tickmarkr 1.56.0 → 1.57.0

package/dist/adapters/codex.d.ts

@@ -5,4 +5,6 @@ export declare function readCodexModelsCache(path?: string): {
 };
 export declare function seedCodexTrust(repoRoot: string, configPath?: string): TrustVerdict;
 export declare function hasCodexTrustedProject(text: string, root: string): boolean;
+export declare function codexConfigMcpServerNames(configPath?: string): string[];
+export declare function codexMcpSuppressionFlags(configPath?: string): string;
 export declare const codex: WorkerAdapter;

package/dist/adapters/codex.js

@@ -123,6 +123,44 @@ export function hasCodexTrustedProject(text, root) {
     const re = new RegExp(String.raw `\[projects\.(?:"${esc}"|'${esc}')\][^\[]*?trust_level\s*=\s*"trusted"`, "s");
     return re.test(text);
 }
+// v1.57 T1 / OBS-82: first key segment after `mcp_servers.` — bare, "basic", or 'literal' TOML
+// keys; sub-tables ([mcp_servers.x.env]) dedupe to their server name. Fails OPEN to [] on a
+// missing/unreadable config (fresh install has none — base flags must still work).
+export function codexConfigMcpServerNames(configPath) {
+    const p = configPath ?? join(process.env.CODEX_HOME || join(homedir(), ".codex"), "config.toml");
+    let text;
+    try {
+        text = readFileSync(p, "utf8");
+    }
+    catch {
+        return [];
+    }
+    const names = new Set();
+    for (const m of text.matchAll(/^[ \t]*\[mcp_servers\.(?:"([^"]+)"|'([^']+)'|([A-Za-z0-9_-]+))/gm)) {
+        names.add((m[1] ?? m[2] ?? m[3]));
+    }
+    return [...names];
+}
+// OBS-24 → OBS-82: -c 'mcp_servers={}' was codex's analog of claude's --strict-mcp-config, but
+// codex ≥0.144 MERGES the empty inline table with config instead of replacing it (live probe
+// 2026-07-18, codex-cli 0.144.5: `mcp list` identical with and without the override) — so a down
+// operator-global MCP server wedges startup indefinitely again (OBS-82: 45m spinner). No global
+// MCP kill switch exists (`codex features list` has nothing MCP-shaped), and plugin-bundled
+// servers (~/.codex/plugins/cache) never appear under [mcp_servers.*], so suppression is
+// two-pronged: --disable plugins kills plugin loading (incl. the OBS-82 sites-design-picker),
+// per-name enabled=false overrides kill every server named in $CODEX_HOME/config.toml. The empty
+// table stays for older codex (replace semantics there; harmless no-op under merge). The LIVE
+// test in real-adapters.test.ts runs this exact builder against the real `codex mcp list`
+// surface and asserts zero enabled servers — executable proof, not a comment claim. Scanned
+// names reach the shell line ONLY through shq — config values flow into shells.
+export function codexMcpSuppressionFlags(configPath) {
+    const flags = ["--disable plugins", `-c 'mcp_servers={}'`];
+    for (const name of codexConfigMcpServerNames(configPath)) {
+        const key = /^[A-Za-z0-9_-]+$/.test(name) ? name : JSON.stringify(name);
+        flags.push(`-c ${shq(`mcp_servers.${key}.enabled=false`)}`);
+    }
+    return flags.join(" ");
+}
 export const codex = {
     id: "codex",
     vendor: "openai",
@@ -132,14 +170,12 @@ export const codex = {
     probe: async () => probeVersion("codex"),
     channels: (cfg) => channelsFromConfig("codex", cfg),
     // --sandbox workspace-write is the autonomous sandbox mode (codex v0.144.1+)
-    // OBS-24: -c 'mcp_servers={}' is codex's analog of claude's --strict-mcp-config — operator-global
-    // MCP servers in ~/.codex/config.toml block startup indefinitely when one is down (live-measured
-    // 2026-07-15: probe wedged >120s with MCP, 30s clean), wedging probes and worktree workers alike.
-    headlessCommand: (promptFile, model) => `codex exec --sandbox workspace-write -c 'mcp_servers={}' ${GITDIR_WRITABLE} --model ${shq(model)} "$(cat ${shq(promptFile)})"`,
+    // MCP suppression built per dispatch (config can change between runs) — see codexMcpSuppressionFlags.
+    headlessCommand: (promptFile, model) => `codex exec --sandbox workspace-write ${codexMcpSuppressionFlags()} ${GITDIR_WRITABLE} --model ${shq(model)} "$(cat ${shq(promptFile)})"`,
     // TUI uses expanded -a never -s workspace-write (exec-only flags do not apply)
     // (--help 2026-07-09: valid approval policies are untrusted|on-request|never; the previously
     // used `on-failure` is invalid and made codex exit 2 pre-inference)
-    interactiveCommand: (promptFile, model) => `codex -a never -s workspace-write -c 'mcp_servers={}' ${GITDIR_WRITABLE} --model ${shq(model)} "$(cat ${shq(promptFile)})"`,
+    interactiveCommand: (promptFile, model) => `codex -a never -s workspace-write ${codexMcpSuppressionFlags()} ${GITDIR_WRITABLE} --model ${shq(model)} "$(cat ${shq(promptFile)})"`,
     invoke(task, _cwd, a, ctx) {
         return { command: this.headlessCommand(ctx.promptFile, a.model) };
     },

package/dist/run/daemon.js

@@ -21,6 +21,7 @@ import { acquireRunLock, releaseRunLock } from "./lock.js";
 import { ensureIntegration, integrationBranch, integrationHead, mergeTask, verifyIntegrationTip } from "./merge.js";
 import { nextChannel, QUALITY_ENV, route } from "../route/router.js";
 import { desiredPanes } from "./reconcile.js";
+import { normalizeStallSnapshot } from "./stall.js";
 const MODE_RANK = { "staff-led": 0, "risk-based": 1, "partner-led": 2 };
 // An override (flag/spec) re-resolves through loadConfigWithMode itself, via a synthesized repo overlay
 // carrying routing.mode — floors, explore, lints, and provenance all come from config.ts's preset
@@ -630,7 +631,10 @@ export async function runDaemon(repoRoot, opts = {}) {
                     // OBS-54: reaping keys on new pane output, not dispatch wall clock. Poll at least twice per
                     // stall window (and at the existing 30s cadence for normal windows) so an active worker resets it.
                     const stallWindowMs = taskTimeoutMinutes * 60_000;
-                    let lastOutput = output;
+                    // OBS-82: the stall clock compares NORMALIZED snapshots so a spinner glyph/elapsed-time
+                    // repaint is silence, not activity. ONLY this inactivity compare sees normalized text —
+                    // trailer detection, harvest, paging, and quota checks all read the raw pane.
+                    let lastStallSnapshot = normalizeStallSnapshot(output);
                     let lastOutputAt = Date.now();
                     while (Date.now() - lastOutputAt < stallWindowMs) {
                         const sliceStart = Date.now();
@@ -649,9 +653,9 @@ export async function runDaemon(repoRoot, opts = {}) {
                                 break;
                             }
                         }
-                        const currentOutput = await driver.read(slot, 1000);
-                        if (currentOutput !== lastOutput) {
-                            lastOutput = currentOutput;
+                        const currentStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 1000));
+                        if (currentStallSnapshot !== lastStallSnapshot) {
+                            lastStallSnapshot = currentStallSnapshot;
                             lastOutputAt = Date.now();
                         }
                         // v1.23 T2: piggyback on this poll slice — same cadence as blocked/idle checks, no new timer.
@@ -709,8 +713,10 @@ export async function runDaemon(repoRoot, opts = {}) {
                     // T5: same brand header as the interactive path and the gate/consult dispatch scripts.
                     await driver.run(slot, `${bannerShell()}; ${inv.command}; ${exitMarkerCmd}`);
                     // OBS-54: headless workers have the same output-inactivity budget as visible panes.
+                    // OBS-82: same normalized-snapshot compare as the interactive site — spinner-only repaints
+                    // exhaust the budget here too; harvest below still reads the raw pane.
                     const stallWindowMs = taskTimeoutMinutes * 60_000;
-                    let lastOutput = await driver.read(slot, 500);
+                    let lastStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 500));
                     let lastOutputAt = Date.now();
                     finished = false;
                     while (Date.now() - lastOutputAt < stallWindowMs) {
@@ -720,9 +726,9 @@ export async function runDaemon(repoRoot, opts = {}) {
                             finished = true;
                             break;
                         }
-                        const currentOutput = await driver.read(slot, 500);
-                        if (currentOutput !== lastOutput) {
-                            lastOutput = currentOutput;
+                        const currentStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 500));
+                        if (currentStallSnapshot !== lastStallSnapshot) {
+                            lastStallSnapshot = currentStallSnapshot;
                             lastOutputAt = Date.now();
                         }
                     }

package/dist/run/stall.d.ts

@@ -0,0 +1,4 @@
+/** Normalize one pane snapshot for the stall-inactivity compare (and ONLY that compare — trailer
+ * parsing, harvest, waitOutput, and paging read the raw text). Two snapshots that normalize equal
+ * are the same frame modulo spinner presentation; any other byte difference is worker activity. */
+export declare function normalizeStallSnapshot(text: string): string;

package/dist/run/stall.js

@@ -0,0 +1,24 @@
+// OBS-82: codex's MCP-startup spinner repaints a braille glyph + elapsed-time cell forever, so the
+// daemon's raw snapshot compare reads a wedged pane as active and the stall clock never fires.
+// This normalizer deletes ONLY presentation tokens from a closed allowlist — ANSI/VT escape
+// sequences, braille-range spinner glyphs, and elapsed-time tokens bound to time-unit suffixes.
+// Every other byte passes through identical: words, paths, server names, and progress counts
+// (a five-of-seven counter change IS activity) all remain change-sensitive. The asymmetry is the
+// design: an allowlist MISS degrades to today's recoverable no-reap behavior, while an over-broad
+// deletion would reap a healthy worker — a new failure class. Grow the allowlist only with
+// captured evidence (tests/fixtures/codex-mcp-spinner/).
+// CSI (with intermediates), OSC (BEL- or ST-terminated), DCS/SOS/PM/APC strings, single-char
+// escapes, and charset selection — the raw-pty forms; herdr pane reads are already rendered.
+// eslint-disable-next-line no-control-regex
+const ANSI_RE = /\x1b\[[0-9;?]*[ -/]*[@-~]|\x9b[0-9;?]*[ -/]*[@-~]|\x1b\][^\x07\x1b]*(?:\x07|\x1b\\)?|\x1b[PX^_][^\x1b]*(?:\x1b\\)?|\x1b[()][0-9A-Za-z]|\x1b[0-~]/g;
+// Braille patterns U+2800–U+28FF — the codex spinner cell (captured fixture: ⠋⠙⠸⠴⠦⠇ …).
+const SPINNER_RE = /[⠀-⣿]/g;
+// A digit run (optionally decimal) bound directly to a time-unit suffix, standing alone as a
+// word: 9s, 41s, 3m, 1h, 800ms. Never bare digits — "(6/7)" and "5 of 7" stay change-sensitive.
+const ELAPSED_RE = /(?<![\w.])\d+(?:\.\d+)?(?:ms|[hms])(?!\w)/g;
+/** Normalize one pane snapshot for the stall-inactivity compare (and ONLY that compare — trailer
+ * parsing, harvest, waitOutput, and paging read the raw text). Two snapshots that normalize equal
+ * are the same frame modulo spinner presentation; any other byte difference is worker activity. */
+export function normalizeStallSnapshot(text) {
+    return text.replace(ANSI_RE, "").replace(SPINNER_RE, "").replace(ELAPSED_RE, "");
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tickmarkr",
-  "version": "1.56.0",
+  "version": "1.57.0",
   "description": "Spec in, verified work out.",
   "type": "module",
   "license": "MIT",