n\t * The input tokens at which the provider compacts the conversation on its side: it replaces\n\t * every turn before that call with its own summary, which only its model reads back. The\n\t * conversation keeps every original turn. Without it, the conversation is never compacted.\n\t * At least 50,000 on Anthropic, where Claude Haiku cannot compact.\n\t */\n\tcompactAt?: number;\n\t/**\n\t * Runs every tool call without asking, `destructive` ones included, for CI where no one can\n\t * answer. Without it, a `destructive` call goes to the run's `approve`, or is denied.\n\t */\n\tautoApprove?: boolean;\n\t/** Defaults to warnings on stderr. */\n\tlogger?: ILogger;\n}\n\n/** A limit per role of `createLLMOrchestrator`; a role left out keeps its default. */\nexport type RoleSteps = Partial<Record<OrchestratorRole, number>>;\n\n/** A number of tokens per role of `createLLMOrchestrator`; a role left out keeps its default. */\nexport type RoleTokens = Partial<Record<OrchestratorRole, number>>;\n\n// The same three roles a command can be given to.\ntype OrchestratorRole = CommandRole;\n\n// The executor reads, edits and checks, so it needs more calls than the roles that only read.\nconst defaultRoleSteps: Record<OrchestratorRole, number> = {\n\tplanner: 65,\n\texecutor: 80,\n\treviewer: 65,\n};\n\n// No role compacts unless asked to.\nconst defaultRoleCompaction: Record<OrchestratorRole, number | undefined> = {\n\tplanner: undefined,\n\texecutor: undefined,\n\treviewer: undefined,\n};\n\n/**\n * Each role's value of the option `name`, from one number for all of them or a number per role.\n * Each number is checked by whatever takes it.\n *\n * @throws {InvalidAgentConfigError} for a value that is neither, and for a role that does not\n * exist, which is a typo that would otherwise leave that role on its default without a word.\n */\nfunction perRole<T extends number | undefined>(\n\tname: string,\n\tvalue: unknown,\n\tdefaults: Record<OrchestratorRole, T>,\n\t{ unit, example }: { unit: string; example: string },\n): Record<OrchestratorRole, T | number> {\n\tif (value === undefined) return defaults;\n\tif (typeof value === 'number') return { planner: value, executor: value, reviewer: value };\n\t// Untyped input, such as the CLI's configuration file, can hold `null` or a list here.\n\tif (typeof value !== 'object' || value === null || Array.isArray(value)) {\n\t\tthrow new InvalidAgentConfigError(\n\t\t\t`${name} must be a number or an object of ${unit} per role, such as ${example}`,\n\t\t);\n\t}\n\tconst unknown = Object.keys(value).filter(role => !(role in defaults));\n\tif (unknown.length > 0) {\n\t\tthrow new InvalidAgentConfigError(\n\t\t\t`${name} names roles that do not exist: ${unknown.join(', ')}; expected planner, executor or reviewer`,\n\t\t);\n\t}\n\t// A role given as `undefined`, which untyped input can do, keeps its default.\n\tconst given = Object.entries(value as Record<string, unknown>).filter(\n\t\t([, each]) => each !== undefined,\n\t);\n\treturn { ...defaults, ...Object.fromEntries(given) };\n}\n\nexport interface CreateLLMOrchestratorOptions extends CreateOrchestratorOptions {\n\t/**\n\t * The folders the agents work on. The executor changes files only in a `write` root, every\n\t * change recorded in the workspace's history. Defaults to the current working directory,\n\t * read only.\n\t */\n\tworkspace?: WorkspaceOptions;\n\t/**\n\t * How many calls to the model one run of each role may make: one number for every role, or a\n\t * number per role. Defaults to 80 for the executor and 65 for the planner and the reviewer,\n\t * with no upper bound. A run that asks for exactly the same tool calls 5 times in a row fails\n\t * before its limit.\n\t */\n\tmaxSteps?: number | RoleSteps;\n\t/**\n\t * The input tokens at which each role's provider compacts its conversation, as\n\t * `CreateLLMAgentOptions.compactAt`: one number for every role, or a number per role. Without\n\t * it, no role compacts.\n\t */\n\tcompactAt?: number | RoleTokens;\n\t/**\n\t * Commands the roles may run, such as the project's tests, through `runCommand`. Each goes to\n\t * the roles its `roles` names, the executor by default. See `createFileTools`. Each needs a\n\t * `write` root to run in.\n\t */\n\tcommands?: Commands;\n\t/**\n\t * A summary of what the user undid, from the planner's provider's cheap model, for every\n\t * role. See `CreateLLMAgentOptions.summarizeUndone`. Defaults to `true`.\n\t */\n\tsummarizeUndone?: boolean;\n}\n\nconst defaultLLMAgentModels = {\n\tanthropic: 'sonnet',\n\topenai: 'gpt-5.6-luna',\n} as const satisfies Record<AgentProvider, AgentModel>;\n\nfunction unknownProvider(provider: never): InvalidAgentConfigError {\n\treturn new InvalidAgentConfigError(\n\t\t`\"${String(provider)}\" is not an agent provider; expected one of: ${agentProviders.join(', ')}`,\n\t);\n}\n\nfunction llmClientFor(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tcompactAt,\n\t\tsystemPrompt,\n\t\tlogger,\n\t}: {\n\t\tmodel?: AgentModel;\n\t\treasoningEffort?: ReasoningEffort;\n\t\tcompactAt?: number | undefined;\n\t\tsystemPrompt: string;\n\t\tlogger: ILogger;\n\t},\n): LLMClient {\n\tswitch (provider) {\n\t\tcase 'anthropic':\n\t\t\treturn new ClaudeLLMClient({\n\t\t\t\tmodel: model ?? defaultLLMAgentModels.anthropic,\n\t\t\t\treasoningEffort,\n\t\t\t\tcompactAt,\n\t\t\t\tsystemPrompt,\n\t\t\t\tlogger,\n\t\t\t});\n\t\tcase 'openai':\n\t\t\treturn new OpenAILLMClient({\n\t\t\t\tmodel: model ?? defaultLLMAgentModels.openai,\n\t\t\t\treasoningEffort,\n\t\t\t\tcompactAt,\n\t\t\t\tsystemPrompt,\n\t\t\t\tlogger,\n\t\t\t});\n\t\tdefault:\n\t\t\t// Reachable from untyped input.\n\t\t\tthrow unknownProvider(provider);\n\t}\n}\n\n/** The provider's cheap model, for the summary of what was undone. */\nfunction summarizerFor(provider: AgentProvider, logger: ILogger): LLMClient {\n\treturn provider === 'anthropic'\n\t\t? new ClaudeLLMClient({ model: 'haiku', systemPrompt: '', logger })\n\t\t: new OpenAILLMClient({\n\t\t\t\tmodel: 'gpt-5.6-luna',\n\t\t\t\treasoningEffort: 'high',\n\t\t\t\tsystemPrompt: '',\n\t\t\t\tlogger,\n\t\t\t});\n}\n\n/**\n * The workspace's history, opened on first use, so a factory that builds an agent stays\n * synchronous. Every agent and every file tool of one workspace reads the same files on disk.\n */\nfunction historyOf(workspace: WorkspaceOptions): HistoryReader {\n\t// Checked now: an agent with no root to follow is a setup error, not a first-run one.\n\tif (workspace.roots.length === 0) {\n\t\tthrow new InvalidAgentConfigError('A workspace needs at least one root');\n\t}\n\tlet store: Promise<RecoveryStore> | undefined;\n\tconst open = () => {\n\t\tstore ??= stateDirectoryOf(workspace)\n\t\t\t.then(directory => openStore(workspace, directory))\n\t\t\t.catch((error: unknown) => {\n\t\t\t\t// Not kept: the next run tries again, as one whose disk was full for a moment.\n\t\t\t\tstore = undefined;\n\t\t\t\tthrow error;\n\t\t\t});\n\t\treturn store;\n\t};\n\treturn {\n\t\tlistRuns: async () => (await open()).listRuns(),\n\t\tlistRevisions: async () => (await open()).listRevisions(),\n\t\treadRun: async runId => (await open()).readRun(runId),\n\t};\n}\n\n/**\n * What the model is told about the workspace after its instructions: the roots, which it may\n * change, and how to name a file. A host path in an error sends the model looking for it, so\n * the real root is given here once.\n */\nfunction workspaceGuidance(roots: WorkspaceRoot[]): string {\n\tconst [first, ...others] = roots.map(root => ({ ...root, path: resolve(root.path) }));\n\tif (!first) return '';\n\tconst described = [first, ...others].map(\n\t\t({ path, access }) => `- ${path}${access === 'write' ? '' : ' (read only)'}`,\n\t);\n\treturn [\n\t\t'The workspace:',\n\t\t...described,\n\t\t`Name a file by its path relative to ${first.path}, such as src/index.ts, and a file in another folder of the workspace by its full path. Read a file before you change it.`,\n\t].join('\\n');\n}\n\nfunction withGuidance(systemPrompt: string, guidance: string): string {\n\treturn guidance === '' ? systemPrompt : `${systemPrompt}\\n\\n${guidance}`;\n}\n\n/** An `LLMAgent` on `provider`, wrapped in the harness's retry policy. */\nfunction buildLLMAgent(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt,\n\t\ttools,\n\t\thistory,\n\t\tsummarizer,\n\t\tmaxSteps,\n\t\tcompactAt,\n\t\tautoApprove,\n\t\tlogger,\n\t}: {\n\t\tmodel?: AgentModel;\n\t\treasoningEffort?: ReasoningEffort;\n\t\tsystemPrompt: string;\n\t\ttools: AgentTool[];\n\t\thistory?: HistoryReader;\n\t\tsummarizer?: LLMClient;\n\t\tmaxSteps?: number;\n\t\tcompactAt?: number | undefined;\n\t\tautoApprove?: boolean;\n\t\tlogger: ILogger;\n\t},\n): Agent {\n\tconst llmClient = llmClientFor(provider, {\n\t\tmodel,\n\t\treasoningEffort,\n\t\tcompactAt,\n\t\tsystemPrompt,\n\t\tlogger,\n\t});\n\t// LLMAgent records nothing from a failed call, so the original prompt is the whole story.\n\treturn new RetryingAgent({\n\t\tinner: new LLMAgent({\n\t\t\tllmClient,\n\t\t\ttools,\n\t\t\tautoApprove,\n\t\t\tlogger,\n\t\t\t...(history ? { history } : {}),\n\t\t\t...(summarizer ? { summarizer } : {}),\n\t\t\t...(maxSteps === undefined ? {} : { maxSteps }),\n\t\t}),\n\t\tlogger,\n\t\tnoteFailures: false,\n\t});\n}\n\n/**\n * Builds an agent whose conversation MiKode owns, on a provider's model API, already wrapped in\n * the harness's retry policy. It needs the provider's API key and bills per token, where\n * `createAgent` uses the Agent SDK's login. Each tool judges the risk of each call; a\n * `destructive` one runs only when the run's `approve` allows it, or with `autoApprove`.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, a model or reasoning effort its\n * client does not support, or two tools that share a name.\n * @throws {UnrecoverableError} when the client cannot be set up, for example because\n * `OPENAI_API_KEY` is missing. The Anthropic SDK resolves credentials lazily, so a missing\n * `ANTHROPIC_API_KEY` fails the first run instead, with an `UnrecoverableError`.\n */\nexport function createLLMAgent(\n\tprovider: AgentProvider,\n\t{\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt,\n\t\ttools = [],\n\t\tworkspace,\n\t\tsummarizeUndone = true,\n\t\tmaxSteps,\n\t\tcompactAt,\n\t\tautoApprove,\n\t\tlogger = new Logger(),\n\t}: CreateLLMAgentOptions,\n): Agent {\n\treturn buildLLMAgent(provider, {\n\t\tmodel,\n\t\treasoningEffort,\n\t\tsystemPrompt: workspace\n\t\t\t? withGuidance(systemPrompt, workspaceGuidance(workspace.roots))\n\t\t\t: systemPrompt,\n\t\t// A file tool is a harness tool under its public name.\n\t\ttools: tools as AgentTool[],\n\t\t...(workspace ? { history: historyOf(workspace) } : {}),\n\t\t...(workspace && summarizeUndone ? { summarizer: summarizerFor(provider, logger) } : {}),\n\t\tmaxSteps,\n\t\tcompactAt,\n\t\tautoApprove,\n\t\tlogger,\n\t});\n}\n\n/**\n * The file tools for an agent from `createLLMAgent` on `provider`, over `workspace`, in the\n * provider's own format:\n *\n * - OpenAI: `listFiles`, `searchText`, `readFile` and its native `apply_patch`;\n * - Anthropic: `listFiles`, `searchText`, its native text editor, whose `view` reads, and\n *   `delete_file`.\n *\n * Every one goes through the workspace's access policy: a path outside the roots, a secret,\n * a file `.gitignore` excludes or the harness protects is refused. A change is made only in a\n * `write` root, only from a version the agent read, and is recorded in the workspace's history,\n * so it can be undone. Without a `write` root the tools only read. Give the agent the same\n * `workspace`, so it follows what the history undoes.\n *\n * With `commands`, they include `runCommand`, which runs one of them by name, with only the\n * parameters it declares and never through a shell. Each run asks the approver, showing the\n * program exactly as it will run, and gets a small environment without the provider keys. The\n * files of the `write` roots are captured before it and compared after, so what it changed is\n * in the history too. It is trusted execution, not a sandbox: a command can reach whatever\n * this user can. Not supported on Windows yet.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, a workspace with no root, or an\n * invalid command, including one whose `cwd` is not in a `write` root, or any on Windows.\n * @throws {UnrecoverableError} when no program can read the first root.\n */\nexport async function createFileTools(\n\tprovider: AgentProvider,\n\tworkspace: WorkspaceOptions,\n\t{ logger = new Logger(), commands }: { logger?: ILogger; commands?: Commands } = {},\n): Promise<FileTool[]> {\n\t// Reachable from untyped input.\n\tif (!isAgentProvider(provider)) throw unknownProvider(provider);\n\tif (commands !== undefined) validateCommands(commands);\n\tconst opened = await openWorkspace(workspace, logger);\n\treturn (await executorTools(provider, opened, commands)) as unknown as FileTool[];\n}\n\n/** The executor's file tools, and `runCommand` when there are commands. */\nasync function executorTools(\n\tprovider: AgentProvider,\n\tworkspace: OpenedWorkspace,\n\tcommands: Commands | undefined,\n): Promise<AgentTool[]> {\n\tconst tools = fileToolsFor(provider, workspace);\n\t// Every command may belong to other roles, leaving none for this one.\n\tif (commands === undefined || Object.keys(commands).length === 0) return tools;\n\treturn [...tools, await commandToolFor(workspace, commands)];\n}\n\n/**\n * What the planner and the reviewer read with: no edits, the current task's changes, and the\n * commands given to their role.\n */\nasync function readingTools(\n\tworkspace: OpenedWorkspace,\n\trole: CommandRole,\n\tcommands: Commands | undefined,\n): Promise<AgentTool[]> {\n\tconst { read, policy, store } = workspace;\n\tconst finding = createWorkspaceTools(read).filter(tool => tool.name !== 'readFile');\n\tconst tools = [\n\t\t...finding,\n\t\tcreateTrackedReadFile(policy),\n\t\tcreateShowChangesTool({ store, policy }),\n\t];\n\tconst own = commands && commandsFor(commands, role);\n\tif (!own || Object.keys(own).length === 0) return tools;\n\treturn [...tools, await commandToolFor(workspace, own)];\n}\n\nconst batching =\n\t'When you know you need several files, or several searches, ask for all of them in one answer rather than one at a time: each answer is a step, and steps are limited.';\n\n// How each role finds its way, sent after the workspace's roots, so it stays short. Only the\n// planner learns the repository; the executor and the reviewer work from its plan, so a plan\n// that leaves them guessing is the planner's to fix, not theirs to make up for.\nconst repositoryGuidance: Record<CommandRole, string> = {\n\tplanner: `Start with AGENTS.md, which says where the architecture and the decisions are, and read only the parts your task needs. Name in the plan every file to change and what to change in it, so the executor needs nothing else. ${batching}`,\n\texecutor: `Work from the plan: read the files it names, and only what you need to change them. Do not study the repository's architecture or decisions; the plan already applied them. ${batching}`,\n\treviewer: `Work from the plan and the changes: read what they touch, and only what you need to judge them. ${batching}`,\n};\n\n/**\n * Builds the planner → executor → reviewer workflow with every role on the model APIs, each\n * holding its role's instructions, or `systemPrompts`, as its system prompt, followed by the\n * workspace's roots.\n *\n * - The planner and the reviewer read: `listFiles`, `searchText`, `readFile`, and\n *   `showChanges`, the current task's changes as far as each may read them.\n * - The executor gets its provider's file tools (see `createFileTools`), and changes files only\n *   in a `write` root; `autoApprove` reaches only it.\n * - Each role has its own step limit: `maxSteps`, 80 for the executor and 65 for the others by\n *   default.\n * - Each role gets `runCommand` with the `commands` whose `roles` name it, the executor by\n *   default. A command asks the approver in any role, and its files are captured around it.\n * - The executor is told to finish only once the tests and checks pass, and the reviewer that\n *   they passed, when the executor has a command; without one, both are told the tests were\n *   not run, and the reviewer judges from the code whether they would pass.\n *\n * One run of the orchestrator is one run of the workspace's history, every round and retry\n * included, and its `AgentResponse` names it. Every role follows what the user undoes.\n * Asynchronous because opening the workspace is.\n *\n * @throws {InvalidAgentConfigError} for an unknown provider, or a workspace with no root.\n * @throws {UnrecoverableError} when a role cannot be set up, for example because\n * `OPENAI_API_KEY` is missing, or when no program can read the repository.\n */\nexport async function createLLMOrchestrator({\n\tprovider,\n\tautoApprove,\n\tsystemPrompts = {},\n\tworkspace = { roots: [{ path: process.cwd(), access: 'read' }] },\n\tmaxSteps,\n\tcompactAt,\n\tcommands,\n\tsummarizeUndone = true,\n\tlogger = new Logger(),\n}: CreateLLMOrchestratorOptions = {}): Promise<Agent> {\n\tconst roles = orchestratorRolesFor(provider);\n\tconst steps = perRole('maxSteps', maxSteps, defaultRoleSteps, {\n\t\tunit: 'steps',\n\t\texample: '{ \"executor\": 120 }',\n\t});\n\tconst compaction = perRole('compactAt', compactAt, defaultRoleCompaction, {\n\t\tunit: 'tokens',\n\t\texample: '{ \"executor\": 200000 }',\n\t});\n\tconst opened = await openWorkspace(workspace, logger);\n\tconst guidance = workspaceGuidance(workspace.roots);\n\tconst summarizer = summarizeUndone ? summarizerFor(roles.planner.provider, logger) : undefined;\n\tconst shared = { history: opened.store, ...(summarizer ? { summarizer } : {}), logger };\n\t// A caller's own prompt replaces our instructions and guidance, not the workspace: the model\n\t// needs the roots to name a file, as `createLLMAgent` gives them.\n\tconst prompt = (role: CommandRole, own: string | undefined, instructions: string) =>\n\t\town === undefined\n\t\t\t? withGuidance(instructions, `${guidance}\\n${repositoryGuidance[role]}`)\n\t\t\t: withGuidance(own, guidance);\n\t// Every command, checked once, before any role takes its share.\n\tif (commands !== undefined) validateCommands(commands);\n\tconst reader = async (\n\t\t{ provider, model, reasoningEffort }: Role,\n\t\trole: CommandRole,\n\t\tsystemPrompt: string,\n\t) =>\n\t\tbuildLLMAgent(provider, {\n\t\t\tmodel,\n\t\t\treasoningEffort,\n\t\t\tsystemPrompt,\n\t\t\ttools: await readingTools(opened, role, commands),\n\t\t\tmaxSteps: steps[role],\n\t\t\tcompactAt: compaction[role],\n\t\t\t...shared,\n\t\t});\n\tconst { executor } = roles;\n\tconst executorToolset = await executorTools(\n\t\texecutor.provider,\n\t\topened,\n\t\tcommands && commandsFor(commands, 'executor'),\n\t);\n\t// The executor and the reviewer are told whether the tests were run, which only an executor\n\t// with a command can do.\n\tconst executorRuns = executorToolset.some(tool => tool.name === 'runCommand');\n\n\treturn new OrchestratorAgent({\n\t\tplannerAgent: await reader(\n\t\t\troles.planner,\n\t\t\t'planner',\n\t\t\tprompt('planner', systemPrompts.planner, plannerInstructions),\n\t\t),\n\t\texecutorAgent: buildLLMAgent(executor.provider, {\n\t\t\tmodel: executor.model,\n\t\t\treasoningEffort: executor.reasoningEffort,\n\t\t\tsystemPrompt: prompt(\n\t\t\t\t'executor',\n\t\t\t\tsystemPrompts.executor,\n\t\t\t\texecutorRuns ? executorInstructions : executorInstructionsWithoutCommands,\n\t\t\t),\n\t\t\ttools: executorToolset,\n\t\t\tmaxSteps: steps.executor,\n\t\t\tcompactAt: compaction.executor,\n\t\t\tautoApprove,\n\t\t\t...shared,\n\t\t}),\n\t\treviewerAgent: await reader(\n\t\t\troles.reviewer,\n\t\t\t'reviewer',\n\t\t\tprompt(\n\t\t\t\t'reviewer',\n\t\t\t\tsystemPrompts.reviewer,\n\t\t\t\texecutorRuns ? reviewerInstructions : reviewerInstructionsWithoutCommands,\n\t\t\t),\n\t\t),\n\t\treviewerDecisionValidator: new ReviewerDecisionValidator(),\n\t\tlogger,\n\t});\n}\n"]}

package/dist/factory/infrastructure/workflowFactory.d.ts

@@ -0,0 +1,9 @@
+import type { CreateWorkflowOptions, Workflow, WorkflowStageName } from '#src/workflow/domain/workflow';
+/**
+ * Builds a workflow from the consumer's stages. The types check every `go` against the stage it
+ * names and the input it hands over. For JavaScript callers, construction checks only what it
+ * can see before a run: each stage's name, the initial stage, `maxAttempts` and `maxTransitions`.
+ * A `go` is a value a stage returns, so the run checks its target when it takes it, and nothing
+ * checks its input: the next stage's `prompt` receives it as it is.
+ */
+export declare function createWorkflow<Inputs, Result, Initial extends WorkflowStageName<Inputs>>({ initialStage, stages, maxAttempts, maxTransitions, logger, }: CreateWorkflowOptions<Inputs, Result, Initial>): Workflow<Inputs[Initial], Result>;

package/dist/factory/infrastructure/workflowFactory.js

@@ -0,0 +1,42 @@
+import { Logger } from '#src/shared/infrastructure/logger';
+import { WorkflowRunner } from '#src/workflow/domain/workflowRunner';
+function positiveInteger(value, name, fallback) {
+    const actual = value ?? fallback;
+    if (!Number.isInteger(actual) || actual < 1) {
+        throw new RangeError(`${name} must be a positive integer`);
+    }
+    return actual;
+}
+/**
+ * Builds a workflow from the consumer's stages. The types check every `go` against the stage it
+ * names and the input it hands over. For JavaScript callers, construction checks only what it
+ * can see before a run: each stage's name, the initial stage, `maxAttempts` and `maxTransitions`.
+ * A `go` is a value a stage returns, so the run checks its target when it takes it, and nothing
+ * checks its input: the next stage's `prompt` receives it as it is.
+ */
+export function createWorkflow({ initialStage, stages, maxAttempts, maxTransitions, logger = new Logger(), }) {
+    const runnable = new Map();
+    for (const [name, stage] of Object.entries(stages)) {
+        if (name.trim() === '')
+            throw new RangeError('workflow stage names must be non-empty');
+        if (stage.name !== name) {
+            throw new RangeError(`workflow stage "${stage.name}" is listed as "${name}"`);
+        }
+        if (stage.maxAttempts !== undefined)
+            positiveInteger(stage.maxAttempts, 'maxAttempts', 1);
+        runnable.set(name, stage);
+    }
+    if (runnable.size === 0)
+        throw new RangeError('stages must not be empty');
+    if (!runnable.has(initialStage)) {
+        throw new RangeError(`initial workflow stage is not among the stages: "${initialStage}"`);
+    }
+    return new WorkflowRunner({
+        initialStage,
+        stages: runnable,
+        maxAttempts: positiveInteger(maxAttempts, 'maxAttempts', 3),
+        maxTransitions: positiveInteger(maxTransitions, 'maxTransitions', 100),
+        logger,
+    });
+}
+//# sourceMappingURL=workflowFactory.js.map

package/dist/factory/infrastructure/workflowFactory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"workflowFactory.js","sourceRoot":"","sources":["../../../src/factory/infrastructure/workflowFactory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,mCAAmC,CAAC;AAC3D,OAAO,EAAE,cAAc,EAAsB,MAAM,qCAAqC,CAAC;AAOzF,SAAS,eAAe,CAAC,KAAyB,EAAE,IAAY,EAAE,QAAgB;IACjF,MAAM,MAAM,GAAG,KAAK,IAAI,QAAQ,CAAC;IACjC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,MAAM,IAAI,UAAU,CAAC,GAAG,IAAI,6BAA6B,CAAC,CAAC;IAC5D,CAAC;IACD,OAAO,MAAM,CAAC;AACf,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,cAAc,CAA4D,EACzF,YAAY,EACZ,MAAM,EACN,WAAW,EACX,cAAc,EACd,MAAM,GAAG,IAAI,MAAM,EAAE,GAC2B;IAChD,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAyB,CAAC;IAClD,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAgB,MAAM,CAAC,EAAE,CAAC;QACnE,IAAI,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE;YAAE,MAAM,IAAI,UAAU,CAAC,wCAAwC,CAAC,CAAC;QACvF,IAAI,KAAK,CAAC,IAAI,KAAK,IAAI,EAAE,CAAC;YACzB,MAAM,IAAI,UAAU,CAAC,mBAAmB,KAAK,CAAC,IAAI,mBAAmB,IAAI,GAAG,CAAC,CAAC;QAC/E,CAAC;QACD,IAAI,KAAK,CAAC,WAAW,KAAK,SAAS;YAAE,eAAe,CAAC,KAAK,CAAC,WAAW,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC;QAC1F,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC3B,CAAC;IACD,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC;QAAE,MAAM,IAAI,UAAU,CAAC,0BAA0B,CAAC,CAAC;IAC1E,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,YAAY,CAAC,EAAE,CAAC;QACjC,MAAM,IAAI,UAAU,CAAC,oDAAoD,YAAY,GAAG,CAAC,CAAC;IAC3F,CAAC;IAED,OAAO,IAAI,cAAc,CAA0B;QAClD,YAAY;QACZ,MAAM,EAAE,QAAQ;QAChB,WAAW,EAAE,eAAe,CAAC,WAAW,EAAE,aAAa,EAAE,CAAC,CAAC;QAC3D,cAAc,EAAE,eAAe,CAAC,cAAc,EAAE,gBAAgB,EAAE,GAAG,CAAC;QACtE,MAAM;KACN,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { Logger } from '#src/shared/infrastructure/logger';\nimport { WorkflowRunner, type RunnableStage } from '#src/workflow/domain/workflowRunner';\nimport type {\n\tCreateWorkflowOptions,\n\tWorkflow,\n\tWorkflowStageName,\n} from '#src/workflow/domain/workflow';\n\nfunction positiveInteger(value: number | undefined, name: string, fallback: number): number {\n\tconst actual = value ?? fallback;\n\tif (!Number.isInteger(actual) || actual < 1) {\n\t\tthrow new RangeError(`${name} must be a positive integer`);\n\t}\n\treturn actual;\n}\n\n/**\n * Builds a workflow from the consumer's stages. The types check every `go` against the stage it\n * names and the input it hands over. For JavaScript callers, construction checks only what it\n * can see before a run: each stage's name, the initial stage, `maxAttempts` and `maxTransitions`.\n * A `go` is a value a stage returns, so the run checks its target when it takes it, and nothing\n * checks its input: the next stage's `prompt` receives it as it is.\n */\nexport function createWorkflow<Inputs, Result, Initial extends WorkflowStageName<Inputs>>({\n\tinitialStage,\n\tstages,\n\tmaxAttempts,\n\tmaxTransitions,\n\tlogger = new Logger(),\n}: CreateWorkflowOptions<Inputs, Result, Initial>): Workflow<Inputs[Initial], Result> {\n\tconst runnable = new Map<string, RunnableStage>();\n\tfor (const [name, stage] of Object.entries<RunnableStage>(stages)) {\n\t\tif (name.trim() === '') throw new RangeError('workflow stage names must be non-empty');\n\t\tif (stage.name !== name) {\n\t\t\tthrow new RangeError(`workflow stage \"${stage.name}\" is listed as \"${name}\"`);\n\t\t}\n\t\tif (stage.maxAttempts !== undefined) positiveInteger(stage.maxAttempts, 'maxAttempts', 1);\n\t\trunnable.set(name, stage);\n\t}\n\tif (runnable.size === 0) throw new RangeError('stages must not be empty');\n\tif (!runnable.has(initialStage)) {\n\t\tthrow new RangeError(`initial workflow stage is not among the stages: \"${initialStage}\"`);\n\t}\n\n\treturn new WorkflowRunner<Inputs[Initial], Result>({\n\t\tinitialStage,\n\t\tstages: runnable,\n\t\tmaxAttempts: positiveInteger(maxAttempts, 'maxAttempts', 3),\n\t\tmaxTransitions: positiveInteger(maxTransitions, 'maxTransitions', 100),\n\t\tlogger,\n\t});\n}\n"]}

package/dist/index.d.ts

@@ -1,6 +1,6 @@
 export type { ILogger } from './shared/domain/logger.ts';
 export { createAgent, createOrchestrator, agentModels, agentReasoningEfforts, isAgentProvider, isModelOf, isReasoningEffortOf, type CreateAgentOptions, } from './factory/infrastructure/agentFactory.ts';
-export { createFileTools, createLLMAgent, createLLMOrchestrator, isLLMModelOf, isLLMReasoningEffortOf, llmModels, llmReasoningEfforts, type CreateLLMAgentOptions, type CreateLLMOrchestratorOptions, type RoleSteps, } from './factory/infrastructure/agentLLMFactory.ts';
+export { createFileTools, createLLMAgent, createLLMOrchestrator, isLLMModelOf, isLLMReasoningEffortOf, llmModels, llmReasoningEfforts, type CreateLLMAgentOptions, type CreateLLMOrchestratorOptions, type RoleSteps, type RoleTokens, } from './factory/infrastructure/agentLLMFactory.ts';
 export type { FileTool, WorkspaceOptions } from './tools/infrastructure/fileTools.ts';
 export type { WorkspaceRoot } from './tools/domain/accessPolicy.ts';
 export type { CommandArgument, CommandDefinition, CommandParameter, CommandRole, Commands, } from './tools/domain/commands.ts';
@@ -22,3 +22,6 @@ export { createHistory } from './recovery/infrastructure/createHistory.ts';
 export type { History, HistoryRun, Move } from './recovery/domain/history.ts';
 export { historyStart, NothingToMoveError, WorkspaceBusyError, WorkspaceMovedError, type MoveOptions, type RunStatus, } from './recovery/domain/recoveryStore.ts';
 export { HistoryExpiredError, UnknownRunError } from './recovery/domain/runTree.ts';
+export { createWorkflow } from './factory/infrastructure/workflowFactory.ts';
+export { defineWorkflowStages, type CreateWorkflowOptions, type Workflow, type WorkflowResult, type WorkflowStage, type WorkflowStageDefinition, type WorkflowStageName, type WorkflowStageReport, type WorkflowTransition, } from './workflow/domain/workflow.ts';
+export type { Validator } from './shared/domain/validator.ts';

package/dist/index.js

@@ -11,4 +11,6 @@ export { createWorkspaceTools } from "./tools/infrastructure/workspaceTools.js";
 export { createHistory } from "./recovery/infrastructure/createHistory.js";
 export { historyStart, NothingToMoveError, WorkspaceBusyError, WorkspaceMovedError, } from "./recovery/domain/recoveryStore.js";
 export { HistoryExpiredError, UnknownRunError } from "./recovery/domain/runTree.js";
+export { createWorkflow } from "./factory/infrastructure/workflowFactory.js";
+export { defineWorkflowStages, } from "./workflow/domain/workflow.js";
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,OAAO,EACN,WAAW,EACX,kBAAkB,EAClB,WAAW,EACX,qBAAqB,EACrB,eAAe,EACf,SAAS,EACT,mBAAmB,GAEnB,MAAM,0CAA0C,CAAC;AAClD,OAAO,EACN,eAAe,EACf,cAAc,EACd,qBAAqB,EACrB,YAAY,EACZ,sBAAsB,EACtB,SAAS,EACT,mBAAmB,GAInB,MAAM,6CAA6C,CAAC;AAWrD,OAAO,EACN,cAAc,GAKd,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAQN,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACN,uBAAuB,EACvB,gBAAgB,EAChB,kBAAkB,GAClB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAC;AAE/D,OAAO,EACN,iBAAiB,GAOjB,MAAM,4BAA4B,CAAC;AAKpC,OAAO,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAClE,OAAO,EAAE,eAAe,EAAE,MAAM,2CAA2C,CAAC;AAC5E,OAAO,EAAE,oBAAoB,EAAE,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAE,aAAa,EAAE,MAAM,4CAA4C,CAAC;AAE3E,OAAO,EACN,YAAY,EACZ,kBAAkB,EAClB,kBAAkB,EAClB,mBAAmB,GAGnB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,mBAAmB,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC","sourcesContent":["export type { ILogger } from './shared/domain/logger.ts';\nexport {\n\tcreateAgent,\n\tcreateOrchestrator,\n\tagentModels,\n\tagentReasoningEfforts,\n\tisAgentProvider,\n\tisModelOf,\n\tisReasoningEffortOf,\n\ttype CreateAgentOptions,\n} from './factory/infrastructure/agentFactory.ts';\nexport {\n\tcreateFileTools,\n\tcreateLLMAgent,\n\tcreateLLMOrchestrator,\n\tisLLMModelOf,\n\tisLLMReasoningEffortOf,\n\tllmModels,\n\tllmReasoningEfforts,\n\ttype CreateLLMAgentOptions,\n\ttype CreateLLMOrchestratorOptions,\n\ttype RoleSteps,\n} from './factory/infrastructure/agentLLMFactory.ts';\nexport type { FileTool, WorkspaceOptions } from './tools/infrastructure/fileTools.ts';\nexport type { WorkspaceRoot } from './tools/domain/accessPolicy.ts';\nexport type {\n\tCommandArgument,\n\tCommandDefinition,\n\tCommandParameter,\n\tCommandRole,\n\tCommands,\n} from './tools/domain/commands.ts';\nexport type { SecretRules } from './tools/infrastructure/rootsAccessPolicy.ts';\nexport {\n\tagentProviders,\n\ttype AgentModel,\n\ttype AgentProvider,\n\ttype CreateOrchestratorOptions,\n\ttype ReasoningEffort,\n} from './factory/infrastructure/types.ts';\nexport {\n\ttype Agent,\n\ttype AgentResponse,\n\ttype Callback,\n\ttype ProgressAgent,\n\ttype ProgressEvent,\n\ttype StepContext,\n\ttype RunOptions,\n} from './agent/domain/agent.ts';\nexport {\n\tInvalidAgentConfigError,\n\tRecoverableError,\n\tUnrecoverableError,\n} from './shared/domain/errors.ts';\nexport { isAbortError } from './shared/domain/isAbortError.ts';\nexport type { Tokens } from './shared/domain/tokens.ts';\nexport {\n\trememberApprovals,\n\ttype ApprovalDecision,\n\ttype ApprovalRequest,\n\ttype Approver,\n\ttype CommandLine,\n\ttype RememberableDecision,\n\ttype ToolRisk,\n} from './agent/domain/approval.ts';\nexport type { Asker, QuestionOption, UserQuestion } from './agent/domain/question.ts';\nexport type { JSONSchema, ToolDefinition } from './llm/domain/tool.ts';\nexport type { Tool, ToolSummarizer } from './tools/domain/tool.ts';\nexport type { TextMatch, Workspace } from './tools/domain/workspace.ts';\nexport { defineTool } from './tools/infrastructure/defineTool.ts';\nexport { createWorkspace } from './tools/infrastructure/createWorkspace.ts';\nexport { createWorkspaceTools } from './tools/infrastructure/workspaceTools.ts';\nexport { createHistory } from './recovery/infrastructure/createHistory.ts';\nexport type { History, HistoryRun, Move } from './recovery/domain/history.ts';\nexport {\n\thistoryStart,\n\tNothingToMoveError,\n\tWorkspaceBusyError,\n\tWorkspaceMovedError,\n\ttype MoveOptions,\n\ttype RunStatus,\n} from './recovery/domain/recoveryStore.ts';\nexport { HistoryExpiredError, UnknownRunError } from './recovery/domain/runTree.ts';\n"]}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,OAAO,EACN,WAAW,EACX,kBAAkB,EAClB,WAAW,EACX,qBAAqB,EACrB,eAAe,EACf,SAAS,EACT,mBAAmB,GAEnB,MAAM,0CAA0C,CAAC;AAClD,OAAO,EACN,eAAe,EACf,cAAc,EACd,qBAAqB,EACrB,YAAY,EACZ,sBAAsB,EACtB,SAAS,EACT,mBAAmB,GAKnB,MAAM,6CAA6C,CAAC;AAWrD,OAAO,EACN,cAAc,GAKd,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAQN,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACN,uBAAuB,EACvB,gBAAgB,EAChB,kBAAkB,GAClB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAC;AAE/D,OAAO,EACN,iBAAiB,GAOjB,MAAM,4BAA4B,CAAC;AAKpC,OAAO,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAClE,OAAO,EAAE,eAAe,EAAE,MAAM,2CAA2C,CAAC;AAC5E,OAAO,EAAE,oBAAoB,EAAE,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAE,aAAa,EAAE,MAAM,4CAA4C,CAAC;AAE3E,OAAO,EACN,YAAY,EACZ,kBAAkB,EAClB,kBAAkB,EAClB,mBAAmB,GAGnB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,mBAAmB,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AACpF,OAAO,EAAE,cAAc,EAAE,MAAM,6CAA6C,CAAC;AAC7E,OAAO,EACN,oBAAoB,GASpB,MAAM,+BAA+B,CAAC","sourcesContent":["export type { ILogger } from './shared/domain/logger.ts';\nexport {\n\tcreateAgent,\n\tcreateOrchestrator,\n\tagentModels,\n\tagentReasoningEfforts,\n\tisAgentProvider,\n\tisModelOf,\n\tisReasoningEffortOf,\n\ttype CreateAgentOptions,\n} from './factory/infrastructure/agentFactory.ts';\nexport {\n\tcreateFileTools,\n\tcreateLLMAgent,\n\tcreateLLMOrchestrator,\n\tisLLMModelOf,\n\tisLLMReasoningEffortOf,\n\tllmModels,\n\tllmReasoningEfforts,\n\ttype CreateLLMAgentOptions,\n\ttype CreateLLMOrchestratorOptions,\n\ttype RoleSteps,\n\ttype RoleTokens,\n} from './factory/infrastructure/agentLLMFactory.ts';\nexport type { FileTool, WorkspaceOptions } from './tools/infrastructure/fileTools.ts';\nexport type { WorkspaceRoot } from './tools/domain/accessPolicy.ts';\nexport type {\n\tCommandArgument,\n\tCommandDefinition,\n\tCommandParameter,\n\tCommandRole,\n\tCommands,\n} from './tools/domain/commands.ts';\nexport type { SecretRules } from './tools/infrastructure/rootsAccessPolicy.ts';\nexport {\n\tagentProviders,\n\ttype AgentModel,\n\ttype AgentProvider,\n\ttype CreateOrchestratorOptions,\n\ttype ReasoningEffort,\n} from './factory/infrastructure/types.ts';\nexport {\n\ttype Agent,\n\ttype AgentResponse,\n\ttype Callback,\n\ttype ProgressAgent,\n\ttype ProgressEvent,\n\ttype StepContext,\n\ttype RunOptions,\n} from './agent/domain/agent.ts';\nexport {\n\tInvalidAgentConfigError,\n\tRecoverableError,\n\tUnrecoverableError,\n} from './shared/domain/errors.ts';\nexport { isAbortError } from './shared/domain/isAbortError.ts';\nexport type { Tokens } from './shared/domain/tokens.ts';\nexport {\n\trememberApprovals,\n\ttype ApprovalDecision,\n\ttype ApprovalRequest,\n\ttype Approver,\n\ttype CommandLine,\n\ttype RememberableDecision,\n\ttype ToolRisk,\n} from './agent/domain/approval.ts';\nexport type { Asker, QuestionOption, UserQuestion } from './agent/domain/question.ts';\nexport type { JSONSchema, ToolDefinition } from './llm/domain/tool.ts';\nexport type { Tool, ToolSummarizer } from './tools/domain/tool.ts';\nexport type { TextMatch, Workspace } from './tools/domain/workspace.ts';\nexport { defineTool } from './tools/infrastructure/defineTool.ts';\nexport { createWorkspace } from './tools/infrastructure/createWorkspace.ts';\nexport { createWorkspaceTools } from './tools/infrastructure/workspaceTools.ts';\nexport { createHistory } from './recovery/infrastructure/createHistory.ts';\nexport type { History, HistoryRun, Move } from './recovery/domain/history.ts';\nexport {\n\thistoryStart,\n\tNothingToMoveError,\n\tWorkspaceBusyError,\n\tWorkspaceMovedError,\n\ttype MoveOptions,\n\ttype RunStatus,\n} from './recovery/domain/recoveryStore.ts';\nexport { HistoryExpiredError, UnknownRunError } from './recovery/domain/runTree.ts';\nexport { createWorkflow } from './factory/infrastructure/workflowFactory.ts';\nexport {\n\tdefineWorkflowStages,\n\ttype CreateWorkflowOptions,\n\ttype Workflow,\n\ttype WorkflowResult,\n\ttype WorkflowStage,\n\ttype WorkflowStageDefinition,\n\ttype WorkflowStageName,\n\ttype WorkflowStageReport,\n\ttype WorkflowTransition,\n} from './workflow/domain/workflow.ts';\nexport type { Validator } from './shared/domain/validator.ts';\n"]}

package/dist/llm/domain/llm.d.ts

@@ -18,8 +18,8 @@ export interface LLMClient {
     /** The system prompt every call sends, for measuring what a call carries. */
     readonly systemPrompt?: string;
     /**
-     * The `source` of the `providerData` this client sends back; it drops every other. For
-     * measuring what a call carries.
+     * The `source` of the `providerData` and `compaction` parts this client sends back; it drops
+     * every other. For measuring what a call carries.
      */
     readonly providerDataSource?: string;
     /** Whether this client can declare a native tool. Missing means it declares none. */

package/dist/llm/domain/llm.js.map

@@ -1 +1 @@
-{"version":3,"file":"llm.js","sourceRoot":"","sources":["../../../src/llm/domain/llm.ts"],"names":[],"mappings":"","sourcesContent":["import type { Tokens } from '#src/shared/domain/tokens';\nimport type { Message } from './message.ts';\nimport type { NativeTool, ToolDefinition } from './tool.ts';\n\nexport type StopReason = 'completed' | 'truncated' | 'refused';\n\nexport interface LLMResponse {\n\tmessage: Message & { role: 'assistant' };\n\t/**\n\t * Missing when the provider did not report it. The call was made and may have been billed,\n\t * so the agent keeps the answer and only its tokens become unknown; a failure raised from\n\t * such a response is marked `usageUnreported`.\n\t */\n\tusage: Tokens | undefined;\n\tstopReason: StopReason;\n}\n\nexport interface LLMClient {\n\t/** The system prompt every call sends, for measuring what a call carries. */\n\treadonly systemPrompt?: string;\n\t/**\n\t * The `source` of the `providerData` this client sends back; it drops every other. For\n\t * measuring what a call carries.\n\t */\n\treadonly providerDataSource?: string;\n\t/** Whether this client can declare a native tool. Missing means it declares none. */\n\tsupportsNative?(tool: NativeTool): boolean;\n\tsend(\n\t\t{ context, tools }: { context: Message[]; tools: ToolDefinition[] },\n\t\tsignal: AbortSignal,\n\t): Promise<LLMResponse>;\n}\n"]}
+{"version":3,"file":"llm.js","sourceRoot":"","sources":["../../../src/llm/domain/llm.ts"],"names":[],"mappings":"","sourcesContent":["import type { Tokens } from '#src/shared/domain/tokens';\nimport type { Message } from './message.ts';\nimport type { NativeTool, ToolDefinition } from './tool.ts';\n\nexport type StopReason = 'completed' | 'truncated' | 'refused';\n\nexport interface LLMResponse {\n\tmessage: Message & { role: 'assistant' };\n\t/**\n\t * Missing when the provider did not report it. The call was made and may have been billed,\n\t * so the agent keeps the answer and only its tokens become unknown; a failure raised from\n\t * such a response is marked `usageUnreported`.\n\t */\n\tusage: Tokens | undefined;\n\tstopReason: StopReason;\n}\n\nexport interface LLMClient {\n\t/** The system prompt every call sends, for measuring what a call carries. */\n\treadonly systemPrompt?: string;\n\t/**\n\t * The `source` of the `providerData` and `compaction` parts this client sends back; it drops\n\t * every other. For measuring what a call carries.\n\t */\n\treadonly providerDataSource?: string;\n\t/** Whether this client can declare a native tool. Missing means it declares none. */\n\tsupportsNative?(tool: NativeTool): boolean;\n\tsend(\n\t\t{ context, tools }: { context: Message[]; tools: ToolDefinition[] },\n\t\tsignal: AbortSignal,\n\t): Promise<LLMResponse>;\n}\n"]}

package/dist/llm/domain/message.d.ts

@@ -1,4 +1,4 @@
-export type MessagePart = TextPart | ReasoningPart | ToolCallPart | ToolResultPart | ProviderDataPart;
+export type MessagePart = TextPart | ReasoningPart | ToolCallPart | ToolResultPart | ProviderDataPart | CompactionPart;
 export interface TextPart {
     type: 'text';
     text: string;
@@ -31,7 +31,23 @@ export interface ProviderDataPart {
     source: string;
     data: unknown;
 }
+/**
+ * A provider's own summary of everything before it, written when its context passed the
+ * client's `compactAt`. Like `ProviderDataPart`, only the client labelled `source` reads it
+ * back: that client sends the conversation from its latest compaction on, and every other client
+ * leaves the part out and sends the conversation whole, which the conversation still holds.
+ */
+export interface CompactionPart {
+    type: 'compaction';
+    source: string;
+    data: unknown;
+}
 export interface Message {
     role: 'user' | 'assistant' | 'tool';
     content: MessagePart[];
 }
+/**
+ * What the client labelled `source` sends of `context`: everything from its latest compaction
+ * on, that compaction included, or the whole context when it has none.
+ */
+export declare function sentContext(context: Message[], source: string | undefined): Message[];