@buildaharness/aielia 0.3.9 → 0.3.11

package/README.md

@@ -275,6 +275,12 @@ runs:
 | `@buildaharness/chat-ui` | Browser | `IndexedDBAdapter`/`DexieExperienceStore` via `PersonalAssistant.create()` (best-effort — see `packages/runtime/README.md`'s persistence section) |
 | `@buildaharness/desktop` | Native window (Tauri, wraps chat-ui) | Same `FileSystemAdapter`/`FileSystemExperienceStore` classes as the CLI, but over `@tauri-apps/plugin-fs` instead of `node:fs`, under `appLocalDataDir()` |
 
+A second browser page, **Aielia Write** (`/write`), uses the same package for writing an
+article with Aielia beside it: it tracks the article's claims and their sources, verifies
+claims against web pages, code files and repositories, and only changes the text as
+suggestions you accept. Its logic lives in `src/writer/` (exported from this package) and the
+page in `@buildaharness/chat-ui`; see `packages/chat-ui/README.md`'s "Aielia Write" section.
+
 Both filesystem backends are the *same* `FileSystemAdapter`/`FileSystemExperienceStore`
 classes — see `packages/runtime/README.md`'s "Filesystem-backed storage"
 section for how the file-I/O seam that makes that possible works.
@@ -927,6 +933,8 @@ ends the session.
 | Command | What it does |
 |---|---|
 | `/help` | Show this list |
+| `/exit` (aliases `/quit`, `exit`, `quit`) | End the session |
+| `/config` | View or change persisted settings (`/config set <key> <value>`; see "Configuration" above) |
 | `/clear` (alias `/new`) | Start a fresh conversation — deletes this session's transcript, extracted facts, and active plan. Leaves learned reminders/experience untouched (those are durable, cross-conversation learning, not conversation-scoped state) |
 | `/status` | Show the resolved config (model, backend, workspace, enabled capabilities — same as the startup banner) plus this session's transcript length and whether a plan is active |
 | `/export [file]` | Save this session's transcript to a markdown file (default: `assistant-transcript-<timestamp>.md` in the current directory) |

package/dist/_memory-core-generated.d.ts

@@ -1762,3 +1762,9 @@ export declare const INV16: {
     readonly neverPromotable: readonly ["agent", "tool", "web"];
     readonly neverReturnedTiers: readonly ["procedural", "commitment"];
 };
+export declare const MEMORY_EVENT_STORE_KEY: "memory:events";
+export declare const MEMORY_EVENT_KEEP: 500;
+export declare const MEMORY_EVENT_KINDS: readonly ["promote", "decline", "correct", "contradict", "supersede", "retire", "forget", "expire", "restore", "recall"];
+export declare const MEMORY_REASON_CODES: readonly ["user_stated", "user_corrected", "user_confirmed", "user_rejected", "inferred_high_conf", "inferred_pending_confirmation", "inferred_low_conf_session_only", "conflicts_with_fact", "superseded_by_newer", "explicit_forget", "retention_expired", "gate_blocked_secret", "gate_blocked_instruction", "duplicate"];
+export declare const MEMORY_SOURCE_TYPES: readonly ["user_message", "assistant_message", "tool_result", "prior_fact", "user_command"];
+export declare const MEMORY_DECIDED_BY: readonly ["user", "classifier", "reviewer", "consolidation", "rule"];

package/dist/assistant.d.ts

@@ -6,9 +6,11 @@ import { WebToolsContext } from './web-tools.js';
 import { ShellToolsContext } from './shell-tools.js';
 import { ActionToolsContext } from './action-tools.js';
 import { SpendCapConfig, SpendState } from './spend-cap.js';
+import { WriterChatPort } from './writer/writer-chat-turn.js';
 import { TraceEvent } from './trace-events.js';
 import { AssistantToolStep } from './tool-step.js';
 import { MemoryService, MemorySummary, MemoryExport, AuditEntry, MemoryStatus, MemoryInjection, ConsolidationOutcome } from './memory-service.js';
+import { EventFilter, MemoryEvent } from './memory-events.js';
 import { MemoryWriteMode } from './memory-governance.js';
 import { ConsolidationProposal } from './memory-consolidation.js';
 import { UserFact } from './fact-extraction.js';
@@ -233,6 +235,12 @@ export interface PersonalAssistantOptions {
      * Off by default.
      */
     dangerouslySkipPermissions?: boolean;
+    /**
+     * Aielia Write (T13): when set, each writer message is read for reviewer comments, decisions and corrections to sources, the result is
+     * stored per article through this port (not per session), and the open items of the active article are added to the system prompt
+     * on every turn. Unset (every other caller) means none of this runs. See plans/aielia_write_article_model_plan.html, section 8.
+     */
+    writerChat?: WriterChatPort;
     /**
      * Opt-in session spend/turn-count ceilings — see spend-cap.ts. Undefined by default, same as
      * every other field here that changes behavior only when a caller sets it: no ceiling means
@@ -323,6 +331,7 @@ export declare class PersonalAssistant {
     private readonly onTrace?;
     private readonly onDebugLog?;
     private readonly dangerouslySkipPermissions;
+    private readonly writerChat;
     /** Whenever any of fileTools/webTools/shellTools is configured, `turn()` routes through AgentLoop instead of a single plain chat call — computed once here since it never changes for the lifetime of an instance. */
     private readonly toolLoopWillRun;
     /** R3 of the internal plan — mirrors the same flag HarnessBridge was given at construction, kept here too so runTurn can decide whether to defer the tool loop into a harness-driven proposer instead of precomputing draftReply. See PersonalAssistantOptions.oneLoopMode's doc comment. */
@@ -555,6 +564,8 @@ export declare class PersonalAssistant {
     forgetFact(selector: string, sessionId: string, erase?: boolean): Promise<MemoryPendingOutcome>;
     /** `/memory history` — the newest audit-log entries (M2). Empty unless `AUDIT_MEMORY_AUDIT_LOG` is on. */
     memoryHistory(limit?: number): Promise<AuditEntry[]>;
+    /** `/memory history events [fact]` — transition events with reason and sources (`AUDIT_MEMORY_EVENTS`, default on). */
+    memoryEvents(limit?: number, filter?: EventFilter): Promise<MemoryEvent[]>;
     /** `/memory undo <seq>` — restores the pre-image of one audit entry (M2). */
     undoMemoryChange(selector: string, sessionId: string): Promise<{
         ok: boolean;

package/dist/{bin-CCwDFxDe.js → bin-BRwA8Kfr.js}

@@ -8,7 +8,7 @@ import { rename, stat, realpath, readdir, mkdir, unlink, writeFile, readFile } f
 import { homedir, tmpdir } from "node:os";
 import { fileURLToPath, pathToFileURL } from "node:url";
 import { OPENROUTER_DEFAULT_MODEL, OPENAI_DEFAULT_MODEL, ANTHROPIC_DEFAULT_MODEL, InMemoryReminderStore, FileSystemAdapter, FileSystemExperienceStore } from "@buildaharness/runtime";
-import { af as buildClaudePrompt, co as stripJsonCodeFence, c2 as recallToolEnabled, bw as lexicalOffEnvValue, a as ALREADY_STAGED_ACTION_TOOL, cn as stagedActionInput, bT as parseClaudeCliOutput, cp as stripMcpToolPrefix, bE as memoryStatusChecks, bR as normalizePlanNodes, V as PROVIDER_SETUP, ao as cleanApiKey, ah as checkApiKeyFormat, aG as envOverridesFromProcessEnv, c6 as resolveConfig, cl as shouldLaunchTuiApp, t as DEFAULT_TUI_MODE, a8 as applyLayerSettings, cj as sanitizeLayerChoices, cu as testApiKey, cx as validateConfig, aW as findInvalidNumericSettings, f as ConfigValidationError, be as formatSpendCapStatus, br as isPlanGraphEnabled, X as PersonalAssistant, ae as buildAssistantOptions, cq as stripTerminalControls, bn as isGoalGraphEnabled, b$ as planToSnapshot, ar as createLlmClient, au as createSmtpSender, at as createResendSender, aZ as formatConfigListing, bm as isConfigKey, e as CONFIG_KEYS, x as ENV_VAR_FOR_CONFIG_KEY, bU as parseConfigValue, g as ConfigValueParseError, ai as checkProxyHealth, a$ as formatDoctorReport, a_ as formatCostSummary, bW as parseProjectCommand, b1 as formatGoalGraphState, bd as formatSearchResults, b6 as formatMemoryHistory, b9 as formatMemoryStatus, b8 as formatMemoryPendingOutcome, b4 as formatMemoryArchive, bY as parseUndoActionCommand, bh as formatUndoLogListing, c0 as planUndo, ax as defaultExportFilename, bg as formatTranscriptMarkdown, aj as classifyError, bf as formatStatus, b2 as formatHelp, aS as explainSteps, aR as explainSources, aP as explainLayers, b3 as formatLayerListing, cA as withLayerChoice, O as LayerSettingError, aT as explainWhy, bt as isQuitCommand, P as LiveSteeringChannel, ay as defaultMemoryExportFilename, b5 as formatMemoryExport, bD as memoryAuditLogEnabled, ba as formatMemorySummary, aa as approvalLabels, U as NO_TRACE, R as MISSING_PLAN_APPROVAL_ID_REASON, c1 as presentTurnResult, bb as formatNextSteps, aQ as explainPlan, bH as nodeToLayer, L as LAYER_DISPLAY_NAME, bG as nodeDisplayName, j as DEFAULT_CLARIFICATION_REASON, aE as emptyDraft, aC as draftToAnswer, bS as orderedOptions, cz as withCostEstimate, C as CLI_FEATURES, ad as braveSearch } from "./provider-setup-Cy9dwqrd.js";
+import { ap as buildClaudePrompt, cU as stripJsonCodeFence, ct as recallToolEnabled, bT as lexicalOffEnvValue, a as ALREADY_STAGED_ACTION_TOOL, cR as stagedActionInput, az as claudeFailureMessage, ch as parseClaudeCliOutput, cV as stripMcpToolPrefix, c0 as memoryStatusChecks, cd as normalizePlanNodes, X as PROVIDER_SETUP, aA as cleanApiKey, as as checkApiKeyFormat, aV as envOverridesFromProcessEnv, cA as resolveConfig, cP as shouldLaunchTuiApp, t as DEFAULT_TUI_MODE, ag as applyLayerSettings, cN as sanitizeLayerChoices, c_ as testApiKey, d2 as validateConfig, bb as findInvalidNumericSettings, f as ConfigValidationError, by as formatSpendCapStatus, bN as isPlanGraphEnabled, Z as PersonalAssistant, ao as buildAssistantOptions, cW as stripTerminalControls, bJ as isGoalGraphEnabled, cp as planToSnapshot, aE as createLlmClient, aI as createSmtpSender, aH as createResendSender, bg as formatConfigListing, bI as isConfigKey, e as CONFIG_KEYS, x as ENV_VAR_FOR_CONFIG_KEY, ci as parseConfigValue, g as ConfigValueParseError, at as checkProxyHealth, bi as formatDoctorReport, bh as formatCostSummary, ck as parseProjectCommand, bk as formatGoalGraphState, bx as formatSearchResults, bo as formatMemoryEvents, bq as formatMemoryHistory, bt as formatMemoryStatus, bs as formatMemoryPendingOutcome, bn as formatMemoryArchive, cm as parseUndoActionCommand, bB as formatUndoLogListing, cq as planUndo, aL as defaultExportFilename, bA as formatTranscriptMarkdown, au as classifyError, bz as formatStatus, bl as formatHelp, b5 as explainSteps, b4 as explainSources, b2 as explainLayers, bm as formatLayerListing, d5 as withLayerChoice, P as LayerSettingError, bR as layerWarnings, b6 as explainWhy, bP as isQuitCommand, Q as LiveSteeringChannel, aM as defaultMemoryExportFilename, bp as formatMemoryExport, b$ as memoryAuditLogEnabled, bu as formatMemorySummary, ai as approvalLabels, W as NO_TRACE, S as MISSING_PLAN_APPROVAL_ID_REASON, cr as presentTurnResult, bv as formatNextSteps, b3 as explainPlan, c3 as nodeToLayer, L as LAYER_DISPLAY_NAME, c2 as nodeDisplayName, j as DEFAULT_CLARIFICATION_REASON, aS as emptyDraft, aQ as draftToAnswer, cg as orderedOptions, d4 as withCostEstimate, C as CLI_FEATURES, an as braveSearch } from "./provider-setup-DG8iYm_F.js";
 import { spawn } from "node:child_process";
 import { createServer, connect } from "node:net";
 import { createRequire } from "node:module";
@@ -129,7 +129,7 @@ function createNodeFsBackend() {
     }
   };
 }
-const CLI_VERSION = "0.3.9";
+const CLI_VERSION = "0.3.11";
 function seaCacheDir(version, home = homedir()) {
   return join(home, ".buildaharness", "personal-assistant", "sea-cache", version);
 }
@@ -166,9 +166,15 @@ function resolveMcpServerPath(seaOverride, home) {
   return fileURLToPath(new URL(MCP_SERVER_FILE_NAME, import.meta.url));
 }
 const EMPTY_MCP_CONFIG = JSON.stringify({ mcpServers: {} });
-function invokeClaude(claudePath, args) {
+function killOnAbort(proc, signal) {
+  if (!signal) return;
+  if (signal.aborted) proc.kill();
+  else signal.addEventListener("abort", () => proc.kill(), { once: true });
+}
+function invokeClaude(claudePath, args, signal) {
   return new Promise((resolvePromise, reject) => {
     const proc = spawn(claudePath, args, { cwd: tmpdir(), stdio: ["ignore", "pipe", "pipe"] });
+    killOnAbort(proc, signal);
     let stdout = "";
     let stderr = "";
     proc.stdout.on("data", (chunk) => {
@@ -180,16 +186,17 @@ function invokeClaude(claudePath, args) {
     proc.on("error", reject);
     proc.on("close", (code) => {
       if (code !== 0) {
-        reject(new Error(stderr.trim() || `claude exited with code ${code}`));
+        reject(new Error(claudeFailureMessage(stdout, stderr, code)));
         return;
       }
       resolvePromise(parseClaudeCliOutput(stdout));
     });
   });
 }
-function invokeClaudeStreaming(claudePath, args, onToolStep) {
+function invokeClaudeStreaming(claudePath, args, onToolStep, signal) {
   return new Promise((resolvePromise, reject) => {
     const proc = spawn(claudePath, args, { cwd: tmpdir(), stdio: ["ignore", "pipe", "pipe"] });
+    killOnAbort(proc, signal);
     let buffer = "";
     let stderr = "";
     let finalResultLine;
@@ -229,7 +236,7 @@ function invokeClaudeStreaming(claudePath, args, onToolStep) {
     proc.on("error", reject);
     proc.on("close", (code) => {
       if (code !== 0) {
-        reject(new Error(stderr.trim() || `claude exited with code ${code}`));
+        reject(new Error(claudeFailureMessage(finalResultLine ?? "", stderr, code)));
         return;
       }
       handleLine(buffer);
@@ -312,6 +319,8 @@ function startToolGateServer(onToolProposal, onToolResult, onToolExecute) {
 }
 class ClaudeCliLLMClient {
   constructor(options = {}) {
+    /** Runs its own tool loop inside `claude -p`; it cannot return tool calls to a caller's loop, so those callers use text tool calls (see writer/text-tool-calls.ts). */
+    __publicField(this, "nativeToolCalls", false);
     __publicField(this, "claudePath");
     __publicField(this, "fileTools");
     __publicField(this, "remindersFile");
@@ -356,7 +365,7 @@ class ClaudeCliLLMClient {
     const model = options.model ?? this.model;
     if (model) args.push("--model", model);
     args.push("--", prompt);
-    const { reply, usage, model: resolved } = await invokeClaude(this.claudePath, args);
+    const { reply, usage, model: resolved } = await invokeClaude(this.claudePath, args, options.signal);
     if (resolved) this.resolvedModelId = resolved;
     if (usage) (_a = options.onUsage) == null ? void 0 : _a.call(options, usage);
     return reply;
@@ -414,7 +423,9 @@ class ClaudeCliLLMClient {
               // M3: recall_memory is registered (gated, then served by the parent via onToolExecute) only under AUDIT_RECALL_TOOL.
               ...recallToolEnabled() ? { ENABLE_RECALL_TOOL: "1" } : {},
               // Always passed (even empty): the server reads an unset value as the default, every family off.
-              ASSISTANT_LEXICAL_RESOLVED_OFF: lexicalOffEnvValue()
+              ASSISTANT_LEXICAL_RESOLVED_OFF: lexicalOffEnvValue(),
+              // Eval harness only: absent in the product, so the env block is unchanged.
+              ...process.env.BENCHMARK_TOOL_FAILURE ? { BENCHMARK_TOOL_FAILURE: process.env.BENCHMARK_TOOL_FAILURE } : {}
             }
           }
         }
@@ -443,7 +454,7 @@ class ClaudeCliLLMClient {
       if (model) args.push("--model", model);
       args.push("--", prompt);
       const callStartedAt = Date.now();
-      const { reply, usage, model: resolved } = await invokeClaudeStreaming(this.claudePath, args, options.onToolStep);
+      const { reply, usage, model: resolved } = await invokeClaudeStreaming(this.claudePath, args, options.onToolStep, options.signal);
       if (resolved) this.resolvedModelId = resolved;
       if (usage) (_f = options.onUsage) == null ? void 0 : _f.call(options, usage);
       const staged = await this.findPendingActionStagedSince(workspaceRoot, callStartedAt);
@@ -3995,6 +4006,12 @@ Forgot ${removed} session digest${removed === 1 ? "" : "s"}.
       await handleMemoryForget(args.slice(1));
       return;
     }
+    if (args[0] === "history" && args[1] === "events") {
+      printSafe(`
+${formatMemoryEvents(await assistant.memoryEvents(50, args[2] ? { fact: args[2] } : {}))}
+`);
+      return;
+    }
     if (args[0] === "history") {
       printSafe(`
 ${formatMemoryHistory(await assistant.memoryHistory())}
@@ -4662,6 +4679,8 @@ Usage: /layers ${sub} <id>
 ✗ Saved, but "${id}" is pinned by its AUDIT_* env flag, so the saved choice has no effect until that is unset.
 ` : `
 ✓ Layers updated (took effect immediately, no restart needed)
+`);
+    for (const warning of layerWarnings(sanitizeLayerChoices(config.layers), layerPins, process.env)) printSafe(`⚠ ${warning}
 `);
   }
   async function handleModel(args) {
@@ -4843,7 +4862,7 @@ async function mainInner(argv) {
     });
   }
   if (shouldLaunchTuiApp(config.tuiMode ?? DEFAULT_TUI_MODE, Boolean(process.stdout.isTTY), Boolean(process.stdin.isTTY))) {
-    const { runTuiApp } = await import("./tui-app-Dt72u8vZ.js");
+    const { runTuiApp } = await import("./tui-app-Dqzmfjvc.js");
     await runTuiApp();
     return;
   }
@@ -4870,4 +4889,4 @@ export {
   sanitizeLabel as s,
   toLines as t
 };
-//# sourceMappingURL=bin-CCwDFxDe.js.map
+//# sourceMappingURL=bin-BRwA8Kfr.js.map