fix(audit): resolve HIGH + MEDIUM + perf/correctness findings from multi-agent repo audit (#185)

* fix(audit): resolve HIGH + MEDIUM + perf/correctness findings from multi-agent repo audit

A 13-subsystem multi-agent audit (adversarially verified, 18 findings refuted)
surfaced 86 confirmed issues. This resolves the 1 HIGH, all 16 MEDIUM, and the
high-value perf/low/doc items — ~42 findings — each with regression tests.
Full suite: 1558 pass. Swift runner build-for-testing SUCCEEDED. The multi-renderer
getTree fix was live-verified on the booted Bridgeless+Reanimated test app over CDP.

HIGH
- collect-feedback.sh redact() failed OPEN on macOS (invalid sed bracket range)
  leaking ghp_/AKIA/JWT/Bearer/sk_ secrets into public feedback issues — now one
  sed program that fails CLOSED ([REDACTION_FAILED]); + bash regression test.

MEDIUM (correctness/inconsistency)
- device_batch swipe was a scroll; now a real swipe gesture honoring duration
- runIOS/postCommand + android client: AbortController per-command HTTP timeout
- ref-map @ref coords gated on snapshot age (isRefMapFresh) -> STALE_REF when stale
- iOS raw screenshot honors png format via extension; NaN tap coords rejected
- repair-engine extract/replaceIdSelector match the maestro-error-parser quoted-id
  grammar (quoted/commented testIDs no longer skip auto-repair)
- experience: candidate-id regex matches real filenames; imported heuristic ids
  deduped; ghost recovery gated to idempotent tools; redaction catches quoted-JSON
  /object-auth secrets; PII regex no longer over-redacts bare numbers
- cross_platform_verify FAIL returns ok:false (was warnResult/ok:true)
- runMaestroInline routes through chooseMaestroDispatch (B59 iOS-only fallback)
- injected getTree() unfiltered branch walks ALL renderers (B143/B145)
- nav-graph strikes namespaced per project (no cross-project cooldown poisoning)
- swift textInputAt restricts query to input types; eas_resolve_artifact.sh portable stat

PERF
- meta.timings_ms helper + scrollintoview instrumentation (CLAUDE.md mandate)
- object_inspect concurrent child fetch; nav_graph go single graph read (was 4x)
- connect candidate loop O(n^2) indexOf -> indexed; auto-login polls instead of 3s wait

LOW / correctness + docs
- guarded JSON.parse (navigation-state, android client); navigateTo loop depth cap
- sidecar missing lastSeenMtimeMs re-seeded (re-enables human-edit guard)
- proof_step no-session verify -> verified:false; device_record path-traversal guard
- device_find surfaces recoveredTier on NOT_FOUND/AMBIGUOUS; detectPrerequisites
  whole-word permission match; learned-actions --max NaN fallback; collect-feedback
  grep double-zero fix; iOS /health probe 127.0.0.1 (was [::1])
- tool-count drift (banner 51, CLAUDE.md 74 -> 75); REACT_READY renderer range 1-5 -> 1-20
- repair budget refusal reports 24h-windowed count, not lifetime total

Swift (xcodebuild build-for-testing verified)
- rn-fast-runner answers GET /health directly (was unparseable -> liveness always stale)
- snapshotHasFocus uses `as?` not `as!` (force-cast trap is uncatchable)

Deferred (~20 moderate-risk perf + cosmetic items) tracked for a focused follow-up.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(audit): clear the deferred perf + cosmetic findings (B12)

Follow-up to the audit fixes — addresses the remaining lower-severity items. Full suite still 1558 pass; tsc clean.

Perf/lifecycle: logger buffered WriteStream (no event-loop block); connectWs backstop terminates a wedged socket (no leak); cdp_network_log truncated/total accurate on the unfiltered path via buffer size.

Correctness/clarity: maestro_run pass/fail no longer false-flags benign 'Error:' (exit code authoritative); fingerprint engine = Hermes only with an RN signal else null; groupFailures unreachable passed++ removed; multiplexer debug-logs unrouted-id drops; RunIOSArgs dead x1/y1 removed; path-safety sidecar-path comment fixed; ensure-agent-device uses npm prefix -g (npm 9+); expo_ensure_running preserves temp logs on failure.

Not-actioned (correct as-is): lockfile release (signals route to exit->release; SIGKILL covered by mtime), dumpsys (action=all runs once), persistRun CAS re-read (required after conflict), test-recorder dedup edge / buildNavigationPlan scan / loadExperience forceReload (dev-only/negligible).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: Lykhoyda

CLAUDE.md

@@ -123,7 +123,7 @@ Fallback: `xcrun simctl` (iOS) + `adb` (Android) for device lifecycle (boot / in
 
 ### MCP Server (cdp-bridge)
 
-**74 tools** exposed via MCP (re-audited 2026-05-18; counted from `trackedTool()` calls in `scripts/cdp-bridge/src/index.ts`). Five conceptual families:
+**75 tools** exposed via MCP (re-audited 2026-05-29; counted from `trackedTool()` calls in `scripts/cdp-bridge/src/index.ts`). Five conceptual families:
 
 **CDP tools** — React internals via Chrome DevTools Protocol over WebSocket:
 - `cdp_status` — health check with domain capabilities + reconnect state

hooks/detect-rn-project.sh

@@ -91,7 +91,7 @@ if [ "$has_rn_config" = true ]; then
   fi
 
   cat <<'EOF'
-React Native project detected. The rn-dev-agent plugin is active with 51 MCP tools.
+React Native project detected. The rn-dev-agent plugin is active with 75 MCP tools.
 
 ## How to interact with the running app
 

scripts/cdp-bridge/dist/agent-device-wrapper.js

@@ -7,7 +7,7 @@ import { homedir } from 'node:os';
 import { createHash } from 'node:crypto';
 import { okResult, failResult } from './utils.js';
 import { isFastRunnerAvailable, startFastRunner, } from './runners/rn-fast-runner-client.js';
-import { refCenter, getScreenRect, clearRefMap } from './fast-runner-ref-map.js';
+import { refCenter, getScreenRect, clearRefMap, isRefMapFresh } from './fast-runner-ref-map.js';
 import { resolveBundleId } from './project-config.js';
 const execFile = promisify(execFileCb);
 /**
@@ -284,22 +284,26 @@ const RN_FAST_RUNNER_COMMANDS = new Set([
 export function getCachedScreenRect() {
     return getScreenRect();
 }
-function buildRunIOSArgs(cliArgs, bundleId) {
+export function buildRunIOSArgs(cliArgs, bundleId) {
     const cmd = cliArgs[0];
     const positionals = cliArgs.slice(1).filter((a) => !a.startsWith('--'));
     switch (cmd) {
         case 'press':
         case 'tap': {
             const ref = positionals[0];
             if (ref && ref.startsWith('@')) {
-                const center = refCenter(ref);
+                const center = isRefMapFresh() ? refCenter(ref) : null;
                 if (!center) {
                     return { command: 'tap', _staleRef: ref, ...(bundleId ? { bundleId } : {}) };
                 }
                 return { command: 'tap', x: center.x, y: center.y, ...(bundleId ? { bundleId } : {}) };
             }
             const [xS, yS] = positionals;
-            return { command: 'tap', x: Number(xS), y: Number(yS), ...(bundleId ? { bundleId } : {}) };
+            const x = Number(xS), y = Number(yS);
+            if (Number.isNaN(x) || Number.isNaN(y)) {
+                throw new Error(`buildRunIOSArgs: tap requires a @ref or numeric x, y`);
+            }
+            return { command: 'tap', x, y, ...(bundleId ? { bundleId } : {}) };
         }
         case 'fill':
         case 'type': {
@@ -310,7 +314,7 @@ function buildRunIOSArgs(cliArgs, bundleId) {
             const ref = positionals[0];
             const text = positionals.slice(1).join(' ');
             if (ref && ref.startsWith('@')) {
-                const center = refCenter(ref);
+                const center = isRefMapFresh() ? refCenter(ref) : null;
                 if (!center) {
                     return { command: 'type', _staleRef: ref, text, ...(bundleId ? { bundleId } : {}) };
                 }
@@ -414,7 +418,7 @@ function androidPositionals(cliArgs) {
     }
     return out;
 }
-function buildRunAndroidArgs(cliArgs, bundleId) {
+export function buildRunAndroidArgs(cliArgs, bundleId) {
     const cmd = cliArgs[0];
     const positionals = androidPositionals(cliArgs);
     const withBundle = bundleId ? { bundleId } : {};
@@ -423,20 +427,24 @@ function buildRunAndroidArgs(cliArgs, bundleId) {
         case 'tap': {
             const ref = positionals[0];
             if (ref && ref.startsWith('@')) {
-                const center = refCenter(ref);
+                const center = isRefMapFresh() ? refCenter(ref) : null;
                 if (!center)
                     return { command: 'tap', _staleRef: ref, ...withBundle };
                 return { command: 'tap', x: center.x, y: center.y, ...withBundle };
             }
             const [xS, yS] = positionals;
-            return { command: 'tap', x: Number(xS), y: Number(yS), ...withBundle };
+            const x = Number(xS), y = Number(yS);
+            if (Number.isNaN(x) || Number.isNaN(y)) {
+                throw new Error(`buildRunAndroidArgs: tap requires a @ref or numeric x, y`);
+            }
+            return { command: 'tap', x, y, ...withBundle };
         }
         case 'fill':
         case 'type': {
             const ref = positionals[0];
             const text = positionals.slice(1).join(' ');
             if (ref && ref.startsWith('@')) {
-                const center = refCenter(ref);
+                const center = isRefMapFresh() ? refCenter(ref) : null;
                 if (!center)
                     return { command: 'type', _staleRef: ref, text, ...withBundle };
                 return { command: 'type', x: center.x, y: center.y, text, ...withBundle };

scripts/cdp-bridge/dist/cdp/connect.js

@@ -67,7 +67,9 @@ discoverFn = discover) {
         throw new Error(selectionWarning ?? 'No matching CDP targets found.');
     }
     let connectedTarget = null;
-    for (const candidate of sorted) {
+    for (let idx = 0; idx < sorted.length; idx++) {
+        const candidate = sorted[idx];
+        const isLast = idx === sorted.length - 1;
         try {
             await connectToTarget(ctx, candidate);
             const devCheck = await ctx.evaluate('typeof __DEV__ !== "undefined" && __DEV__ === true');
@@ -76,7 +78,7 @@ discoverFn = discover) {
                 break;
             }
             console.error(`CDP: target ${candidate.id} (${candidate.title}) has __DEV__=${devCheck.value}, skipping`);
-            if (sorted.indexOf(candidate) < sorted.length - 1) {
+            if (!isLast) {
                 closeAndResetWs(ctx);
                 ctx.setState('disconnected');
                 ctx.setHelpersInjected(false);
@@ -87,7 +89,7 @@ discoverFn = discover) {
             connectedTarget = candidate;
         }
         catch (err) {
-            if (sorted.indexOf(candidate) < sorted.length - 1)
+            if (!isLast)
                 continue;
             throw err;
         }
@@ -210,15 +212,34 @@ function connectWs(ctx, url) {
             maxPayload: 100 * 1024 * 1024,
         });
         let settled = false;
+        // Backstop: handshakeTimeout should emit 'error', but if the socket ever
+        // wedges without firing open/error/close it would leak with its listeners.
+        // Terminate it after a grace window so it can't linger.
+        const guard = setTimeout(() => {
+            if (settled)
+                return;
+            settled = true;
+            try {
+                ws.terminate();
+            }
+            catch { /* already gone */ }
+            reject(new Error('WebSocket connect timed out'));
+        }, 7000);
         ws.on('open', () => {
             settled = true;
+            clearTimeout(guard);
             ctx.setWs(ws);
             ctx.setState('connected');
             resolve();
         });
         ws.on('error', (err) => {
             if (!settled) {
                 settled = true;
+                clearTimeout(guard);
+                try {
+                    ws.terminate();
+                }
+                catch { /* already closing */ }
                 reject(err);
             }
             else {
@@ -231,6 +252,7 @@ function connectWs(ctx, url) {
         ws.on('close', (code) => {
             if (!settled) {
                 settled = true;
+                clearTimeout(guard);
                 reject(new Error(`WebSocket closed before connecting: ${code}`));
                 return;
             }

scripts/cdp-bridge/dist/cdp/multiplexer.js

@@ -236,8 +236,12 @@ export class CDPMultiplexer {
             return;
         }
         const route = this.routingTable.get(upstreamId);
-        if (!route)
+        if (!route) {
+            // No routing entry for this upstream id — a late reply after the entry was
+            // swept, or an id-space collision. Log it (debug) so it isn't silent.
+            logger.debug(this.opts.logTag, `dropping Hermes reply with unrouted id=${upstreamId}`);
             return;
+        }
         this.routingTable.delete(upstreamId);
         m.id = route.consumerOriginalId;
         const rewritten = JSON.stringify(m);

scripts/cdp-bridge/dist/domain/path-safety.js

@@ -27,7 +27,7 @@ export class PathTraversalError extends Error {
 }
 // ── Action ID validation ────────────────────────────────────────────
 // Action IDs flow into `.rn-agent/actions/<id>.yaml` and the matching
-// sidecar `<id>.action.json`. They must start with an alphanumeric
+// sidecar `.rn-agent/state/<id>.state.json`. They must start with an alphanumeric
 // character, then accept hyphen/underscore/dot/alphanumeric. We
 // deliberately exclude `..`, `/`, `\`, and control characters so the
 // ID can never escape its parent directory.

scripts/cdp-bridge/dist/domain/repair-engine.js

@@ -136,24 +136,35 @@ export function replaceIdSelector(body, oldId, newId) {
     const lines = body.split('\n');
     const out = [];
     let replacements = 0;
-    // Match the line-trimmed `id: <quoted-or-bare><oldId><quoted-or-bare>`.
-    // Capture leading whitespace + quote style so we can preserve them.
-    // Allowed quotes: `"X"`, `'X'`, or bare `X` if it has no spaces.
+    // Match the line-trimmed `id: <quoted-or-bare><oldId>`, preserving leading
+    // whitespace, quote style, and any trailing `# comment`. Three explicit
+    // shapes (double / single / bare) keep this in lockstep with
+    // extractIdSelectors and the maestro-error-parser matched-quote grammar:
+    // a double-quoted value may contain `'` and vice-versa.
     const escapedOld = oldId.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-    const re = new RegExp(`^(\\s*)id:\\s*("?)${escapedOld}("?)\\s*$`);
-    const reSingle = new RegExp(`^(\\s*)id:\\s*(')${escapedOld}(')\\s*$`);
+    const dq = new RegExp(`^(\\s*)id:\\s*"${escapedOld}"(\\s*(?:#.*)?)$`);
+    const sq = new RegExp(`^(\\s*)id:\\s*'${escapedOld}'(\\s*(?:#.*)?)$`);
+    const bare = new RegExp(`^(\\s*)id:\\s*${escapedOld}(\\s*(?:#.*)?)$`);
     for (const line of lines) {
-        const m = line.match(re) ?? line.match(reSingle);
+        let m = line.match(dq);
         if (m) {
-            const indent = m[1];
-            const quoteOpen = m[2] ?? '';
-            const quoteClose = m[3] ?? '';
-            out.push(`${indent}id: ${quoteOpen}${newId}${quoteClose}`);
+            out.push(`${m[1]}id: "${newId}"${m[2]}`);
             replacements++;
+            continue;
         }
-        else {
-            out.push(line);
+        m = line.match(sq);
+        if (m) {
+            out.push(`${m[1]}id: '${newId}'${m[2]}`);
+            replacements++;
+            continue;
+        }
+        m = line.match(bare);
+        if (m) {
+            out.push(`${m[1]}id: ${newId}${m[2]}`);
+            replacements++;
+            continue;
         }
+        out.push(line);
     }
     return { body: out.join('\n'), replacements };
 }
@@ -168,22 +179,27 @@ export function replaceIdSelector(body, oldId, newId) {
 export function extractIdSelectors(body) {
     const out = [];
     const lines = body.split('\n');
-    // Issue #102 A2: previous regex `[^"'\\n]*?` greedily included
-    // trailing inline comments — `id: foo-bar  # human note` captured
-    // `foo-bar  # human note` as the testID. Hand-authored YAML
-    // sometimes has these comments; the bug fails safely (no fuzzy
-    // match) but produces a UX issue. Fix: strip trailing
-    // `\s+#.*` from the captured group before pushing. Also tighten
-    // the bare-form regex to reject `#` directly so quoted forms are
-    // unaffected.
-    const re = /^\s*id:\s*"?'?([^"'\s][^"'\n]*?)"?'?\s*$/;
+    // Mirror the maestro-error-parser matched-quote grammar (PR #115) so the
+    // failure parser and this extractor agree on what a testID is. Previously
+    // the char class `[^"'\s]` rejected any testID containing a quote (e.g.
+    // `user's-task`), so attemptRepair's gate short-circuited to
+    // 'no-stale-selector' and auto-repair silently no-op'd for ids the parser
+    // had correctly extracted. Three explicit shapes:
+    //   id: "value"   — value may contain '
+    //   id: 'value'   — value may contain "
+    //   id: value     — bare; strip a trailing ` # comment` (Issue #102 A2)
+    const dq = /^\s*id:\s*"([^"\n]*)"\s*(?:#.*)?$/;
+    const sq = /^\s*id:\s*'([^'\n]*)'\s*(?:#.*)?$/;
+    const bare = /^\s*id:\s*([^"'#\s][^#\n]*?)\s*(?:#.*)?$/;
     for (const line of lines) {
-        const m = line.match(re);
+        const m = line.match(dq) ?? line.match(sq);
         if (m) {
-            // Trim any trailing `<space>#<rest-of-line>` that the lazy
-            // capture could not exclude.
-            const cleaned = m[1].replace(/\s+#.*$/, '');
-            out.push(cleaned);
+            out.push(m[1]);
+            continue;
+        }
+        const b = line.match(bare);
+        if (b) {
+            out.push(b[1].trimEnd());
         }
     }
     return out;

scripts/cdp-bridge/dist/domain/reusable-action.js

@@ -102,10 +102,12 @@ export function appendRepairRecord(state, record) {
  * Check whether a self-repair attempt is within the rolling-24h budget.
  * Pure function — `now` is injectable for tests.
  */
-export function repairBudgetAvailable(state, now = () => new Date()) {
+export function recentRepairCount(state, now = () => new Date()) {
     const cutoff = now().getTime() - 24 * 60 * 60 * 1000;
-    const recent = state.repairHistory.filter((r) => new Date(r.timestamp).getTime() >= cutoff);
-    return recent.length < REPAIR_BUDGET.ATTEMPTS_PER_24H;
+    return state.repairHistory.filter((r) => new Date(r.timestamp).getTime() >= cutoff).length;
+}
+export function repairBudgetAvailable(state, now = () => new Date()) {
+    return recentRepairCount(state, now) < REPAIR_BUDGET.ATTEMPTS_PER_24H;
 }
 /**
  * Promote `experimental → active` after a clean replay.

scripts/cdp-bridge/dist/domain/sidecar-io.js

@@ -46,6 +46,18 @@ export function loadOrInitSidecar(yamlFilePath, now = () => new Date()) {
                 Array.isArray(parsed.runHistory) &&
                 Array.isArray(parsed.repairHistory) &&
                 typeof parsed.stats === 'object') {
+                // A sidecar missing lastSeenMtimeMs (e.g. written before the field
+                // existed) would silently disable the human-edit guard, since
+                // yamlEditedSinceLastSeen compares against undefined. Re-seed just that
+                // field from the YAML's mtime — preserves run/repair history.
+                if (typeof parsed.lastSeenMtimeMs !== 'number') {
+                    try {
+                        parsed.lastSeenMtimeMs = statSync(yamlFilePath).mtimeMs;
+                    }
+                    catch {
+                        parsed.lastSeenMtimeMs = 0;
+                    }
+                }
                 return parsed;
             }
             // Fall through — corrupted; return fresh.

scripts/cdp-bridge/dist/experience/compact.js

@@ -80,11 +80,11 @@ export function groupFailures(events) {
         if (isGhostRecovery) {
             stats.ghost_recovered++;
         }
-        else if (event.result === 'FAIL' || event.result === 'ERROR') {
-            stats.failed++;
-        }
         else {
-            stats.passed++;
+            // The filter above only admits fails/ghost-recoveries, so a non-ghost
+            // event here is always FAIL/ERROR — the prior `else { passed++ }` branch
+            // was unreachable (stats.passed stayed 0 forever).
+            stats.failed++;
         }
         if (event.family_id && !stats.family_id) {
             stats.family_id = event.family_id;
@@ -96,19 +96,27 @@ export function groupFailures(events) {
  * Generate candidate heuristics from failure stats.
  * Only generates for patterns with >= MIN_OCCURRENCES and >= MIN_SUCCESS_RATE.
  */
+// Files are written as `candidate-<id>.md` where id is RS-C<n>/FP-C<n>
+// (writeCandidateFile lowercases it → `candidate-rs-c1.md`). The prior
+// `/candidate-(\d+)/` never matched (a letter always follows the dash), so the
+// counter stayed 1 and each compaction cycle clobbered earlier candidate files.
+export function nextCandidateId(existingFilenames) {
+    let nextId = 1;
+    for (const f of existingFilenames) {
+        const match = f.match(/candidate-(?:rs|fp)-c(\d+)/i);
+        if (match)
+            nextId = Math.max(nextId, parseInt(match[1], 10) + 1);
+    }
+    return nextId;
+}
 export function generateCandidates(stats, events) {
     const candidates = [];
     let nextId = 1;
     // Find the highest existing candidate ID
     const candidatesDir = join(AGENT_DIR, 'candidates');
     if (existsSync(candidatesDir)) {
         try {
-            const existing = readdirSync(candidatesDir).filter(f => f.startsWith('candidate-'));
-            for (const f of existing) {
-                const match = f.match(/candidate-(\d+)/);
-                if (match)
-                    nextId = Math.max(nextId, parseInt(match[1], 10) + 1);
-            }
+            nextId = nextCandidateId(readdirSync(candidatesDir).filter(f => f.startsWith('candidate-')));
         }
         catch { /* best-effort */ }
     }