Skip to content

Node-based terminal manager for AI coding agents — tmux-backed terminals and parallel agent sessions as draggable nodes on an infinite pan/zoom canvas. macOS, Linux, and a browser Server Edition.

chore: drop references to an audit that is not in the repo

reviewace7ceb619a8
diff --git a/src/core/memory-pressure.ts b/src/core/memory-pressure.ts
index 87d8a3b21..98766d6d2 100644
--- a/src/core/memory-pressure.ts
+++ b/src/core/memory-pressure.ts
@@ -10,13 +10,13 @@
// RSS at 63-200 MB against a 4096 MB threshold, the two legs above never fire on macOS, so the
// renderer's reclaim levers (`releaseAllHiddenGrants` + the park drop) never ran while the window
// was visible — #179 removed the timed hidden-context release on the promise that these would.
-// Measured (2026-10 GPU audit, dev build): the lever pair on 12 hidden warm terminals took the GPU
+// Measured (2026-10, dev build): the lever pair on 12 hidden warm terminals took the GPU
// process footprint 331 → 152 MB and IOSurface 235 → 91 MB; panning back re-granted all 12.
// This leg is deliberately NOT part of check()/onPressure:
// - it fires on TRANSITIONS into warn/critical only, never on a held level. The 60 s re-fire
// floor above at a held WARN (which a Mac can sit in for hours) would be #179's timed release
// again, pan-back flicker included. A level that FLAPS (normal, warn, normal, warn — the
-// audited Mac moved between 1 and 2) is the same timed release by another route, so a run at
+// measured Mac moved between 1 and 2) is the same timed release by another route, so a run at
// the same or a lower severity as the last one also waits out KERNEL_LEVER_GAP_MS; only an
// escalation above the last run's severity (warning → critical) fires inside the gap;
// - it never goes through hostMemReader/MemInfo, which the reaper's planReap shares — a kernel
diff --git a/src/core/pty-manager.ts b/src/core/pty-manager.ts
index 6085b8677..80a53b098 100644
--- a/src/core/pty-manager.ts
+++ b/src/core/pty-manager.ts
@@ -307,7 +307,7 @@ const PROBE_TIMEOUT_MS = 6_000
* `ensureUpdateEnvKeys`' `show-options` / `set-option`). They block the main thread, and one runs
* before `createWindow`: against a tmux server that accepts the connection but never answers
* (stuck on a blocked client write, or SIGSTOPped) an unbounded call froze main with no window
- * and no error — the 2026-10 audit (main-5) reproduced a client held 4.05 s, exactly the wedged
+ * and no error — a 2026-10 repro held a client 4.05 s, exactly the wedged
* listener's lifetime, i.e. forever against a server that never recovers. A healthy server
* answers these in milliseconds, so 2 s only ever fires on a wedge. `vm_stat` already has 5 s.
*/
@@ -2055,7 +2055,7 @@ export class PtyManager {
// The tmux server outlives the app, so it won't re-read `-f` on relaunch. Push the
// (possibly updated) config into a running server now so new bindings apply immediately;
// a no-op error when no server exists yet (the next session loads it fresh via `-f`).
- // Bounded: this runs before the first window exists (audit main-5). A timeout keeps tmux
+ // Bounded: this runs before the first window exists. A timeout keeps tmux
// resolved — only the config push is skipped, and runTmuxSync has already logged it.
this.runTmuxSync(['source-file', this.confPath], false)
}
diff --git a/src/core/pty-tmux-boot-timeout.test.ts b/src/core/pty-tmux-boot-timeout.test.ts
index f5e36ff6d..43dc00641 100644
--- a/src/core/pty-tmux-boot-timeout.test.ts
+++ b/src/core/pty-tmux-boot-timeout.test.ts
@@ -4,12 +4,12 @@ import { initPlatform, resetPlatformForTests } from './platform'
import { fakePlatform } from './platform-fake'
/**
- * THE BOOT-PATH tmux CALLS ARE BOUNDED (audit main-5).
+ * THE BOOT-PATH tmux CALLS ARE BOUNDED.
*
* `ensureTmux`'s `source-file` runs on the main thread before the first window exists, and
* `ensureUpdateEnvKeys`' `show-options` / `set-option` run on a spawn. All three were
* `execFileSync` with no timeout, so a tmux server that accepts the connection and never answers
- * held main for as long as it stayed wedged (4.05 s in the audit's repro — the listener's whole
+ * held main for as long as it stayed wedged (4.05 s in a repro — the listener's whole
* lifetime). These pin the contract with an injected `execFileSync`: every call carries the bound,
* a timeout is WARNED and fails open and starts a cooldown with no further spawn-path probe, and
* the ordinary "no server yet" failure stays as silent as it always was. The real-tmux proof is
diff --git a/src/core/pty-tmux-wedged.realtmux.test.ts b/src/core/pty-tmux-wedged.realtmux.test.ts
index e249a22df..96add2243 100644
--- a/src/core/pty-tmux-wedged.realtmux.test.ts
+++ b/src/core/pty-tmux-wedged.realtmux.test.ts
@@ -1,9 +1,9 @@
-// A WEDGED tmux SERVER NO LONGER HOLDS BOOT (audit main-5), PROVEN AGAINST A REAL tmux CLIENT.
+// A WEDGED tmux SERVER NO LONGER HOLDS BOOT, PROVEN AGAINST A REAL tmux CLIENT.
//
// The unit suite (`pty-tmux-boot-timeout.test.ts`) pins that every boot-path call carries the
// bound; whether the bound actually ends a real client stuck on a server that never answers is a
// property of tmux and the kernel, so it is measured here. The "server" is a unix listener bound
-// where `-L node-terminal` resolves that accepts and then says nothing — the audit's repro, which
+// where `-L node-terminal` resolves that accepts and then says nothing — the original repro, which
// held an unbounded client 4.05 s (the listener's whole lifetime). It works with main blocked in
// `execFileSync`: the kernel completes the connect from the listen backlog, no JS has to run.
//
diff --git a/src/core/remote-ssh/native/native-mux.lazy.test.ts b/src/core/remote-ssh/native/native-mux.lazy.test.ts
index 64184d330..67a2914a9 100644
--- a/src/core/remote-ssh/native/native-mux.lazy.test.ts
+++ b/src/core/remote-ssh/native/native-mux.lazy.test.ts
@@ -1,4 +1,4 @@
-// ssh2 must not load when the native transport is merely IMPORTED (audit startup-6 / roadmap #34).
+// ssh2 must not load when the native transport is merely IMPORTED.
// native-runtime → native-mux sits on main's module-scope import graph, and main's module scope runs
// before Chromium's `ready`: a top-level ssh2 import cost every launch 14–25 ms warm (~680 ms in one
// cold-cache sample) for a transport that only runs on Windows or with NODETERM_NATIVE_SSH=1.
diff --git a/src/core/remote-ssh/native/native-mux.ts b/src/core/remote-ssh/native/native-mux.ts
index 488dac15a..d17fbe18f 100644
--- a/src/core/remote-ssh/native/native-mux.ts
+++ b/src/core/remote-ssh/native/native-mux.ts
@@ -48,8 +48,8 @@ import { addKeyToAgent, type AgentAddResult } from './agent-add'
/**
* ssh2, loaded on first use. This module sits on main's module-scope import graph (via
* native-runtime / native-invoke), so a top-level `import { Client } from 'ssh2'` evaluated ssh2 on
- * every launch, before Chromium's `ready`: 14–25 ms warm, ~680 ms in one cold-cache sample (audit startup-6 /
- * roadmap #34) — for a transport that only runs on Windows or with `NODETERM_NATIVE_SSH=1`. A plain
+ * every launch, before Chromium's `ready`: 14–25 ms warm, ~680 ms in one cold-cache sample — for a
+ * transport that only runs on Windows or with `NODETERM_NATIVE_SSH=1`. A plain
* `require` stays a runtime require in both bundles: electron-vite externalises dependencies, and
* `server:build` passes `--external:ssh2` (server-build.test.ts). Pinned by native-mux.lazy.test.ts.
*/
diff --git a/src/core/session-host-client-staged.test.ts b/src/core/session-host-client-staged.test.ts
index d44d7ef2d..c6869517f 100644
--- a/src/core/session-host-client-staged.test.ts
+++ b/src/core/session-host-client-staged.test.ts
@@ -24,7 +24,7 @@ beforeEach(() => {
})
// Each failing connect polls 30 × 150 ms for a host that never appears (the launcher is mocked).
-// On the real clock that was 4.5 s per connect and 22.6 s for this file on CI (audit code-devx-4);
+// On the real clock that was 4.5 s per connect and 22.6 s for this file on CI;
// the pause is injected instead, so the loop still runs every attempt of its budget.
const connectPollSleep = vi.fn(async (_ms: number) => {})
const newClient = (
diff --git a/src/core/session-host-client-startup-wait.test.ts b/src/core/session-host-client-startup-wait.test.ts
index ea1bffb34..32106451a 100644
--- a/src/core/session-host-client-startup-wait.test.ts
+++ b/src/core/session-host-client-startup-wait.test.ts
@@ -28,7 +28,7 @@ const emptyLock = (): never => {
const tempDir = (): string => testTmpDir('nt-host-wait-')
// The client's poll pause, injected so the 45- and 35-poll waits below cost no wall-clock: on the
-// real 150 ms timer this file took 12 s on CI (audit code-devx-4). Only the pause is replaced, not
+// real 150 ms timer this file took 12 s on CI. Only the pause is replaced, not
// the loop, so the attempt counts these tests pin are exactly what production runs.
const connectPollSleep = vi.fn(async (_ms: number) => {})
const newClient = (): SessionHostClient =>
diff --git a/src/core/session-host-client.ts b/src/core/session-host-client.ts
index f9c84a7bf..9f9157bbf 100644
--- a/src/core/session-host-client.ts
+++ b/src/core/session-host-client.ts
@@ -444,7 +444,7 @@ export class SessionHostClient {
/** The pause between connect polls (the pre-launch publication retries and the post-launch
* wait for the host). Absent = a real timer. Tests inject an immediate one: each failing
* connect waited out 30 × 150 ms = 4.5 s of real time, and the staged/startup-wait suites
- * spent ~35 s of CI doing nothing else (audit code-devx-4). Not fake timers: 94005975. */
+ * spent ~35 s of CI doing nothing else. Not fake timers: 94005975. */
connectPollSleep?: (ms: number) => Promise<void>
}
) {}
diff --git a/src/core/session-memory-service.test.ts b/src/core/session-memory-service.test.ts
index 43934d691..05b3ffd86 100644
--- a/src/core/session-memory-service.test.ts
+++ b/src/core/session-memory-service.test.ts
@@ -394,7 +394,7 @@ describe('mergeLocalLegs', () => {
})
})
-describe('startSessionMemoryService — "nodeterm itself" (roadmap #4)', () => {
+describe('startSessionMemoryService — "nodeterm itself"', () => {
const MEM = { availableMb: 8000, totalMb: 16000 }
const listing = {
running: true,
diff --git a/src/core/session-memory-service.ts b/src/core/session-memory-service.ts
index 7ddd4d923..14a42cd0e 100644
--- a/src/core/session-memory-service.ts
+++ b/src/core/session-memory-service.ts
@@ -34,7 +34,7 @@ export interface SessionMemoryServiceOptions {
/** Host RAM reader, injectable for tests. Defaults to the real `/proc/meminfo` read. */
readMem?: () => MemInfo | null
/**
- * nodeterm's own processes, for the panel's "nodeterm itself" group (audit roadmap #4). Only the
+ * nodeterm's own processes, for the panel's "nodeterm itself" group. Only the
* desktop shell has them (`app.getAppMetrics()`); the Server Edition omits it, and its reports
* then carry no `app` — "not available", which is not "could not measure". Consulted for LOCAL
* sweeps only: an SSH host's report must never include this machine's processes.
diff --git a/src/core/session-memory.test.ts b/src/core/session-memory.test.ts
index 6353fb3dc..d3b525f2e 100644
--- a/src/core/session-memory.test.ts
+++ b/src/core/session-memory.test.ts
@@ -561,10 +561,10 @@ import {
type AppProcessSample
} from './session-memory'
-/** Captured live on the audited Mac, 2026-10-19 (`sysctl vm.swapusage`). */
+/** Captured live on a Mac, 2026-10-19 (`sysctl vm.swapusage`). */
const SWAPUSAGE = 'vm.swapusage: total = 9216.00M used = 7853.62M free = 1362.38M (encrypted)\n'
-describe('parseSwapUsage (darwin, roadmap #14)', () => {
+describe('parseSwapUsage (darwin)', () => {
it('reads used and total from the live string, prefixed or not', () => {
expect(parseSwapUsage(SWAPUSAGE)).toEqual({ swapUsedMb: 7854, swapTotalMb: 9216 })
expect(parseSwapUsage(SWAPUSAGE.replace('vm.swapusage: ', ''))).toEqual({ swapUsedMb: 7854, swapTotalMb: 9216 })
@@ -582,7 +582,7 @@ describe('parseSwapUsage (darwin, roadmap #14)', () => {
})
})
-describe('parsePressureLevel (darwin, roadmap #14)', () => {
+describe('parsePressureLevel (darwin)', () => {
it('maps xnu levels 1/2/4 and nothing else', () => {
expect(parsePressureLevel('1')).toBe('normal')
expect(parsePressureLevel('kern.memorystatus_vm_pressure_level: 2\n')).toBe('warn')
@@ -659,7 +659,7 @@ describe('darwinMemInfo display extras', () => {
})
})
-describe('appSelfMemory — "nodeterm itself" (roadmap #4)', () => {
+describe('appSelfMemory — "nodeterm itself"', () => {
const S = (pid: number, type: string, cpuSeconds?: number): AppProcessSample => ({
pid,
type,
diff --git a/src/core/session-memory.ts b/src/core/session-memory.ts
index cd6d7ee07..95db5389d 100644
--- a/src/core/session-memory.ts
+++ b/src/core/session-memory.ts
@@ -246,8 +246,8 @@ function runDarwinSysctl(names: string[]): string {
}
/**
- * Parse `sysctl vm.swapusage` (with or without the `vm.swapusage:` prefix). Live shape on the
- * audited Mac (2026-10-19):
+ * Parse `sysctl vm.swapusage` (with or without the `vm.swapusage:` prefix). Live shape on a
+ * measured Mac (2026-10-19):
* vm.swapusage: total = 9216.00M used = 7853.62M free = 1362.38M (encrypted)
* `K`/`M`/`G` are accepted; an unknown suffix or a missing half yields NOTHING rather than a guessed
* scale. `total = 0.00M` is a real answer (no swap file yet) and passes through as 0.
@@ -270,7 +270,7 @@ export function parseSwapUsage(text: string): Pick<MemDisplay, 'swapUsedMb' | 's
}
/** `kern.memorystatus_vm_pressure_level`: 1 NORMAL, 2 WARN, 4 CRITICAL (xnu's
- * kVMPressureNormal/Warning/Critical; 2 observed live while the audit's Mac had 7.5-8.7 GB
+ * kVMPressureNormal/Warning/Critical; 2 observed live while a measured Mac had 7.5-8.7 GB
* swapped). Anything else is undefined — never mapped to "normal". */
export function parsePressureLevel(text: string): MemDisplay['pressureLevel'] {
const m = /(?:^|:\s*)(\d+)\s*$/.exec(text.trim())
@@ -682,7 +682,7 @@ export function hostOverhead(
*
* Memory is DELIBERATELY not taken from `getAppMetrics()`. On macOS neither field tracks
* phys_footprint: `memory.workingSetSize` is resident size, as blind to IOSurfaces as `ps` rss
- * (audit tmux-2/obs-1 — the GPU helper's 1557-1845 MB footprint shows as 96-153 MB there), and
+ * (the GPU helper's 1557-1845 MB footprint shows as 96-153 MB there), and
* `memory.privateBytes` is Windows-only. The pid is looked up in the panel sweep's own process
* table instead, so the group uses exactly the accounting the session rows use, with no extra
* `top` (~0.5 s per call, measured).
@@ -708,7 +708,7 @@ export function appProcessKind(type: string): AppProcessKind {
const APP_KIND_ORDER: readonly AppProcessKind[] = ['gpu', 'renderer', 'main', 'helper']
/**
- * Build the "nodeterm itself" group from the sweep's table. Rules (audit §5b invariants 2 and 3):
+ * Build the "nodeterm itself" group from the sweep's table. Rules:
* - a null process list or table → `ok:false`, never an empty-but-ok group;
* - darwin: a pid counts as measured only if `top` gave its footprint (`ProcEntry.footprint`).
* The rss fallback the session rows tolerate is the very number this group exists to replace;
@@ -917,7 +917,7 @@ export async function collectSessionMemory(
if (answered === 0) return { ok: false, rows: [], mem }
const report = buildReport([...bySession.values()], table, mem)
// Same table, no second read: on darwin it already carries `top`'s footprint for EVERY pid,
- // Electron's included (audit §5b: reuse the sweep, the pill never polls `top`).
+ // Electron's included (reuse the sweep; the pill never polls `top`).
if (!deps.appProcesses) return report
return { ...report, app: appSelfMemory(sampleApp(deps.appProcesses), table, process.platform) }
}
diff --git a/src/core/transcript-index-core.test.ts b/src/core/transcript-index-core.test.ts
index a5f933fe5..1fcf6ffda 100644
--- a/src/core/transcript-index-core.test.ts
+++ b/src/core/transcript-index-core.test.ts
@@ -80,7 +80,7 @@ describe('searchEntries', () => {
})
})
-describe('searchEntries matching (RegExp iu, audit #5)', () => {
+describe('searchEntries matching (RegExp iu)', () => {
const one = (text: string, title = 'x'): TranscriptIndexEntry[] => [
{ sessionId: 's', transcriptPath: '/p/s.jsonl', cwd: '/c', mtime: 1, title, text }
]
diff --git a/src/core/transcript-index-core.ts b/src/core/transcript-index-core.ts
index df977bfd1..7feab8613 100644
--- a/src/core/transcript-index-core.ts
+++ b/src/core/transcript-index-core.ts
@@ -46,8 +46,8 @@ export function extractEntryFields(rawText: string): { cwd: string; title: strin
// One case-insensitive matcher per query, shared by the filter and the snippet. The previous
// `e.text.toLowerCase().includes(q)` built a lowered copy of every entry on every debounced
// keystroke: on the user's 16 MB / 235-entry index that measured 20-52 ms and 7-15 MB of garbage
-// per search on the main thread; this RegExp measured 2-13 ms and 0 MB (audit #5, main-2/agents-2,
-// `main-process-bench/search.ts`). Storing a lowered copy instead would retain +32 MB — don't.
+// per search on the main thread; this RegExp measured 2-13 ms and 0 MB.
+// Storing a lowered copy instead would retain +32 MB — don't.
//
// `iu`, not `i`: neither flag reproduces `toLowerCase().includes` exactly, so the criterion is "no
// match the old code found is lost". Over every BMP case pair, plain `i` disagreed 160 times and
diff --git a/src/core/transcript-index.ts b/src/core/transcript-index.ts
index cef0c4bee..259ca2ab9 100644
--- a/src/core/transcript-index.ts
+++ b/src/core/transcript-index.ts
@@ -135,7 +135,7 @@ export function loadPersistedIndex(file: string): TranscriptIndexEntry[] {
*
* It was a bare `writeFile` of ~16 MB every 5 minutes while agents are active. A crash or power
* loss mid-write left truncated JSON, which loads as an empty index and forces a cold rescan —
- * measured at ~779 MB read and ~3.2 s (audit #7, main-3/agents-3/persistence-io-2). Temp + rename
+ * measured at ~779 MB read and ~3.2 s. Temp + rename
* means a reader sees the old index or the new one, never half of either. Mode 0600 because the
* file holds transcript text (it was 0644).
*/
diff --git a/src/main/index.ts b/src/main/index.ts
index 68706cdfe..5bb65a033 100644
--- a/src/main/index.ts
+++ b/src/main/index.ts
@@ -22,7 +22,7 @@ import { IPC } from '../shared/ipc'
// in-memory and redacted at its push boundary; the panel/IPC side is gated on the setting.
const logBuffer = new LogBuffer()
installLogSink(logBuffer)
-// Process-death telemetry (audit roadmap #6, tmux-2 / obs-5): a GPU-process crash used to leave
+// Process-death telemetry: a GPU-process crash used to leave
// no trace at all. Registered here, before 'ready', because the GPU process launches during boot
// and can die there. LOG ONLY — Chromium respawns the GPU process itself; see process-health-log.ts.
const childProcessGoneLine = createChildProcessGoneLog()
@@ -1398,8 +1398,8 @@ function createWindow(): BrowserWindow {
})
}
const crashReload = createCrashReloadPolicy()
- // Renderer hangs (audit roadmap #6): Electron declares one after its own input-ack timeout and
- // clears it on the next ack. Paired into one line with a duration; logged, never acted on.
+ // Renderer hangs: Electron declares one after its own input-ack timeout and clears it on the
+ // next ack. One line when declared, one with the duration when it ends; logged, never acted on.
const hangs = createHangTracker()
win.on('unresponsive', () => {
const line = hangs.unresponsive(Date.now())
@@ -3609,7 +3609,7 @@ app.whenReady().then(async () => {
unmeasuredSessions: () => ptyManager.zellijSessionCount(),
// Windows (no tmux): the session host's sessions; undefined where it is not the backend.
sessionHost: () => ptyManager.sessionHostListing(),
- // The panel's "nodeterm itself" group (audit roadmap #4): pids, types and CPU time only. Memory
+ // The panel's "nodeterm itself" group: pids, types and CPU time only. Memory
// is NOT taken from here — on macOS `memory.workingSetSize` is resident size and misses the GPU
// helper's IOSurfaces (footprint 1557-1845 MB vs 96-153 MB), and `privateBytes` is Windows-only;
// core looks each pid up in the sweep's own footprint table instead. Called once per panel
diff --git a/src/main/lazy-deps.source.test.ts b/src/main/lazy-deps.source.test.ts
index 3cc516711..9ecfa8c89 100644
--- a/src/main/lazy-deps.source.test.ts
+++ b/src/main/lazy-deps.source.test.ts
@@ -1,4 +1,4 @@
-// Source-level pin (audit startup-6 / roadmap #34): `ssh2` and `electron-updater` are never
+// Source-level pin: `ssh2` and `electron-updater` are never
// VALUE-imported by shipped main/core/shared/server code. Main's module scope runs before Chromium's
// `ready`, and the two used to cost every launch 14–25 ms (ssh2) + 32–47 ms (electron-updater) warm
// there. Each now loads at first use — `ssh2()` in native-mux.ts, `UpdaterDeps.loadUpdater` in
diff --git a/src/main/process-health-log.ts b/src/main/process-health-log.ts
index 3924a7f8b..b87d8b3c3 100644
--- a/src/main/process-health-log.ts
+++ b/src/main/process-health-log.ts
@@ -1,4 +1,4 @@
-// Process-death and hang telemetry for the debug log ring (audit roadmap #6, tmux-2 / obs-5).
+// Process-death and hang telemetry for the debug log ring.
//
// Before this, the only process-death handler was the main window's `render-process-gone`, and
// it logged nothing: a GPU-process crash, a utility-process OOM or a multi-second renderer hang
@@ -9,7 +9,7 @@
//
// OBSERVE, NEVER ACT. Nothing here reloads, relaunches or kills anything. Chromium respawns a
// dead GPU process itself; relaunching on top of that would turn a recoverable GPU reset into a
-// lost session (and the audit's tmux-4 note: repeated GPU deaths can trip Chromium's per-domain
+// lost session (and repeated GPU deaths can trip Chromium's per-domain
// 3D-API block "until restart", so anything that provokes more of them degrades the session).
//
// Electron-free on purpose (structural types only) so it unit-tests without an Electron runtime.
@@ -34,7 +34,7 @@ export interface ChildProcessGoneDetails {
* Format app 'child-process-gone' and count it per process type for the session. Returns a
* stateful formatter so the count lives with the app, not in a module global a test would leak.
*
- * Only `{type, reason, exitCode, name}` — the audit's field list. `name` / `serviceName` are
+ * Only `{type, reason, exitCode, name}`. `name` / `serviceName` are
* Chromium's fixed service labels ("Audio Service", "Network Service"), never a path or argv.
*
* Level: a `clean-exit` is Chromium tearing down a utility process it no longer needs — routine,
diff --git a/src/main/remote/host-service.ts b/src/main/remote/host-service.ts
index 6cab02e51..82dbcbebc 100644
--- a/src/main/remote/host-service.ts
+++ b/src/main/remote/host-service.ts
@@ -79,7 +79,7 @@ const FRESH_PROBE_BUDGET_MS = 750
* Relay backpressure recovery for a paused phone stream. `sendFrame` refuses (and DROPS) a frame
* once the socket holds more than 8 MB (`MAX_BINARY_BUFFERED_AMOUNT`); the host then pauses the
* pty. A paused node-pty emits no `onData`, so a resume driven off "the next successful send" can
- * never happen — the audit (REPORT §4.7 net-1) found exactly that: after one >8 MB burst the phone's
+ * never happen — and it did not: after one >8 MB burst the phone's
* terminal froze until it detached. The sweep is the drain trigger instead, mirroring
* `core/ui-sink-registry.ts` (RESYNC_SWEEP_MS / WS_LOW_WATER: "never the next chunk as the drain
* trigger"). It exists only while some stream is paused, so the healthy path pays nothing.
diff --git a/src/main/remote/relay-advertise.test.ts b/src/main/remote/relay-advertise.test.ts
index 28768e70f..ac673c8fd 100644
--- a/src/main/remote/relay-advertise.test.ts
+++ b/src/main/remote/relay-advertise.test.ts
@@ -1,4 +1,4 @@
-// Audit net-5 (REPORT §4.7): the standing host rewrote ~/.nodeterm/relay.json with identical
+// The standing host used to rewrite ~/.nodeterm/relay.json with identical
// content on every ~90 s listener cycle. The advertiser now skips an unchanged write, but must still
// rewrite when the content changes, when the file has gone, and after a remove. Every test runs
// against a temp dir through the factory — never the live ~/.nodeterm.
diff --git a/src/main/remote/relay-advertise.ts b/src/main/remote/relay-advertise.ts
index b0768072c..3fca502df 100644
--- a/src/main/remote/relay-advertise.ts
+++ b/src/main/remote/relay-advertise.ts
@@ -27,8 +27,8 @@ export interface RelayAdvertisement {
/**
* The advertisement writer for one file. The standing host calls `write` on EVERY successful
- * listener open — every ~90 s, 24/7, with byte-identical content (audit REPORT §4.7 net-5 measured
- * the mtime refreshing each cycle: ~960 mkdir + temp-write + rename a day for nothing). So an
+ * listener open — every ~90 s, 24/7, with byte-identical content (measured: the mtime
+ * refreshing each cycle, ~960 mkdir + temp-write + rename a day for nothing). So an
* unchanged advertisement is skipped: the last string we wrote is kept in memory, and the file is
* rewritten only when the content changes or the file has gone (a user or another tool deleted it —
* one stat per cycle is what that costs). `remove` forgets it, so the next start always writes.
diff --git a/src/main/updater.no-channel.test.ts b/src/main/updater.no-channel.test.ts
index aa0701570..47df321ab 100644
--- a/src/main/updater.no-channel.test.ts
+++ b/src/main/updater.no-channel.test.ts
@@ -65,7 +65,7 @@ vi.mock('./notifications', () => ({ retainUntilDismissed: () => {} }))
import { describeCheckFailure, initUpdater as initUpdaterReal, lastUpdateCheck, type UpdaterHooks } from './updater'
-// electron-updater is loaded lazily with `require` (startup-6), which `vi.mock` does not reach, so
+// electron-updater is loaded lazily with `require`, which `vi.mock` does not reach, so
// the double goes in through the loader seam instead of a module mock.
const initUpdater = (hooks?: UpdaterHooks) =>
initUpdaterReal(hooks, { loadUpdater: () => autoUpdater as unknown as import('electron-updater').AppUpdater })
@@ -336,7 +336,7 @@ describe('restart-to-update elsewhere', () => {
})
/**
- * Audit startup-6 / roadmap #34: electron-updater is no longer required at main's module scope
+ * electron-updater is no longer required at main's module scope
* (32–47 ms warm, before Chromium's `ready`, on every launch). It is loaded only on a path that uses
* it, and the IPC handlers stay registered first whatever the build: the renderer invokes them on
* mount, and an invoke with no handler rejects.
diff --git a/src/main/updater.restart.test.ts b/src/main/updater.restart.test.ts
index 0a5d93b26..10fc1227a 100644
--- a/src/main/updater.restart.test.ts
+++ b/src/main/updater.restart.test.ts
@@ -75,7 +75,7 @@ vi.mock('./notifications', () => ({ retainUntilDismissed: () => {} }))
import { initUpdater as initUpdaterReal, PREPARE_TIMEOUT_MS, type UpdaterHooks } from './updater'
-// electron-updater is loaded lazily with `require` (startup-6), which `vi.mock` does not reach, so
+// electron-updater is loaded lazily with `require`, which `vi.mock` does not reach, so
// the double goes in through the loader seam instead of a module mock.
const initUpdater = (hooks?: UpdaterHooks) =>
initUpdaterReal(hooks, { loadUpdater: () => autoUpdater as unknown as import('electron-updater').AppUpdater })
diff --git a/src/main/updater.ts b/src/main/updater.ts
index 48c276f2b..e4a5bba2b 100644
--- a/src/main/updater.ts
+++ b/src/main/updater.ts
@@ -8,7 +8,7 @@
import { app, autoUpdater as nativeUpdater, ipcMain, Notification } from 'electron'
import fs from 'fs'
import path from 'path'
-// electron-updater is NOT imported at module scope (audit startup-6 / roadmap #34): main's module
+// electron-updater is NOT imported at module scope: main's module
// scope runs before Chromium's `ready`, and requiring electron-updater there measured 32–47 ms warm
// on every launch — including dev and `no-channel` builds that never touch it. It is loaded by
// `initUpdater` (after createWindow) only on a path that uses it; see `UpdaterDeps`. Type-only here.
diff --git a/src/renderer/canvas/SharedGlyphLayer.test.ts b/src/renderer/canvas/SharedGlyphLayer.test.ts
index 14d52e897..054ee6dfe 100644
--- a/src/renderer/canvas/SharedGlyphLayer.test.ts
+++ b/src/renderer/canvas/SharedGlyphLayer.test.ts
@@ -1244,7 +1244,7 @@ describe('createPixelRatioWatcher', () => {
*/
describe('boardCoversCanvas', () => {
// The predicate the board frame gate and the blink clock read. It must say "covered" for BOTH
- // full-page boards, because Canvas's `kanbanOpen` does: the Omni board was missing (audit GT-6),
+ // full-page boards, because Canvas's `kanbanOpen` does: the Omni board was missing,
// so the shared canvas kept drawing under the all-projects board.
const setOmni = (enabled: boolean): void => {
useSettings.setState((s) => ({ settings: { ...s.settings, omniKanbanEnabled: enabled } }))
diff --git a/src/renderer/canvas/SharedGlyphLayer.tsx b/src/renderer/canvas/SharedGlyphLayer.tsx
index c13a4eb6e..322e17b77 100644
--- a/src/renderer/canvas/SharedGlyphLayer.tsx
+++ b/src/renderer/canvas/SharedGlyphLayer.tsx
@@ -1551,7 +1551,7 @@ export function setSharedGlyphCamera(cam: Camera): void {
* Including the OMNI board (`isGlobalKanbanOpen`, which already folds in the feature flag): it is
* the same opaque full-page overlay, and Canvas's `kanbanOpen` is `globalKanbanOpen ||
* perProjectKanbanOpen`. Leaving it out kept the shared canvas drawing under the all-projects
- * board — the same omission 12564a20 fixed for the window-activation focus restore (audit GT-6).
+ * board — the same omission 12564a20 fixed for the window-activation focus restore.
* Like Canvas's own term, Omni needs no active project.
*/
export function boardCoversCanvas(): boolean {
diff --git a/src/renderer/components/SessionMemoryPanel.test.tsx b/src/renderer/components/SessionMemoryPanel.test.tsx
index d45772880..3f7081909 100644
--- a/src/renderer/components/SessionMemoryPanel.test.tsx
+++ b/src/renderer/components/SessionMemoryPanel.test.tsx
@@ -483,7 +483,7 @@ describe('SessionMemoryPanel', () => {
})
})
- describe('"nodeterm itself" (roadmap #4)', () => {
+ describe('"nodeterm itself"', () => {
const APP = {
ok: true,
groups: [
diff --git a/src/renderer/components/SessionMemoryPanel.tsx b/src/renderer/components/SessionMemoryPanel.tsx
index 2280b21d6..99268f8ae 100644
--- a/src/renderer/components/SessionMemoryPanel.tsx
+++ b/src/renderer/components/SessionMemoryPanel.tsx
@@ -291,7 +291,7 @@ export function SessionMemoryPanel({
nodeterm itself.
</div>
- {/* nodeterm's OWN processes (audit roadmap #4), below the attribution line on purpose: they
+ {/* nodeterm's OWN processes, below the attribution line on purpose: they
are not a session and are NOT in the header total. The GPU helper is the reason this
exists — `ps` said 96-153 MB while its footprint was 1557-1845 MB, mostly IOSurfaces (the
window's swap chain and raster tiles, and the terminals' WebGL buffers), and no surface
diff --git a/src/renderer/components/SystemResourcePill.test.tsx b/src/renderer/components/SystemResourcePill.test.tsx
index ad986bd5a..18549f26e 100644
--- a/src/renderer/components/SystemResourcePill.test.tsx
+++ b/src/renderer/components/SystemResourcePill.test.tsx
@@ -230,13 +230,13 @@ describe('SystemResourcePill', () => {
expect(panel?.classList.contains('sessmem-panel')).toBe(true)
})
- describe('swap and kernel pressure (roadmap #14, display-only)', () => {
+ describe('swap and kernel pressure (display-only)', () => {
const pillText = (): string => host.querySelector('.sysres-pill')?.textContent ?? ''
const title = (): string => host.querySelector('.sysres-pill')?.getAttribute('title') ?? ''
const icon = (): string => host.querySelector<HTMLElement>('.sysres-pill__icon')?.style.color ?? ''
it('shows swap in use beside the RAM figure, and used-of-total in the title', () => {
- // The audited Mac: 7.7 GB of 9 GB swapped while the pill read only "86%".
+ // A measured Mac: 7.7 GB of 9 GB swapped while the pill read only "86%".
mount({ ...MEM, display: { swapUsedMb: 7854, swapTotalMb: 9216 } })
expect(pillText()).toContain('7.7 GB swap')
expect(title()).toContain('swap 7.7 GB of 9.0 GB used')
diff --git a/src/renderer/components/SystemResourcePill.tsx b/src/renderer/components/SystemResourcePill.tsx
index 32bd71524..f2171e8ee 100644
--- a/src/renderer/components/SystemResourcePill.tsx
+++ b/src/renderer/components/SystemResourcePill.tsx
@@ -134,12 +134,12 @@ export function SystemResourcePill({
// The ICON carries the pressure reading now that the bar is gone: it tints as memory is
// consumed, so a glance still says "this machine is tight" without the pill claiming a row of
// the canvas. Same thresholds the bar used, raised (never lowered) by the kernel's pressure
- // level where the OS reports one (audit roadmap #14).
+ // level where the OS reports one.
const color = TINT[Math.max(percentSeverity(usedPercent), pressureSeverity(mem)) as Severity]
const swap = mem.display
// Swap is shown only when BOTH halves were measured (`display` is display-only: the reaper never
- // reads it). On the audited Mac the pill said 86% "warn" with 8.5 GB swapped and nothing on
- // screen said so (audit agent-trees-3). No figure when the read failed or the OS has none —
+ // reads it). On a measured Mac the pill said 86% "warn" with 8.5 GB swapped and nothing on
+ // screen said so. No figure when the read failed or the OS has none —
// never "+0 GB"; a measured 0 with swap provisioned is a fact and only goes in the title.
const swapShown =
swap?.swapUsedMb !== undefined && swap.swapTotalMb !== undefined && swap.swapUsedMb > 0
diff --git a/src/renderer/editor/monaco-setup.ts b/src/renderer/editor/monaco-setup.ts
index 90850bb7f..bde52a40e 100644
--- a/src/renderer/editor/monaco-setup.ts
+++ b/src/renderer/editor/monaco-setup.ts
@@ -19,7 +19,7 @@ self.MonacoEnvironment = {
}
}
-// The TS/JS worker never idles out on its own (audit #18, startup-4): stop it once the last
+// The TS/JS worker never idles out on its own: stop it once the last
// TS/JS model is gone. `monaco.typescript` is the 0.57 namespace; `monaco.languages.typescript`
// is a deprecated stub there. See tsWorkerReaper.ts for the mechanism and the grace period.
const tsDefaults = {
diff --git a/src/renderer/editor/tsWorkerReaper.test.ts b/src/renderer/editor/tsWorkerReaper.test.ts
index 64c60a691..05aa588a4 100644
--- a/src/renderer/editor/tsWorkerReaper.test.ts
+++ b/src/renderer/editor/tsWorkerReaper.test.ts
@@ -4,7 +4,7 @@ import { createTsWorkerReaper, TS_WORKER_STOP_GRACE_MS, type TsWorkerLanguage }
// The reaper's decision, driven the way monaco-setup.ts drives it: `live` stands in for
// `editor.getModels()`, and — like Monaco's `onWillDisposeModel` — a model is still in `live`
// when `modelLeft` fires and only removed after.
-describe('tsWorkerReaper (audit #18: stop the TS/JS worker once no TS/JS model remains)', () => {
+describe('tsWorkerReaper (stop the TS/JS worker once no TS/JS model remains)', () => {
let live: string[]
let stopped: TsWorkerLanguage[]
let reaper: ReturnType<typeof createTsWorkerReaper>
diff --git a/src/renderer/editor/tsWorkerReaper.ts b/src/renderer/editor/tsWorkerReaper.ts
index 8c500cf12..9d75f908d 100644
--- a/src/renderer/editor/tsWorkerReaper.ts
+++ b/src/renderer/editor/tsWorkerReaper.ts
@@ -1,6 +1,6 @@
// Stop Monaco's TypeScript / JavaScript worker once no model of that language is left.
//
-// Why (audit roadmap #18, startup-4): in monaco-editor 0.57 the css/html/json WorkerManagers stop
+// Why: in monaco-editor 0.57 the css/html/json WorkerManagers stop
// their worker after 2 min idle, but the TypeScript WorkerManager has NO idle path — it only stops
// when its defaults fire `onDidChange` (or on dispose), and it is created with `keepIdleModels`.
// So the first .ts/.tsx/.js file opened in an EditorNode or DiffNode starts a worker isolate holding
@@ -9,7 +9,6 @@
// unmeasured). The supported way to stop one is a no-op `setCompilerOptions(getCompilerOptions())`:
// `onDidChange` -> `WorkerManager._stopWorker()`, and the next request lazily starts a fresh worker
// with the same options and extra libs (both live on the defaults, not the worker).
-// `noSemanticValidation` does NOT stop the worker (REPORT §4.9) — don't reach for it here.
//
// The decision is kept Monaco-free so it is testable in node: monaco-setup.ts passes the real
// model list and the real stop; tests pass fakes and fake timers.
diff --git a/src/renderer/lib/chatLive.test.ts b/src/renderer/lib/chatLive.test.ts
index 6bd047f35..cf0c03b2b 100644
--- a/src/renderer/lib/chatLive.test.ts
+++ b/src/renderer/lib/chatLive.test.ts
@@ -101,7 +101,7 @@ describe('shouldPollScreen — reading the pane for the agent\'s own dialogs', (
})
})
-describe('screenPollTickAllowed — one tick of the screen poll (audit main-6)', () => {
+describe('screenPollTickAllowed — one tick of the screen poll', () => {
it('runs while the window is visible and focused', () => {
expect(screenPollTickAllowed({ hidden: false, focused: true })).toBe(true)
})
diff --git a/src/renderer/lib/chatLive.ts b/src/renderer/lib/chatLive.ts
index 6abc7a152..687bc9ea4 100644
--- a/src/renderer/lib/chatLive.ts
+++ b/src/renderer/lib/chatLive.ts
@@ -71,7 +71,7 @@ export function shouldPollScreen(s: {
* while the window is unfocused: a nodeterm window left visible on a second monitor is not hidden
* by Chromium's definition, so the hidden gate alone kept every chat view spawning a `tmux
* capture-pane` every 2 s — ~30 spawns a minute per open chat view — for a composer nobody could
- * type into (perf audit 2026-10, main-6 / roadmap #21). The panel re-ticks on `focus`, so the
+ * type into. The panel re-ticks on `focus`, so the
* screen is re-read the moment someone can act on it, and a send re-checks the screen in core
* (`pty.sendChatPrompt`) regardless, so skipping a tick never weakens dialog safety.
*
diff --git a/src/renderer/lib/sessionMemoryRows.test.ts b/src/renderer/lib/sessionMemoryRows.test.ts
index 8ea640870..dd22252f9 100644
--- a/src/renderer/lib/sessionMemoryRows.test.ts
+++ b/src/renderer/lib/sessionMemoryRows.test.ts
@@ -87,7 +87,7 @@ describe('totalMb', () => {
})
})
-describe('"nodeterm itself" labels (roadmap #4)', () => {
+describe('"nodeterm itself" labels', () => {
it('names each group, with a count only when there is more than one', () => {
expect(appGroupLabel({ kind: 'gpu', count: 1 })).toBe('GPU')
expect(appGroupLabel({ kind: 'renderer', count: 2 })).toBe('Renderers ×2')
diff --git a/src/renderer/nodes/ChatPanel.screenDialog.test.tsx b/src/renderer/nodes/ChatPanel.screenDialog.test.tsx
index f01325c74..6831df374 100644
--- a/src/renderer/nodes/ChatPanel.screenDialog.test.tsx
+++ b/src/renderer/nodes/ChatPanel.screenDialog.test.tsx
@@ -71,7 +71,7 @@ async function typeAndEnter(text: string): Promise<void> {
}
// jsdom's `hasFocus()` is false (nothing focused the test window), and the poll skips unfocused
-// windows (audit main-6) — so every test starts with a focused window, and the focus tests flip it.
+// windows — so every test starts with a focused window, and the focus tests flip it.
let focused = true
beforeEach(() => {
diff --git a/src/renderer/nodes/ChatPanel.tsx b/src/renderer/nodes/ChatPanel.tsx
index c3354a1dc..34ddd0a71 100644
--- a/src/renderer/nodes/ChatPanel.tsx
+++ b/src/renderer/nodes/ChatPanel.tsx
@@ -686,7 +686,7 @@ export function ChatPanel({
// Re-read the pane's screen while the view is visible (see `screenBlock`). A read that fails or
// finds a blank screen changes nothing: unknown is not evidence either way.
// Ticks are skipped while the window is unfocused (`screenPollTickAllowed`: ~30 tmux spawns a
- // minute per chat view otherwise, audit main-6), and coming back — `focus`, or the page turning
+ // minute per chat view otherwise), and coming back — `focus`, or the page turning
// visible — re-reads at once, so a dialog that opened meanwhile is on screen before anyone types.
useEffect(() => {
if (!pollScreen) {
diff --git a/src/renderer/state/sessionMemory.test.ts b/src/renderer/state/sessionMemory.test.ts
index 6ddad07dd..b6fc282d0 100644
--- a/src/renderer/state/sessionMemory.test.ts
+++ b/src/renderer/state/sessionMemory.test.ts
@@ -154,7 +154,7 @@ describe('failure semantics', () => {
})
})
-describe('"nodeterm itself" group (roadmap #4)', () => {
+describe('"nodeterm itself" group', () => {
const APP = { ok: true, groups: [{ kind: 'gpu', count: 1, mb: 1600 }], totalMb: 1600 }
it('carries the group from an answered sweep, and drops it when the sweep failed or the scope changed', async () => {
diff --git a/src/renderer/terminal/board-render-pause.ts b/src/renderer/terminal/board-render-pause.ts
index 8f7febd63..818186f2b 100644
--- a/src/renderer/terminal/board-render-pause.ts
+++ b/src/renderer/terminal/board-render-pause.ts
@@ -4,7 +4,7 @@
* WHY. The canvas stays mounted under the board on purpose (`display: none` would 0x0-resize every
* terminal into a tmux SIGWINCH — see the kanban section of CLAUDE.md), and nothing told xterm it
* was covered: its only pause is its own GEOMETRIC IntersectionObserver (`RenderService`), and a
- * node under an opaque overlay still intersects the viewport. MEASURED (audit GT-6, dev build, 12
+ * node under an opaque overlay still intersects the viewport. MEASURED (dev build, 12
* streaming terminals, glass on, 60 s windows): board open over streaming terminals 36.5 % GPU /
* 15.8 % renderer — no better than the canvas itself (30-32 %) — against 0.0 % / 2.4 % with the same
* board over SILENT terminals, so essentially all of it was xterm painting output nobody could see
@@ -27,7 +27,7 @@
* WHAT THIS DOES NOT DO, ON PURPOSE. It never feeds the app's own visibility verdict
* (`reportVisible` / the WebGL budget's `setVisible` / hibernation / the offscreen teardown):
* folding the board into that verdict would let the board hibernate or tear down the sessions it
- * is showing cards for (audit GT-6, "refuted fix shapes"). The WebGL context stays held while the
+ * is showing cards for. The WebGL context stays held while the
* board is up; only the draws stop.
*
* Private xterm API (`_core._renderService`), exactly as TerminalNode's renderer-swap net and
@@ -110,8 +110,8 @@ export function installBoardRenderPause(
// (if any) handed its real verdict back. Not a blanket `true`: a terminal re-adopted after a park
// was detached, so its observer already delivered not-intersecting (`_isPaused = true`); reattached
// still off screen, the observer has nothing new to report, and a `true` seed made the next board
- // cover/uncover UNPAUSE it — streaming output painted off screen until a pan in and out (review
- // of the GT-6 fix, reproduced against xterm 5.5's handler). A fresh xterm starts unpaused with no
+ // cover/uncover UNPAUSE it — streaming output painted off screen until a pan in and out (found in
+ // review, reproduced against xterm 5.5's handler). A fresh xterm starts unpaused with no
// delivery yet, so the same read gives intersecting there. A build without the field: `true`.
const pausedNow = (rs as { _isPaused?: unknown })._isPaused
let geometric = typeof pausedNow === 'boolean' ? !pausedNow : true
diff --git a/src/renderer/terminal/webgl-budget.test.ts b/src/renderer/terminal/webgl-budget.test.ts
index fcebf4757..b7fdc0ef7 100644
--- a/src/renderer/terminal/webgl-budget.test.ts
+++ b/src/renderer/terminal/webgl-budget.test.ts
@@ -819,7 +819,7 @@ describe('loseWebglContexts', () => {
})
})
-describe('WebGL loss telemetry (audit roadmap #6)', () => {
+describe('WebGL loss telemetry', () => {
let warn: ReturnType<typeof vi.spyOn>
beforeEach(() => {
vi.useFakeTimers()
diff --git a/src/renderer/terminal/webgl-budget.ts b/src/renderer/terminal/webgl-budget.ts
index a56e4a3ce..294c6c73d 100644
--- a/src/renderer/terminal/webgl-budget.ts
+++ b/src/renderer/terminal/webgl-budget.ts
@@ -360,7 +360,7 @@ export function loseWebglContexts(canvases: ArrayLike<HTMLCanvasElement> | null)
}
/**
- * Per-session WebGL loss telemetry (audit roadmap #6, tmux-2 / obs-5). Before this a GPU reset
+ * Per-session WebGL loss telemetry. Before this a GPU reset
* left no trace: the terminals silently fell back to the DOM renderer and were re-granted, and
* the only evidence was a user saying "the text went soft for a second". Each EXTERNAL loss now
* writes one `[webgl]` warn — which main's `console-message` mirror carries into the debug log
diff --git a/src/shared/keybindings.test.ts b/src/shared/keybindings.test.ts
index 235a94500..711afedf1 100644
--- a/src/shared/keybindings.test.ts
+++ b/src/shared/keybindings.test.ts
@@ -151,7 +151,7 @@ describe('registry invariants', () => {
darwin: ['Cmd+ArrowUp'], other: ['Ctrl+Shift+ArrowUp'], allowInTerminal: true },
{ id: 'node.focusDown', title: 'Focus node below', group: 'Nodes', scope: 'canvas',
darwin: ['Cmd+ArrowDown'], other: ['Ctrl+Shift+ArrowDown'], allowInTerminal: true },
- // Pool only (audit roadmap #19): no default chord on either platform.
+ // Pool only: no default chord on either platform.
{ id: 'node.nextNeedingYou', title: 'Go to next agent needing you', group: 'Nodes',
scope: 'canvas', darwin: [], other: [], allowInTerminal: true },
{ id: 'node.maximize', title: 'Maximize / restore node', group: 'Nodes', scope: 'canvas',
diff --git a/src/shared/keybindings.ts b/src/shared/keybindings.ts
index 64132da7b..1af4148ea 100644
--- a/src/shared/keybindings.ts
+++ b/src/shared/keybindings.ts
@@ -246,8 +246,7 @@ export const COMMAND_DEFINITIONS: readonly CommandDefinition[] = [
defaultBindings: { darwin: ['Cmd+ArrowDown'], other: ['Ctrl+Shift+ArrowDown'] },
allowInTerminal: true },
// Travel to the agent that has waited longest for you (waiting/blocked), across projects; each
- // press steps to the next one (audit roadmap #19 / qol-2: one chord instead of 3-5 clicks
- // through the sidebar). Ships UNBOUND — every obvious chord already has an owner, and a default
+ // press steps to the next one, without going through the sidebar. Ships UNBOUND — every obvious chord already has an owner, and a default
// would steal it from a user who never asked. Scope 'canvas' + allowInTerminal for the
// node.focus* reason: the hand that wants to move is in a terminal, and over the kanban board
// there is no canvas to land on. The queue is built on demand per press (lib/nextNeedingYou).
diff --git a/src/shared/types.ts b/src/shared/types.ts
index 64acacba9..08a8bcb5d 100644
--- a/src/shared/types.ts
+++ b/src/shared/types.ts
@@ -3159,11 +3159,11 @@ export interface MemInfo {
psiFullAvg60?: number
/**
* What the RAM pill SHOWS beyond used/total. **Display only: never an input to the reaper, the
- * memory-pressure monitor or a reclaim lever** (audit §5b invariant 1). It is a separate object,
+ * memory-pressure monitor or a reclaim lever**. It is a separate object,
* rather than more top-level fields, so that rule is structural: `hostUnderMemoryPressure` reads
* `swapTotalMb`/`swapFreeMb`, and filling THOSE on darwin would fire the reaper's swap term all
- * day on a Mac that keeps its swap ~85% used as a matter of course (audit agent-trees-2/3: 16%
- * swap free while healthy). Absent field = not measured, never zero.
+ * day on a Mac that can keep most of its swap in use for hours (16% swap free on the measured
+ * one). Absent field = not measured, never zero.
*/
display?: MemDisplay
}
@@ -3252,7 +3252,7 @@ export interface AppSelfGroup {
}
/**
- * The audit's roadmap #4: the GPU helper's real footprint (1557-1845 MB live, mostly IOSurfaces)
+ * The GPU helper's real footprint (1557-1845 MB live, mostly IOSurfaces)
* was invisible because `ps` RSS says 96-153 MB and nothing attributed it. `ok:false` (no groups)
* when the process list or the table could not be read, or no pid of ours was in it.
*/
diff --git a/test/remote/host-handlers.test.ts b/test/remote/host-handlers.test.ts
index 86e1bc6c3..4f7b9c22d 100644
--- a/test/remote/host-handlers.test.ts
+++ b/test/remote/host-handlers.test.ts
@@ -257,7 +257,7 @@ describe('createHostHandlers', () => {
expect(pty.calls).toHaveLength(0)
})
- // Audit net-1 (REPORT §4.7): a paused node-pty emits NO further onData, so a resume that waits for
+ // A paused node-pty emits NO further onData, so a resume that waits for
// "the next successful send" never comes and the phone's terminal froze for good after one >8 MB
// burst. These tests deliver nothing after the pause — the drain sweep alone must resume it.
describe('relay backpressure: the drain sweep resumes a paused stream', () => {