aiden-runtime 4.1.4 → 4.1.5

package/dist/cli/v4/display.js

@@ -18,12 +18,13 @@
  *
  */
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.TOOL_ROW_ARG_CAP = exports.TOOL_ROW_NAME_PAD = exports.Display = exports.SPINNER_PHRASES = exports.TOOL_ICONS = void 0;
+exports.TRAIL_HIDE_TOOLS = exports.TOOL_ROW_ARG_CAP = exports.TOOL_ROW_NAME_PAD = exports.Display = exports.SPINNER_PHRASES = exports.TOOL_ICONS = void 0;
 exports.iconForTool = iconForTool;
 exports.detectConfiguredChannels = detectConfiguredChannels;
 exports.summarizeConfiguredChannels = summarizeConfiguredChannels;
 exports.summarizeChannelState = summarizeChannelState;
 exports.voiceIndicator = voiceIndicator;
+exports.makeNoOpToolRowHandle = makeNoOpToolRowHandle;
 exports.previewToolArgs = previewToolArgs;
 exports.verbForActivity = verbForActivity;
 exports.isPreFramedLine = isPreFramedLine;
@@ -772,7 +773,21 @@ class Display {
      * content below MUST call `pause()` first; otherwise their content
      * lands on the indicator line and the next tick clobbers it.
      */
-    activityIndicator(initialVerb = 'thinking') {
+    /**
+     * v4.1.5 Issue K — wave-bar option.
+     *
+     * When `opts.waveBar === true` (DEFAULT), the indicator paints a
+     * second row BELOW the verb line — a 10-cell `▰▱` snake-scroll
+     * heartbeat that gives visible motion during long pre-first-token
+     * gaps even when the verb doesn't change. The bar is NOT progress:
+     * it's a constant-cadence heartbeat (250ms shared with the dot
+     * pulse), explicitly not a percentage indicator.
+     *
+     * Pass `{ waveBar: false }` for back-compat with v4.1.4 tests that
+     * assert single-row geometry. Production callers (chatSession) get
+     * the wave bar by default.
+     */
+    activityIndicator(initialVerb = 'thinking', opts = {}) {
         const sk = this.skin;
         const out = this.out;
         const isTty = !!out.isTTY;
@@ -799,6 +814,13 @@ class Display {
         // feedback; a separate bottom-of-screen footer can be added in
         // v4.1.5 if wanted, but it must NOT be glued to the indicator.
         const glyph = sk.applyColors('▲', 'brand');
+        // v4.1.5 Issue K — wave-bar state. Snake-scroll: a 3-cell `▰`
+        // block slides across 10 cells, wrapping at the right edge. Same
+        // 250ms tick as the verb dot pulse — one timer drives both rows.
+        const waveBarEnabled = opts.waveBar !== false; // default true
+        const WAVE_CELLS = 10;
+        const WAVE_BLOCK = 3;
+        let waveFrame = 0;
         const buildLine = () => {
             const dots = '.'.repeat(dotFrame); // 0..3 dots
             const elapsedSec = Math.floor((Date.now() - startTime) / 1000);
@@ -808,14 +830,92 @@ class Display {
             // `▲ {verb}{dots-padded-to-3}{elapsed?}`
             return `${glyph} ${verb}${dots.padEnd(3, ' ')}${elapsedStr}`;
         };
+        /**
+         * v4.1.5 Issue K — render the wave-bar row. A 3-cell `▰` block at
+         * positions `[waveFrame, waveFrame+1, waveFrame+2]` mod 10. The
+         * filled cells paint brand orange, empty cells paint warm-muted.
+         * Same width + glyph set as the token progress bar so the two
+         * rows feel like a coherent palette (one is heartbeat, the other
+         * is real progress).
+         *
+         * Heartbeat semantics: this is NOT progress. The wave moves at a
+         * constant 250ms cadence regardless of any backend metric. It
+         * exists purely so the user sees motion during the unobservable
+         * TTFT (time-to-first-token) wait. The verb row above carries
+         * any real lifecycle signal via `setVerb()`.
+         */
+        const buildWave = () => {
+            // v4.1.5 Phase 1d (Q-P1) — glyph palette switch. Was `▰`/`▱`
+            // (U+25B0/B1, Geometric Shapes) which legacy Windows console
+            // fonts render as tofu. Now `▓`/`░` (U+2593/91, Block Elements
+            // — in CP437, universally supported). Matches the existing
+            // statusFooter chrome that's shipped since v3 without ever
+            // being garbled.
+            const filled = new Set();
+            for (let i = 0; i < WAVE_BLOCK; i += 1) {
+                filled.add((waveFrame + i) % WAVE_CELLS);
+            }
+            // Render cells in order so the snake-scroll visually slides:
+            // we paint cell-by-cell with the right color, joined into one
+            // string. ANSI runs reset per cell — slight overhead but keeps
+            // glyph order true to position. Brand orange filled, warm-muted
+            // empty.
+            const cells = [];
+            for (let c = 0; c < WAVE_CELLS; c += 1) {
+                cells.push(filled.has(c)
+                    ? sk.applyColors('▓', 'brand')
+                    : sk.applyColors('░', 'muted'));
+            }
+            return cells.join('');
+        };
+        // v4.1.5 Part 1a — Issue M (Windows ConPTY buffering fix).
+        //
+        // Prior pattern wrote `\r\x1b[K{indicator}` with NO trailing
+        // newline. On Windows ConPTY, `process.stdout` buffers no-newline
+        // writes — none of the 60 indicator ticks during a 15s gap
+        // actually rendered. The final reply's `\n` chars eventually
+        // flushed the buffer, but by then the indicator's stop()-erase
+        // had also been buffered + flushed, so the user saw 15s of blank
+        // followed by the reply dumping all at once.
+        //
+        // Fix: indicator OWNS one terminal row. Every write that paints
+        // the indicator ends with `\n`, which forces a flush on every
+        // platform. The cursor sits on the LINE BELOW the indicator
+        // while it's running (one visible empty row gap). When the
+        // indicator stops/pauses, we walk back UP to the indicator's
+        // row and erase it — the cursor then sits at col 0 of that
+        // (now empty) row, ready for the caller to write whatever
+        // content follows (header, tool row, stream output).
+        //
+        // ANSI primitives:
+        //   `\x1b[1A` — cursor up 1 line
+        //   `\x1b[2K` — erase the whole current line
+        // Sequence on tick:    walk up → erase → paint → `\n` → cursor below.
+        // Sequence on erase:   walk up → erase (no newline). Cursor on the
+        //                      now-empty indicator row, ready for caller.
+        const ANSI_UP_ERASE = '\x1b[1A\x1b[2K';
         const renderTick = () => {
             if (stopped || paused || !isTty)
                 return;
             dotFrame = (dotFrame + 1) % 4;
-            // `\r\x1b[K` — carriage return + clear line — then write the
-            // fresh indicator. No newline at end: cursor stays at end of
-            // the indicator line, ready for the next overwrite.
-            out.write(`\r\x1b[K${buildLine()}`);
+            // v4.1.5 Issue K — wave snake-scroll advances 1 cell per tick.
+            // Same 250ms cadence as the dot pulse, so both rows move in
+            // visible lockstep. Modulo WAVE_CELLS wraps the leading block
+            // back to the left edge.
+            waveFrame = (waveFrame + 1) % WAVE_CELLS;
+            if (waveBarEnabled) {
+                // 2-row layout: walk up TWO rows (two separate up-1+erase
+                // sequences, which keeps the `\x1b[1A\x1b[2K` substring
+                // assertion-compatible), repaint both, drop newlines so the
+                // cursor lands on the row below the wave bar.
+                out.write(`${ANSI_UP_ERASE}${ANSI_UP_ERASE}` +
+                    `${buildLine()}\n` +
+                    `${buildWave()}\n`);
+            }
+            else {
+                // Single-row layout (back-compat with v4.1.4 tests).
+                out.write(`${ANSI_UP_ERASE}${buildLine()}\n`);
+            }
         };
         const startTick = () => {
             if (stopped || !isTty || tickTimer !== null)
@@ -829,12 +929,38 @@ class Display {
             }
         };
         const eraseLine = () => {
-            if (isTty && printed)
-                out.write('\r\x1b[K');
+            // Walk up to the indicator's row(s) + erase. Cursor lands at
+            // col 0 of the (now empty) verb row. NO trailing newline here:
+            // the caller is about to write content on this row, and
+            // whatever they write will include their own `\n` to flush
+            // the buffer. If we emitted `\n` here, we'd leave a phantom
+            // blank row before the caller's content.
+            //
+            // v4.1.5 Issue K — with wave bar enabled, walk up 2 rows (two
+            // up-1+erase sequences). Without the bar, walk up 1 row.
+            if (!isTty || !printed)
+                return;
+            if (waveBarEnabled) {
+                out.write(`${ANSI_UP_ERASE}${ANSI_UP_ERASE}`);
+            }
+            else {
+                out.write(ANSI_UP_ERASE);
+            }
         };
-        // Initial paint — only on TTY.
+        // Initial paint — only on TTY. Indicator + `\n` so the buffer
+        // flushes and the cursor sits on the row below, ready for the
+        // first tick to walk back up.
+        //
+        // v4.1.5 Issue K — when wave bar is enabled, paint TWO rows:
+        // verb row + wave row, each with trailing `\n`. Cursor lands on
+        // the row below the wave bar. The first tick will walk up 2.
         if (isTty) {
-            out.write(buildLine());
+            if (waveBarEnabled) {
+                out.write(`${buildLine()}\n${buildWave()}\n`);
+            }
+            else {
+                out.write(`${buildLine()}\n`);
+            }
             printed = true;
             startTick();
         }
@@ -861,9 +987,18 @@ class Display {
                     return;
                 // Caller has just finished writing its own content (typically
                 // ending with `\n`), so the cursor is on a fresh line below
-                // whatever was there. Render the indicator there and arm the
-                // tick again.
-                out.write(buildLine());
+                // whatever was there. Paint the indicator + `\n` to claim the
+                // current row(s) and leave the cursor on the row below — same
+                // invariant the initial paint and tick maintain. Trailing `\n`
+                // also flushes Windows ConPTY buffering (Issue M).
+                //
+                // v4.1.5 Issue K — repaint BOTH rows when wave bar enabled.
+                if (waveBarEnabled) {
+                    out.write(`${buildLine()}\n${buildWave()}\n`);
+                }
+                else {
+                    out.write(`${buildLine()}\n`);
+                }
                 printed = true;
                 startTick();
             },
@@ -894,6 +1029,31 @@ class Display {
     // completion so each line in the log carries the final state — no
     // ANSI cursor games on a dumb sink.
     toolRow(name, args) {
+        // v4.1.5 Phase 1d (Q-Q2-a) — TRAIL_HIDE_TOOLS suppression.
+        //
+        // Some tools are pure agent plumbing — the model calls them to
+        // introspect its own registry, not to do user-visible work.
+        // `lookup_tool_schema` is the canonical case: during planning
+        // the agent may invoke it 30+ times to discover unfamiliar tool
+        // shapes. Each call is a sub-millisecond in-memory lookup, but
+        // they flood the visible trail with noise that obscures the
+        // actual user-relevant tool calls.
+        //
+        // Short-circuit: hidden tools get a NO-OP handle that satisfies
+        // the `ToolRowHandle` contract (ok/fail/degraded/retry/blocked/
+        // emptyRetry/emptyFail all defined but write nothing). The
+        // execution path itself is unaffected — the agent still calls
+        // the tool, the planner / skill-enforcement trackers still
+        // record it. Only the visual row is suppressed.
+        //
+        // CRITICAL invariant: `setBeforeFirstToolHook` is fired by
+        // callbacks.ts BEFORE `toolRow()` is called (see callbacks.ts
+        // onToolCall 'before' branch), so `turnHadTools` flips even for
+        // hidden tools. The separator logic stays correct regardless of
+        // whether ONLY hidden tools fired this turn.
+        if (exports.TRAIL_HIDE_TOOLS.has(name)) {
+            return makeNoOpToolRowHandle();
+        }
         const sk = this.skin;
         // ── Build the fixed left portion (icon + verb + detail) ────────────
         // v4.1.3-repl-polish: icons default ON; set AIDEN_UI_ICONS=0 to
@@ -995,8 +1155,31 @@ class Display {
                     writeFinal(`ok ${formatToolDuration(durationMs)} after ${retries} ${retries === 1 ? 'retry' : 'retries'}`, 'warn');
                 }
                 else {
-                    // Clean success — SILENT. Erase on TTY; emit nothing on non-TTY.
-                    eraseLast();
+                    // v4.1.5 Issue N — persistent tool trail in scrollback.
+                    //
+                    // Prior behaviour: silent erase on clean success (`eraseLast()`
+                    // with no replacement write). Tool rows for successful tools
+                    // vanished, leaving only the markdown reply visible afterward.
+                    // The user couldn't see WHAT actions Aiden took unless a tool
+                    // failed or degraded.
+                    //
+                    // Fix: replace the silent erase with a completed-state row
+                    // painted entirely in warm-muted (`#b8a89a` from v4.1.4). The
+                    // duration suffix replaces the live `running Ns…` chrome; the
+                    // whole row reads "done" via reduced visual weight. Failed /
+                    // degraded / retry outcomes keep their existing coloured paint
+                    // (error red, degraded yellow, warn amber) — only clean success
+                    // shifts from "silent" to "muted-persistent."
+                    //
+                    // The persistence mechanism is the existing `writeFinal` path:
+                    // it walks up + erases the running row, then writes the final
+                    // row with trailing `\n`. The row sits in scrollback because
+                    // `streamComplete` rerenders only the post-tool stream chunk
+                    // (via `streamLineCount` which was reset to 0 inside
+                    // `commitStreamChunk` before this row wrote). No additional
+                    // isolation machinery needed — already verified by 13/13
+                    // `smoke-stream-rerender.ts` regressions.
+                    writeFinal(formatToolDuration(durationMs), 'muted');
                 }
             },
             fail(durationMs, retries = 0) {
@@ -1698,6 +1881,56 @@ function renderRmsBar(rms) {
 exports.TOOL_ROW_NAME_PAD = toolTrail_1.TRAIL_VERB_PAD;
 /** @deprecated Use TRAIL_DETAIL_CAP. */
 exports.TOOL_ROW_ARG_CAP = toolTrail_1.TRAIL_DETAIL_CAP;
+/**
+ * v4.1.5 Phase 1d (Q-Q2-a) — names of tools that should be SUPPRESSED
+ * from the visible tool-trail row, even though they still execute
+ * normally through the agent loop.
+ *
+ * The canonical case is `lookup_tool_schema`: the agent calls it
+ * during planning to introspect tool registry entries (in-memory
+ * registry get, sub-millisecond per call). On complex prompts the
+ * model may fire it 30+ times in a row, flooding the visible trail
+ * with rows that don't represent user-meaningful work. Suppressing
+ * them keeps the trail focused on the tools that did real work
+ * (web_search, file_read, etc.).
+ *
+ * Suppression happens at `Display.toolRow()` entry — it returns a
+ * no-op handle that satisfies the `ToolRowHandle` contract but
+ * never writes to stdout. The agent's `callbacks.onToolCall`
+ * dispatch is unchanged: `setBeforeFirstToolHook` still fires (so
+ * `turnHadTools` flips for the separator-emission logic), and
+ * skill-enforcement / honesty-trace tracking still records the
+ * call. Only the visual row is hidden.
+ *
+ * Exported as a `Set` so callers can mutate at runtime if they
+ * need to hide additional tools (e.g. user customization, MCP
+ * plumbing tools). Mutation-of-shared-state is intentional — there's
+ * no per-session config plumbing for "trail hidden tools" yet, so
+ * the env-var pattern (`AIDEN_TRAIL_HIDE=tool1,tool2`) would be the
+ * v4.1.6 evolution.
+ */
+exports.TRAIL_HIDE_TOOLS = new Set([
+    'lookup_tool_schema',
+]);
+/**
+ * v4.1.5 Phase 1d helper — produces a `ToolRowHandle` that satisfies
+ * the contract but writes nothing. Used by hidden tools (see
+ * `TRAIL_HIDE_TOOLS`) and as a safe fallback. All methods are inert.
+ *
+ * Pure — no side effects, no closures over Display state. Safe to
+ * call from any thread / phase.
+ */
+function makeNoOpToolRowHandle() {
+    return {
+        ok: () => { },
+        fail: () => { },
+        degraded: () => { },
+        retry: () => { },
+        blocked: () => { },
+        emptyRetry: () => { },
+        emptyFail: () => { },
+    };
+}
 /**
  * Build a compact, single-line preview of the tool's arguments. Picks
  * the most informative scalar fields when the args are an object, then

package/dist/cli/v4/toolPreview.js

@@ -69,6 +69,16 @@ exports.TOOL_PRIMARY_ARG = {
     skill_view: 'name',
     skill_manage: 'action',
     skills_list: '',
+    // v4.1.5 Phase 1d (Q-Q1-a) — registry introspection tool. Args
+    // shape: `{ toolName: 'web_search' }`. The agent uses this to
+    // discover unfamiliar tool schemas during planning. Surface the
+    // target tool name so the trail row (when not suppressed via
+    // TRAIL_HIDE_TOOLS) reads as the introspected tool, not raw JSON.
+    // Note: most callers see this tool suppressed entirely from the
+    // visible trail via the TRAIL_HIDE_TOOLS set in display.ts; the
+    // extractor exists for code paths that DON'T suppress (verbose
+    // mode, log-file capture).
+    lookup_tool_schema: 'toolName',
     // ── sessions ─────────────────────────────────────────────────────────
     session_search: 'query',
     session_list: '',

package/dist/core/toolRegistry.js

@@ -1388,7 +1388,13 @@ exports.TOOLS = {
             return { success: false, output: '', error: `No research results for: ${topic}` };
         }
         const combined = results.join('\n\n');
-        console.log(`[deep_research] Complete: ${combined.length} chars across ${results.length} passes`);
+        // v4.1.5 Issue O — gated behind AIDEN_DEBUG_WEB to match the
+        // webSearch.ts debug-helper convention. Default off; power users
+        // export the env var to see the research chain.
+        if (process.env.AIDEN_DEBUG_WEB === '1') {
+            // eslint-disable-next-line no-console
+            console.log(`[deep_research] Complete: ${combined.length} chars across ${results.length} passes`);
+        }
         return { success: true, output: combined.slice(0, 15000) };
     },
     // Activate a specialist agent persona — actual synthesis happens in respond phase

package/dist/core/v4/aidenAgent.js

@@ -103,6 +103,10 @@ class AidenAgent {
         this.onCompression = opts.onCompression;
         this.refreshMemorySnapshot = opts.refreshMemorySnapshot;
         this.onMemoryRefresh = opts.onMemoryRefresh;
+        // v4.1.5 Issue K — phase hooks (all optional, fire defensively).
+        this.onMemoryRefreshStart = opts.onMemoryRefreshStart;
+        this.onPromptBuilt = opts.onPromptBuilt;
+        this.onProviderRequestStart = opts.onProviderRequestStart;
         this.lookupSkillRequiredTools = opts.lookupSkillRequiredTools;
         // Phase v4.1.2-slice3: optional health registry (constructor-
         // injected per the slice3 decision tree — no singleton). When
@@ -386,6 +390,14 @@ class AidenAgent {
         // / 'user' need a snapshot refresh first.
         const needsSnapshot = this.memoryDirty.has('memory') || this.memoryDirty.has('user');
         if (needsSnapshot && this.refreshMemorySnapshot) {
+            // v4.1.5 Issue K — fire BEFORE the file I/O so the display layer
+            // can switch the activity verb to "refreshing memory" while the
+            // read is in flight. Defensive try/catch so a misbehaving hook
+            // never blocks the refresh.
+            try {
+                this.onMemoryRefreshStart?.();
+            }
+            catch { /* defensive */ }
             let snapshot;
             try {
                 snapshot = await this.refreshMemorySnapshot();
@@ -410,6 +422,21 @@ class AidenAgent {
         if (this.cachedSystemPrompt !== null)
             return this.cachedSystemPrompt;
         this.cachedSystemPrompt = await this.promptBuilder.build(this.promptBuilderOptions);
+        // v4.1.5 Issue K — fire AFTER the prompt has been assembled, with
+        // cardinality so the display layer can surface "preparing prompt:
+        // N tools, M skills" or similar. Only fires when the cache MISSED
+        // (which is what made us actually build); cached returns skip the
+        // hook because nothing was prepared this turn. Defensive try/catch.
+        if (this.onPromptBuilt) {
+            try {
+                this.onPromptBuilt({
+                    tools: this.tools.length,
+                    skills: this.promptBuilderOptions.skillsList?.length ?? 0,
+                    memoryFacts: countMemoryFacts(this.promptBuilderOptions.memorySnapshot),
+                });
+            }
+            catch { /* defensive */ }
+        }
         return this.cachedSystemPrompt;
     }
     async narrowTools(userMsg, history) {
@@ -629,6 +656,18 @@ class AidenAgent {
      */
     async callProvider(messages, tools, runOptions) {
         const wantStream = runOptions.stream === true && typeof this.provider.callStream === 'function';
+        // v4.1.5 Issue K — fire just before the HTTP request opens, so the
+        // display layer can transition the activity verb from local-prep
+        // ("preparing prompt", "selecting tools") to a network verb
+        // ("calling provider"). The wait for TTFT (time-to-first-token) is
+        // the longest gap in most turns and is what the wave bar covers.
+        // Fires for both streaming and non-streaming paths — caller may use
+        // it to add a one-shot indicator on non-streaming providers too.
+        // Defensive try/catch (a misbehaving hook must not block dispatch).
+        try {
+            this.onProviderRequestStart?.(this.providerId);
+        }
+        catch { /* defensive */ }
         if (!wantStream) {
             return this.provider.call({ messages, tools });
         }
@@ -671,6 +710,30 @@ class AidenAgent {
 }
 exports.AidenAgent = AidenAgent;
 // ── Free helpers ────────────────────────────────────────────────────────
+/**
+ * v4.1.5 Issue K — best-effort count of "memory facts" from a
+ * MemorySnapshot. Counts markdown bullet-list lines (`- `) in both
+ * MEMORY.md and USER.md. This is a fuzzy proxy — the agent stores
+ * facts as bullets by convention but free-form prose can also carry
+ * fact-like content. Surfaced verbatim to the display layer; treat as
+ * "approximately N items in the persistent memory file" rather than
+ * a precise inventory.
+ */
+function countMemoryFacts(snapshot) {
+    if (!snapshot || typeof snapshot !== 'object')
+        return 0;
+    const s = snapshot;
+    let count = 0;
+    for (const md of [s.memoryMd, s.userMd]) {
+        if (typeof md !== 'string' || md.length === 0)
+            continue;
+        for (const line of md.split('\n')) {
+            if (line.trim().startsWith('- '))
+                count += 1;
+        }
+    }
+    return count;
+}
 function lastUserMessageContent(history) {
     for (let i = history.length - 1; i >= 0; i--) {
         const m = history[i];