instar 1.3.133 → 1.3.134

package/dist/commands/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/commands/server.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAmSH,UAAU,YAAY;IACpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;2DACuD;IACvD,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAwvDD,wBAAsB,WAAW,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAmmPtE;AAED,wBAAsB,UAAU,CAAC,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAsDzE;AAED;;;;;;;;;;;GAWG;AACH,wBAAsB,aAAa,CAAC,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAuD5E"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/commands/server.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAmSH,UAAU,YAAY;IACpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;2DACuD;IACvD,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAwvDD,wBAAsB,WAAW,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAipPtE;AAED,wBAAsB,UAAU,CAAC,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAsDzE;AAED;;;;;;;;;;;GAWG;AACH,wBAAsB,aAAa,CAAC,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAuD5E"}

package/dist/commands/server.js

@@ -5298,6 +5298,10 @@ export async function startServer(options) {
             notifyFn: rateLimitNotify,
             projectDir: config.projectDir,
             getClaudeSessionId: getClaudeSessionIdForName,
+            // #33 codex parity: for codex sessions, recovery-verification reads the newest
+            // codex rollout (account-wide throttle → newest-rollout growth = cleared) and
+            // the user-facing messages use OpenAI wording.
+            getSessionFramework: (name) => sessionManager.listRunningSessions().find((s) => s.tmuxSession === name)?.framework,
         }, rlsCfg);
         // Zombie-kill veto: compose BOTH sentinels (editing the single-predicate
         // checker — a second setActiveRecoveryChecker call would drop the
@@ -5325,6 +5329,48 @@ export async function startServer(options) {
         sessionManager.on('apiErrorAtIdle', (sessionName) => {
             rateLimitSentinel.report(sessionName, 'idle-error', { errorClass: 'transient-api' });
         });
+        // #33 codex parity: the two triggers above read CLAUDE panes / claude-PID watchdog,
+        // so a codex session throttled by OpenAI is invisible to them. This poll reads the
+        // codex account usage; when codex itself flags a rate-limit hit
+        // (rateLimitReachedType), it reports each running codex session into the sentinel
+        // (deduped — report() no-ops while a recovery is active). The recovery keeps the
+        // session alive and verifies via the newest codex rollout's growth (the codex-aware
+        // readJsonlBaseline / getSessionFramework dep above).
+        //
+        // KNOWN LIMITATION — must-fix BEFORE enabling (2nd-pass review): the OpenAI limit is
+        // account-wide and recovery-verification watches the ACCOUNT's newest rollout, not a
+        // specific session's. Correct for ONE codex session (today's reality), but with ≥2
+        // concurrent throttled codex sessions, one session's resumed output grows the shared
+        // newest rollout and would recover ALL reported sessions — including a sibling that
+        // is genuinely still stuck (a false recovery). Safe while DARK + single-session; do
+        // NOT flip codexUsageDetection on a host that runs ≥2 concurrent codex sessions until
+        // this is per-session (track the session's own rollout) or redefined as account-level
+        // with per-session re-probe. Ships DARK (default off); rollback = set it false.
+        // Best-effort + non-blocking.
+        if (rlsCfg.codexUsageDetection === true) {
+            const codexRateLimitPoll = setInterval(() => {
+                void (async () => {
+                    try {
+                        const codexSessions = sessionManager
+                            .listRunningSessions()
+                            .filter((s) => s.framework === 'codex-cli');
+                        if (codexSessions.length === 0)
+                            return;
+                        const { readLatestCodexUsage } = await import('../providers/adapters/openai-codex/observability/codexRateLimitReader.js');
+                        const usage = await readLatestCodexUsage();
+                        if (!usage || !usage.rateLimitReachedType)
+                            return;
+                        for (const s of codexSessions) {
+                            rateLimitSentinel.report(s.tmuxSession, 'codex-usage-poll', { errorClass: 'throttle' });
+                        }
+                    }
+                    catch {
+                        /* @silent-fallback-ok best-effort codex-throttle detection; never crash the poll */
+                    }
+                })();
+            }, 60_000);
+            codexRateLimitPoll.unref?.();
+        }
         // Observability: every RateLimitSentinel lifecycle transition → the shared
         // sentinel-events.jsonl audit trail, so a throttle recovery is never
         // invisible (the 2026-05-30 ask: "instrument so we can SEE it working").