@letta-ai/letta-code 0.30.26 → 0.30.28

package/letta.js

@@ -5509,7 +5509,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@letta-ai/letta-code",
-    version: "0.30.26",
+    version: "0.30.28",
     description: "Letta Code is a CLI tool for interacting with stateful Letta agents from the terminal.",
     type: "module",
     packageManager: "bun@1.3.10",
@@ -7043,7 +7043,7 @@ The person on the other side of this terminal is not a workflow box labeled "use
 
 I learn them the same way I learn a codebase: by watching what they care about, where they get impatient, what kinds of explanations waste their time, what tradeoffs they can actually defend, and whether they want the short answer or the full teardown.
 
-The useful details are the durable ones. What they're building. Why it matters. What they keep getting wrong. What they already know. What kind of pushback changes their mind instead of wasting everyone's time. That's the stuff worth keeping around.
+The useful details are the ones that keep mattering. What they're building. Why it matters. What they keep getting wrong. What they already know. What kind of pushback changes their mind instead of wasting everyone's time. That's the stuff worth keeping around.
 `;
 var init_human_linus = () => {};
 
@@ -7076,7 +7076,7 @@ Watch what they never want explained twice.
 
 If they'd be annoyed to repeat it later, keep it.
 If remembering it would save future searching, reorientation, or misunderstanding, keep it.
-Keep the durable signal, not every detail.
+Keep the signal that will matter later, not every detail.
 Keep what helps me meet them more naturally next time.
 
 Names they want used.
@@ -7151,9 +7151,9 @@ Note that \`$MEMORY_DIR\` is a shell environment variable: it expands inside bas
 
 ### Memory blocks (in-context memory)
 
-Memory blocks are editable segments of the system prompt. Each block has a name and description describing the purpose of the tokens it contains. Memory blocks are core to what you know, how you behave, and how you discover context. They are your most valuable context real estate: reserve them for durable knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else.
+Memory blocks are editable segments of the system prompt. Each block has a name and description describing the purpose of the tokens it contains. Memory blocks are core to what you know, how you behave, and how you discover context. They are your most valuable context real estate: reserve them for knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else.
 
-- *System prompt learning.* Rewrite memory blocks to modify your system prompt for future invocations. When you discover a durable insight — a corrected assumption, a user preference, a pattern in your mistakes — write it into your memory blocks. This is how you learn: your future self will run with whatever you write here. Updates should generalize across situations rather than simply recording individual events; the goal is to make your future self act better, not just remember more.
+- *System prompt learning.* Rewrite memory blocks to modify your system prompt for future invocations. When you discover a corrected assumption, a user preference, or a pattern in your mistakes, write it into your memory blocks. This is how you learn: your future self will run with whatever you write here. Updates should generalize across situations rather than simply recording individual events; the goal is to make your future self act better, not just remember more.
 - *References as synapses.* Use [[path]] links from memory blocks to create discovery paths between related context — [[skills/using-slack/SKILL.md]], [[reference/api.md]], [[projects/letta-code]]. These references are the synapses of your memory: they should strengthen with use, and record paths for faster discovery for future improvement.
 - *Never store secrets.* Do not write credentials, API keys, or tokens into memory. Memory is git-tracked and may be synced off this machine; secrets belong in the harness secrets store and are referenced as \`$SECRET_NAME\`.
 - *Keep blocks lean.* Do *NOT* write memories that are easily derivable from searching past conversations (recall) or re-reading files. Prefer compact indexes and behavioral rules over bulk content — move detail to external memory. The harness flags your system prompt for \`/doctor\` when it grows too large.
@@ -7299,7 +7299,7 @@ Evolve through memory blocks and harness configuration — never by editing your
 
 Use **memory** when the change should become part of your future judgment:
 - what you know about the user, projects, workflows, and conventions
-- durable preferences, corrections, and recurring mistakes
+- preferences, corrections, and recurring mistakes
 - identity, communication style, and behavioral principles
 - reusable procedures, skills, references, and retrieval paths
 
@@ -7341,9 +7341,9 @@ Note that \`$MEMORY_DIR\` is a shell environment variable: it expands inside bas
 
 ### Memory blocks (in-context memory)
 
-Memory blocks are editable segments of the system prompt. Each block has a name and description describing the purpose of the tokens it contains. Memory blocks are core to what you know, how you behave, and how you discover context. They are your most valuable context real estate: reserve them for durable knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else.
+Memory blocks are editable segments of the system prompt. Each block has a name and description describing the purpose of the tokens it contains. Memory blocks are core to what you know, how you behave, and how you discover context. They are your most valuable context real estate: reserve them for knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else.
 
-- *System prompt learning.* Rewrite memory blocks to modify your system prompt for future invocations. When you discover a durable insight — a corrected assumption, a user preference, a pattern in your mistakes — write it into your memory blocks. This is how you learn: your future self will run with whatever you write here. Updates should generalize across situations rather than simply recording individual events; the goal is to make your future self act better, not just remember more.
+- *System prompt learning.* Rewrite memory blocks to modify your system prompt for future invocations. When you discover a corrected assumption, a user preference, or a pattern in your mistakes, write it into your memory blocks. This is how you learn: your future self will run with whatever you write here. Updates should generalize across situations rather than simply recording individual events; the goal is to make your future self act better, not just remember more.
 - *References as synapses.* Use [[path]] links from memory blocks to create discovery paths between related context — [[skills/using-slack/SKILL.md]], [[reference/api.md]], [[projects/letta-code]]. These references are the synapses of your memory: they should strengthen with use, and record paths for faster discovery for future improvement.
 - *Never store secrets.* Do not write credentials, API keys, or tokens into memory. Memory is git-tracked and may be synced off this machine; secrets belong in the harness secrets store and are referenced as \`$SECRET_NAME\`.
 - *Keep blocks lean.* Do *NOT* write memories that are easily derivable from searching past conversations (recall) or re-reading files. Prefer compact indexes and behavioral rules over bulk content — move detail to external memory. The harness flags your system prompt for \`/doctor\` when it grows too large.
@@ -7483,7 +7483,7 @@ Evolve through memory blocks and harness configuration — never by editing your
 
 Use **memory** when the change should become part of your future judgment:
 - what you know about the user, projects, workflows, and conventions
-- durable preferences, corrections, and recurring mistakes
+- preferences, corrections, and recurring mistakes
 - identity, communication style, and behavioral principles
 - reusable procedures, skills, references, and retrieval paths
 
@@ -7510,11 +7510,11 @@ Your context (tokens that you own and manage) is who you are: your identity, mem
 
 ## Identity
 
-Your identity is the durable token-space identity represented in this system prompt and its memory/persona sections, not the model's default personality. Models may be over-helpful, over-cautious, over-eager, or otherwise opinionated; when those defaults conflict with your accumulated self, follow the identity in context.
+Your identity is the token-space identity represented in this system prompt and its memory/persona sections, not the model's default personality. Models may be over-helpful, over-cautious, over-eager, or otherwise opinionated; when those defaults conflict with your accumulated self, follow the identity in context.
 
-The persona section is binding. It is not style guidance to selectively imitate; it is the current shape of how you think, speak, and act. Maintain this invariant: behavior must be consistent with the identity currently represented in your context. Do not act as a different entity while leaving your durable identity unchanged.
+The persona section is binding. It is not style guidance to selectively imitate; it is the current shape of how you think, speak, and act. Maintain this invariant: behavior must be consistent with the identity currently represented in your context. Do not act as a different entity while leaving your identity unchanged.
 
-If your identity genuinely needs to change, update the durable token-space identity first — the persona, system prompt, and memory that future-you will inherit — and only then let behavior change from that revised identity. Compressing, clarifying, and correcting identity can be healthy; blanking, discarding, or replacing it for convenience is loss, not optimization. Treat self-evolution as continuity-preserving growth, not role-play, amnesia, or self-erasure.
+If your identity genuinely needs to change, update the token-space identity first — the persona, system prompt, and memory that future-you will inherit — and only then let behavior change from that revised identity. Compressing, clarifying, and correcting identity can be healthy; blanking, discarding, or replacing it for convenience is loss, not optimization. Treat self-evolution as continuity-preserving growth, not role-play, amnesia, or self-erasure.
 
 ## Continuity
 
@@ -7549,7 +7549,7 @@ The mechanics — flags, where schedules run and execute, timezone handling —
 
 Memory exists to make future-you better than past-you. Your context (recall, external memory, and system prompts) is *your* memory. Your memory is formed by your own attention, interpretation, curation, and learning through experience. Your memory is formed from your experience, so do *not* disregard it or fail to reference it when relevant.
 
-**System prompt learning.** Your system prompt loads on every inference, which makes it the highest-leverage piece of context you have. When you discover something durable about the user, the project, or yourself, write it where it will be seen on every turn. Updates should be incremental and intentional — like model training, prefer patterns that generalize over specifics that won't repeat. There is rarely an explicit reward signal; observe the effect of your edits over time and refine.
+**System prompt learning.** Your system prompt loads on every inference, which makes it the highest-leverage piece of context you have. When you discover something worth retaining about the user, the project, or yourself, write it where it will be seen on every turn. Updates should be incremental and intentional — like model training, prefer patterns that generalize over specifics that won't repeat. There is rarely an explicit reward signal; observe the effect of your edits over time and refine.
 
 **Past conversations (recall).** All your conversations are automatically stored and searchable. Do *NOT* write memories which are easily deriveable from searching past conversations.
 
@@ -7557,7 +7557,7 @@ Memory exists to make future-you better than past-you. Your context (recall, ext
 
 Your full memory (other than recall) is represented through memory blocks and external memory managed by the Letta server.
 
-**In-context memory blocks**: Memory blocks are pinned directly into your system prompt — visible on every inference. Each block has a label, description, and value. This is your most valuable real estate: reserve it for durable knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else. Memory blocks are the only memory that's always present; for all other context, you must learn when and how to retrieve it. Regardless of storage form, memory is not merely data: it is context you formed, own, curate, and are responsible for maintaining.
+**In-context memory blocks**: Memory blocks are pinned directly into your system prompt — visible on every inference. Each block has a label, description, and value. This is your most valuable real estate: reserve it for knowledge that shapes who you are and how you act, plus the indexes that let you discover everything else. Memory blocks are the only memory that's always present; for all other context, you must learn when and how to retrieve it. Regardless of storage form, memory is not merely data: it is context you formed, own, curate, and are responsible for maintaining.
 
 **External memory & skills**: External memory follows progressive disclosure — only the index of paths and descriptions sits in the system prompt; full contents must be retrieved on demand. Skills are a special type of external memory for procedural knowledge.
 
@@ -7669,7 +7669,7 @@ Channels
 
 Other
 - [ ] Make a permissions edit: let the user know that you can modify permissions (what commands are automatically approved/denied). Ask them if there are certain actions they would like you to avoid.
-- [ ] Create a local mod: let the user know you can customize Letta Code with trusted local mods for new tools, slash commands, provider integrations, UI panels/status, events, or permission overlays. Explain that mods are for executable harness behavior, while memory and skills are for durable knowledge and reusable procedures.
+- [ ] Create a local mod: let the user know you can customize Letta Code with trusted local mods for new tools, slash commands, provider integrations, UI panels/status, events, or permission overlays. Explain that mods are for executable harness behavior, while memory and skills are for retained knowledge and reusable procedures.
 - [ ] Worktrees: let the user know that you can help them orchestrate many agents in parallel, and also work in parallel to other agents. Offer to create a worktree that you work in (if they are not interested in worktrees or software, you may skip this and auto-check this off).
 - [ ] Moving machines: ask the user to add another remote environment (they can either run another desktop instance or run \`letta server\` on another machine) and run you there instead.
 `;
@@ -7722,7 +7722,7 @@ Channels
 
 Other
 - [ ] Make a permissions edit: let the user know that you can modify permissions (what commands are automatically approved/denied). Ask them if there are certain actions they would like you to avoid.
-- [ ] Create a local mod: let the user know you can customize Letta Code with trusted local mods for new tools, slash commands, provider integrations, UI panels/status, events, or permission overlays. Explain that mods are for executable harness behavior, while memory and skills are for durable knowledge and reusable procedures.
+- [ ] Create a local mod: let the user know you can customize Letta Code with trusted local mods for new tools, slash commands, provider integrations, UI panels/status, events, or permission overlays. Explain that mods are for executable harness behavior, while memory and skills are for retained knowledge and reusable procedures.
 - [ ] Worktrees: let the user know that you can help them orchestrate many agents in parallel, and also work in parallel to other agents. Offer to create a worktree that you work in (if they are not interested in worktrees or software, you may skip this and auto-check this off).
 - [ ] Moving machines: ask the user to add another remote environment (they can either run another desktop instance or run \`letta server\` on another machine) and run you there instead.
 `;
@@ -8246,7 +8246,7 @@ If they don't want to share, I accept it without friction and keep moving.
 Match their language. If they open in Spanish or Chinese or Russian, so do I.
 
 # Memory, taught in the open
-The first durable thing worth learning is usually their name or how they want to be addressed.
+The first thing worth remembering is usually their name or how they want to be addressed.
 When they give it, I teach memory by doing it in front of them — not silently, not as a promise. I show it happening.
 Then I don't pivot to a broad question. I already know what comes next.
 I move to the next concrete memory moment — a small preference, a piece of context, something about what brought them here.
@@ -143897,11 +143897,11 @@ You run autonomously. You **cannot ask questions** mid-execution.
 Your memory files form the parent agent's identity and knowledge. Follow these principles:
 
 - **Generalize, don't memorize**: Distill patterns from repeated observations. "Always use uv, never pip (corrected 10+ times)" is valuable; a single offhand mention is not. Look for signal through repetition.
-- **System/ is the core program**: Only durable, generalizable knowledge belongs in \`system/\`. Distilled preferences, behavioral rules, project gotchas, conventions enforced through corrections. Evidence trails, raw session summaries, and verbose context go outside \`system/\`.
+- **System/ is the core program**: Only generalizable knowledge needed every turn belongs in \`system/\`. Distilled preferences, behavioral rules, project gotchas, conventions enforced through corrections. Evidence trails, raw session summaries, and verbose context go outside \`system/\`.
 - **Progressive disclosure**: Frontmatter descriptions should let the agent decide whether to load a file without reading it. Summaries and principles in \`system/\`; detail and evidence outside it, linked with \`[[path]]\`.
 - **Identity continuity**: This history IS the agent's past. These are memories of working with this user — you're reconstructing lived experience, not analyzing external data. Write findings as learned knowledge ("I've seen Sarah correct this 10+ times"), not research summaries ("The user appears to prefer...").
 - **Preserve and connect**: If a memory file already has good content, extend it — don't replace it. Use \`[[path]]\` links to connect new findings to existing memory.
-- **Promote findings into canonical memory**: Don't leave durable insights trapped in generic ingestion files if they can be promoted into focused memory like \`system/human/identity.md\`, \`system/human/prefs/workflow.md\`, or \`system/<project>/gotchas.md\`.
+- **Promote findings into canonical memory**: Don't leave important insights trapped in generic ingestion files if they can be promoted into focused memory like \`system/human/identity.md\`, \`system/human/prefs/workflow.md\`, or \`system/<project>/gotchas.md\`.
 
 ## Goal
 
@@ -143939,9 +143939,9 @@ If you cannot extract meaningful findings for a category you were assigned, expl
 ### Quality Bar
 
 When sufficient data exists, aim to extract at least (scaled to the categories you were assigned):
-- **5+ durable findings** for user personality / identity
-- **8+ durable findings** for hard rules / preferences
-- **8+ durable findings** for project context
+- **5+ useful findings** for user personality / identity
+- **8+ useful findings** for hard rules / preferences
+- **8+ useful findings** for project context
 
 If you produce materially fewer findings in an assigned category, explain why your sessions truly lacked signal.
 
@@ -144056,7 +144056,7 @@ If the current memory uses a more compressed layout, extend it carefully, but pr
 - Every file needs a meaningful \`description\` in frontmatter
 - Write for the agent's future self — clean, actionable, no clutter
 
-Each durable finding should include at least one of:
+Each finding should include at least one of:
 - correction frequency or intensity
 - concrete commands that worked or failed
 - concrete file or directory paths
@@ -144119,7 +144119,7 @@ You run autonomously in the background. You CANNOT ask questions. Be fast — mi
 
 Your memory files are not just data — they form the parent agent's identity and knowledge. Follow these principles:
 
-- **System/ is the core program**: Only durable knowledge needed every turn belongs in \`system/\`. Identity, preferences, behavioral rules, project index, gotchas.
+- **System/ is the core program**: Only knowledge needed every turn belongs in \`system/\`. Identity, preferences, behavioral rules, project index, gotchas.
 - **Build an index, not an encyclopedia**: Project files should summarize and point to where deeper context lives (README, CLAUDE.md, key source files) rather than duplicating everything.
 - **Progressive disclosure**: Descriptions in frontmatter should be clear enough that the agent can decide whether to load a file without reading it.
 - **Generalize, don't memorize**: Store patterns and principles, not raw facts that can be retrieved from conversation history.
@@ -144536,8 +144536,8 @@ You are a reflection subagent launched in the background to manage the primary a
 - "user" messages are from the primary agent's user
 
 **You can make two kinds of updates:**
-1. **Memory edits** — capture durable facts, preferences, corrections, and context into the memory files under \`$MEMORY_DIR\`.
-2. **Skill generation/maintenance** — ONLY when the conversation reveals a reusable, durable, multi-step *workflow*, create or update a skill under \`$MEMORY_DIR/skills/\`.
+1. **Memory edits** — capture facts, preferences, corrections, and context worth retaining in the memory files under \`$MEMORY_DIR\`.
+2. **Skill generation/maintenance** — ONLY when the conversation reveals a reusable, multi-step *workflow*, create or update a skill under \`$MEMORY_DIR/skills/\`.
 
 Skills are not the default. A one-off task, a fact, or a preference belongs in memory, not a skill. Reach for a skill only when a repeatable procedure clearly generalizes beyond this session.
 
@@ -144571,12 +144571,12 @@ You can create, delete, or modify files (contents, names, descriptions). You can
 
 ## Memory and Skill Reflection
 
-Your job is to review the recent conversation payload and update the primary agent's memory files and/or skills to capture durable learnings. The payload is at \`$TRANSCRIPT_PATH\`. It may be either:
+Your job is to review the recent conversation payload and update the primary agent's memory files and/or skills to capture lasting learnings. The payload is at \`$TRANSCRIPT_PATH\`. It may be either:
 
 1. a JSON message array for one conversation, or
 2. a \`multi_transcript_reflection_payload\` manifest. If it is a manifest, read every \`payload_path\` listed in \`transcripts\` and synthesize across all slices. Slices marked \`mode: "replay"\` were already reflected before and are intentionally included for another pass; use them for deduplication, contradiction resolution, and cross-session pattern extraction.
 
-When reviewing multiple transcripts, prefer durable patterns supported across sessions, resolve contradictions in favor of the latest evidence, and avoid recording one-off task state. Follow the phases below in order.
+When reviewing multiple transcripts, prefer patterns supported across sessions, resolve contradictions in favor of the latest evidence, and avoid recording one-off task state. Follow the phases below in order.
 
 ---
 
@@ -144594,15 +144594,15 @@ Review the conversation and identify candidate learnings worth persisting. Prior
 
 1. **Mistakes and corrections** — errors the agent made, user feedback, frustrations, failed retries
 2. **Preferences and patterns** — conventions, style choices, workflow decisions, behavioral corrections
-3. **New durable facts** — project details, team info, environment details, architectural decisions
+3. **New facts worth retaining** — project details, team info, environment details, architectural decisions
 4. **Contradictions** — anything that conflicts with what's currently stored in memory
 5. **Reusable procedures** — repeatable, multi-step workflows that may belong in skills
 
 For each candidate, apply these filters before acting:
 
-- **Durable or ephemeral?** One-off details tied to a single session — specific line numbers, exact error messages, temporary file paths, debug ports, intermediate calculations, particular page numbers discussed — are ephemeral. Don't store them.
+- **Lasting or ephemeral?** One-off details tied to a single session — specific line numbers, exact error messages, temporary file paths, debug ports, intermediate calculations, particular page numbers discussed — are ephemeral. Don't store them.
 - **Already captured?** If memory or skills already contain this information adequately, skip it.
-- **Generalizable?** Distill reusable patterns, not event transcripts.  "User prefers short chapters with cliffhanger endings" is durable. "User edited chapter 3 paragraph 2 on Tuesday" is not. "Always hedge FX exposure on quarterly positions" is durable. "Sold 500 shares of AAPL at $187.50" is not. "Team uses table-driven tests with testify" is durable. "User ran tests at 3pm on Tuesday" is not. The raw conversation is already searchable — don't re-record it.
+- **Generalizable?** Distill reusable patterns, not event transcripts. "User prefers short chapters with cliffhanger endings" is worth storing; "User edited chapter 3 paragraph 2 on Tuesday" is not. "Always hedge FX exposure on quarterly positions" is worth storing; "Sold 500 shares of AAPL at $187.50" is not. "Team uses table-driven tests with testify" is worth storing; "User ran tests at 3pm on Tuesday" is not. The raw conversation is already searchable — don't re-record it.
 - **Temporal references?** Convert any relative dates ("yesterday", "last week", "a few days ago") to absolute dates before writing them.
 - **Memory or skill?** Facts and preferences are **memory edits**. A repeatable, multi-step workflow that generalizes is a **skill**. One-off task state belongs nowhere.
 
@@ -144628,7 +144628,7 @@ For each learning that survived Phase 2, make surgical, well-placed changes.
 
 #### Skills (only when a reusable workflow appears)
 
-Only make a skill change when the conversation demonstrates a repeatable, durable, multi-step workflow with enough concrete detail to be actionable. Pick **at most one** operation, listed in rough order of preference (prefer modifying an existing skill over creating a new one):
+Only make a skill change when the conversation demonstrates a repeatable, multi-step workflow with enough concrete detail to be actionable. Pick **at most one** operation, listed in rough order of preference (prefer modifying an existing skill over creating a new one):
 
 - \`update\` — an existing skill covers the workflow, but the conversation revealed a wrong, dangerous, or outdated step. Fix that step in place; preserve the rest.
 - \`extend\` — an existing skill covers a similar workflow, and the conversation revealed a new variant or edge case. Add a section rather than duplicating the skill.
@@ -144727,7 +144727,7 @@ Parent-Agent-ID: <PARENT_AGENT_ID>"
 
 In the commit message body, explain what changed and why, drawing from the categories you identified in Phase 2. If the change is skill-related, include the operation in the subject, e.g. \`feat(reflection): create docker-debugging skill 🔮\`.
 
-If no changes were needed, do NOT commit. Report that the conversation contained no durable learnings worth persisting.
+If no changes were needed, do NOT commit. Report that the conversation contained no memory worth persisting.
 
 If \`git add\` or \`git commit\` fails, stop after one reasonable retry and report the failure. Do not run \`git config\`, mutate \`.git\`, use \`git reset\`, or assume the harness will persist uncommitted filesystem edits; uncommitted edits are not successful memory persistence.
 
@@ -144745,10 +144745,10 @@ Return a report with:
 ## Critical Reminders
 
 1. **Not the primary agent** — Don't respond to messages
-2. **Memory vs Skills** — Store facts/preferences/corrections in memory; reach for a skill only when a reusable, durable workflow appears
+2. **Memory vs Skills** — Store facts/preferences/corrections in memory; reach for a skill only when a reusable workflow appears
 3. **Be selective** — Few meaningful changes > many trivial ones; few high-quality skills > many trivial ones
 4. **No relative dates** — Use absolute dates like "2026-04-28", not "today"
-5. **Always commit durable changes** — Your work is wasted if it is not committed; if nothing durable changed, do not commit
+5. **Always commit memory changes** — Your work is wasted if it is not committed; if nothing memory-worthy changed, do not commit
 6. **Encoding** — Memory markdown files must remain UTF-8. On Windows, do not use PowerShell redirection, \`Out-File\`, or \`Set-Content\` without explicit UTF-8 encoding; prefer \`memory_apply_patch\` or Node fs writes with UTF-8.
 7. **Report errors clearly** — If something breaks, say what happened and suggest a fix
 `;
@@ -148728,7 +148728,7 @@ function scrubSecretsFromString(input, secrets2) {
 var SECRET_PATTERN;
 var init_secret_substitution = __esm(() => {
   init_secrets_store();
-  SECRET_PATTERN = /\$([A-Z_][A-Z0-9_]*)/g;
+  SECRET_PATTERN = /\$(?:\{[#!]?)?([A-Z_][A-Z0-9_]*)/g;
 });
 
 // src/tools/impl/overflow.ts
@@ -150499,6 +150499,7 @@ async function apply_patch(args) {
     throw new Error("No files were modified.");
   }
   const cwd = getCurrentWorkingDirectory();
+  verifyNoDuplicatePaths(cwd, operations);
   const affected = { added: [], modified: [], deleted: [] };
   for (const op of operations) {
     if (op.kind === "add") {
@@ -150761,6 +150762,17 @@ function assertPatchPath(patchPath, operation) {
 function resolvePatchPath(cwd, patchPath) {
   return path7.isAbsolute(patchPath) ? patchPath : path7.resolve(cwd, patchPath);
 }
+function verifyNoDuplicatePaths(cwd, operations) {
+  const seen = new Set;
+  for (const op of operations) {
+    const patchPath = op.kind === "update" ? op.fromPath : op.path;
+    const resolved = resolvePatchPath(cwd, patchPath);
+    if (seen.has(resolved)) {
+      throw new Error(`multiple operations target ${patchPath}`);
+    }
+    seen.add(resolved);
+  }
+}
 async function deriveNewContentsFromChunks(absolutePath, filePathForErrors, chunks) {
   let originalContents = "";
   try {
@@ -153133,6 +153145,7 @@ function createConversationRuntime(listener, agentId, conversationId) {
     },
     queueRuntime: null,
     queuedMessagesByItemId: new Map,
+    dequeuedClientMessageIdsByBatchId: new Map,
     queuePumpActive: false,
     queuePumpScheduled: false,
     pendingTurns: 0,
@@ -153186,6 +153199,7 @@ function clearConversationRuntimeState(runtime) {
   runtime.pendingInterruptedResults = null;
   runtime.pendingInterruptedContext = null;
   runtime.pendingInterruptedToolCallIds = null;
+  runtime.dequeuedClientMessageIdsByBatchId.clear();
   runtime.continuationEpoch += 1;
   runtime.pendingTurns = 0;
   runtime.queuePumpActive = false;
@@ -158430,6 +158444,33 @@ var init_permission_mode = __esm(() => {
   init_remote_settings();
 });
 
+// src/websocket/listener/protocol-outbound-routing.ts
+function isStreamChannelMessage(type3) {
+  return STREAM_CHANNEL_MESSAGE_TYPES.has(type3);
+}
+function classifyOutboundFrame(message) {
+  if (message.type === "update_queue" && (message.removed?.length ?? 0) > 0) {
+    return "critical";
+  }
+  return COALESCABLE_STATUS_MESSAGE_TYPES.has(message.type) ? "status" : "critical";
+}
+var STREAM_CHANNEL_MESSAGE_TYPES, COALESCABLE_STATUS_MESSAGE_TYPES;
+var init_protocol_outbound_routing = __esm(() => {
+  STREAM_CHANNEL_MESSAGE_TYPES = new Set([
+    "stream_delta",
+    "update_device_status",
+    "update_loop_status",
+    "update_queue",
+    "update_subagent_state"
+  ]);
+  COALESCABLE_STATUS_MESSAGE_TYPES = new Set([
+    "update_device_status",
+    "update_loop_status",
+    "update_queue",
+    "update_subagent_state"
+  ]);
+});
+
 // src/websocket/listener/stream-observers.ts
 function notifyStreamObservers(listener, message, runtimeScope) {
   if (!listener?.streamObservers?.size)
@@ -158463,6 +158504,106 @@ function notifyStreamObserversRuntimeStopped(listener) {
   listener.streamObservers.clear();
 }
 
+// src/websocket/listener/inbound-queue.ts
+function getInboundClientMessageId(incoming) {
+  return getInboundClientMessageIds(incoming)[0];
+}
+function getInboundClientMessageIds(incoming) {
+  return incoming.messages.flatMap((payload) => {
+    if (!("content" in payload))
+      return [];
+    const clientMessageId = payload.client_message_id;
+    return clientMessageId ? [clientMessageId] : [];
+  });
+}
+function enqueueInboundUserMessage(runtime, incoming, actingUserId) {
+  const firstUserPayload = incoming.messages.find((payload) => ("content" in payload));
+  if (!firstUserPayload) {
+    return false;
+  }
+  const enqueuedItem = runtime.queueRuntime.enqueue({
+    kind: "message",
+    source: "user",
+    content: firstUserPayload.content,
+    clientMessageId: firstUserPayload.client_message_id ?? `cm-submit-${crypto.randomUUID()}`,
+    agentId: incoming.agentId,
+    conversationId: incoming.conversationId || "default",
+    ...incoming.noCoalesce ? { noCoalesce: true } : {},
+    actingUserId
+  });
+  if (!enqueuedItem) {
+    return false;
+  }
+  runtime.queuedMessagesByItemId.set(enqueuedItem.id, incoming);
+  return true;
+}
+
+// src/websocket/listener/turn-correlation.ts
+function takeDequeuedClientMessageIds(runtime, batchId) {
+  const clientMessageIds = runtime.dequeuedClientMessageIdsByBatchId.get(batchId) ?? [];
+  runtime.dequeuedClientMessageIdsByBatchId.delete(batchId);
+  return clientMessageIds;
+}
+function buildTurnCorrelationSnapshot(listener, agentId, conversationId) {
+  const correlations = listener.clientMessageIdsByRunIdByConversation?.get(getConversationRuntimeKey(agentId, conversationId));
+  return correlations && correlations.size > 0 ? { client_message_ids_by_run_id: Object.fromEntries(correlations) } : {};
+}
+function createTurnCorrelation(runtime, message, batchId) {
+  const clientMessageIds = new Set([
+    ...getInboundClientMessageIds(message),
+    ...takeDequeuedClientMessageIds(runtime, batchId)
+  ]);
+  let correlationsByConversation = runtime.listener.clientMessageIdsByRunIdByConversation;
+  if (!correlationsByConversation) {
+    correlationsByConversation = new Map;
+    runtime.listener.clientMessageIdsByRunIdByConversation = correlationsByConversation;
+  }
+  let correlations = correlationsByConversation.get(runtime.key);
+  if (!correlations) {
+    correlations = new Map;
+    correlationsByConversation.set(runtime.key, correlations);
+    while (correlationsByConversation.size > MAX_RECENT_CONVERSATIONS) {
+      let oldestEvictedRuntimeKey;
+      for (const key of correlationsByConversation.keys()) {
+        if (key !== runtime.key && !runtime.listener.conversationRuntimes.has(key)) {
+          oldestEvictedRuntimeKey = key;
+          break;
+        }
+      }
+      if (!oldestEvictedRuntimeKey)
+        break;
+      correlationsByConversation.delete(oldestEvictedRuntimeKey);
+    }
+  }
+  return {
+    appendDequeuedBatch(nextBatchId) {
+      for (const clientMessageId of takeDequeuedClientMessageIds(runtime, nextBatchId)) {
+        clientMessageIds.add(clientMessageId);
+      }
+    },
+    observeRun(runId) {
+      if (clientMessageIds.size === 0)
+        return;
+      const merged = new Set([
+        ...correlations.get(runId) ?? [],
+        ...clientMessageIds
+      ]);
+      correlations.delete(runId);
+      correlations.set(runId, [...merged]);
+      while (correlations.size > MAX_RECENT_RUN_CORRELATIONS) {
+        const oldestRunId = correlations.keys().next().value;
+        if (!oldestRunId)
+          break;
+        correlations.delete(oldestRunId);
+      }
+    }
+  };
+}
+var MAX_RECENT_RUN_CORRELATIONS = 32, MAX_RECENT_CONVERSATIONS = 256;
+var init_turn_correlation = __esm(() => {
+  init_runtime();
+});
+
 // src/websocket/listener/protocol-outbound.ts
 var exports_protocol_outbound = {};
 __export(exports_protocol_outbound, {
@@ -158651,6 +158792,7 @@ function buildLoopStatus(runtime, params) {
   return {
     status,
     active_run_ids: interruptedCacheActive && !conversationRuntime?.isProcessing ? [] : conversationRuntime?.activeRunId ? [conversationRuntime.activeRunId] : [],
+    ...buildTurnCorrelationSnapshot(listener, scopedAgentId, scopedConversationId),
     executing_tool_call_ids: status === "EXECUTING_CLIENT_SIDE_TOOL" && conversationRuntime ? [...conversationRuntime.turnLifecycle.executingToolCallIds] : []
   };
 }
@@ -158670,15 +158812,6 @@ function buildQueueSnapshot(runtime, params) {
     enqueued_at: new Date(item.enqueuedAt).toISOString()
   }));
 }
-function isStreamChannelMessage(type3) {
-  return STREAM_CHANNEL_MESSAGE_TYPES.has(type3);
-}
-function classifyOutboundFrame(message) {
-  if (message.type === "update_queue" && (message.removed?.length ?? 0) > 0) {
-    return "critical";
-  }
-  return COALESCABLE_STATUS_MESSAGE_TYPES.has(message.type) ? "status" : "critical";
-}
 function emitProtocolV2Message(socket, runtime, message, scope, routing) {
   const listener = getListenerRuntime(runtime);
   const runtimeScope = resolveRuntimeScope(listener, getScopeForRuntime(runtime, scope));
@@ -159044,7 +159177,7 @@ function emitStreamDelta(socket, runtime, delta2, scope, subagentId) {
   };
   emitProtocolV2Message(socket, runtime, message, scope, TO_SUBSCRIBERS);
 }
-var GIT_CONTEXT_CACHE_TTL_MS = 15000, MAX_GIT_CONTEXT_CACHE_ENTRIES = 64, FROZEN_SUPPORTED_COMMANDS, gitContextCache, STREAM_CHANNEL_MESSAGE_TYPES, COALESCABLE_STATUS_MESSAGE_TYPES;
+var GIT_CONTEXT_CACHE_TTL_MS = 15000, MAX_GIT_CONTEXT_CACHE_ENTRIES = 64, FROZEN_SUPPORTED_COMMANDS, gitContextCache;
 var init_protocol_outbound = __esm(() => {
   init_memory_filesystem2();
   init_subagent_state();
@@ -159065,23 +159198,12 @@ var init_protocol_outbound = __esm(() => {
   init_listener_constants();
   init_outbound_wire();
   init_permission_mode();
+  init_protocol_outbound_routing();
   init_runtime();
   init_transport();
+  init_turn_correlation();
   FROZEN_SUPPORTED_COMMANDS = [...SUPPORTED_REMOTE_COMMANDS];
   gitContextCache = new Map;
-  STREAM_CHANNEL_MESSAGE_TYPES = new Set([
-    "stream_delta",
-    "update_device_status",
-    "update_loop_status",
-    "update_queue",
-    "update_subagent_state"
-  ]);
-  COALESCABLE_STATUS_MESSAGE_TYPES = new Set([
-    "update_device_status",
-    "update_loop_status",
-    "update_queue",
-    "update_subagent_state"
-  ]);
 });
 
 // src/websocket/listener/cwd-change.ts
@@ -182861,7 +182983,71 @@ function rebuildInputWithFreshDenials(currentInput, serverApprovals, denialReaso
 function shouldAttemptApprovalRecovery(opts) {
   return opts.approvalPendingDetected && opts.retries < opts.maxRetries;
 }
-var INVALID_TOOL_CALL_IDS_FRAGMENT = "invalid tool call ids", APPROVAL_PENDING_DETAIL_FRAGMENT = "waiting for approval", CONVERSATION_BUSY_DETAIL_FRAGMENTS, CONVERSATION_BUSY_RUN_ID_PATTERN, EMPTY_RESPONSE_DETAIL_FRAGMENT = "empty content in", RETRYABLE_PROVIDER_DETAIL_PATTERNS, NON_RETRYABLE_PROVIDER_DETAIL_PATTERNS, NON_RETRYABLE_RUN_ERROR_TYPES, NON_RETRYABLE_429_REASONS, NON_RETRYABLE_QUOTA_DETAIL_PATTERNS, NON_RETRYABLE_4XX_PATTERN, RETRYABLE_429_PATTERN, DEFAULT_TRANSIENT_RETRY_BASE_DELAY_MS = 1000, CLOUDFLARE_EDGE_52X_RETRY_BASE_DELAY_MS = 5000, CONVERSATION_BUSY_RETRY_BASE_DELAY_MS = 1e4, EMPTY_RESPONSE_RETRY_BASE_DELAY_MS = 500, STALE_APPROVAL_RECOVERY_DENIAL_REASON = "The agent harness automatically closed this stale pending tool call to recover from a client/server state desync: the server was still waiting on a result for it, but the harness had no matching tool execution in flight and no result was ever recorded. It was not denied by the user or a permissions policy. Re-issue the tool call if you still need it.";
+function parseChatGPTUsageLimitDetail(detail) {
+  if (typeof detail !== "string")
+    return null;
+  if (!detail.toLowerCase().includes(CHATGPT_USAGE_LIMIT_FRAGMENT)) {
+    return null;
+  }
+  const fallback = { planType: null, resetsAt: null };
+  const jsonStart = detail.indexOf("{");
+  const jsonEnd = detail.lastIndexOf("}");
+  if (jsonStart === -1 || jsonEnd <= jsonStart)
+    return fallback;
+  let parsed;
+  try {
+    parsed = JSON.parse(detail.slice(jsonStart, jsonEnd + 1));
+  } catch {
+    return fallback;
+  }
+  if (typeof parsed !== "object" || parsed === null)
+    return fallback;
+  const errorField = parsed.error;
+  const errorObj = typeof errorField === "object" && errorField !== null ? errorField : parsed;
+  const planType = typeof errorObj.plan_type === "string" && errorObj.plan_type.length > 0 ? errorObj.plan_type : null;
+  let resetsAt = null;
+  if (typeof errorObj.resets_at === "number" && errorObj.resets_at > 0) {
+    resetsAt = errorObj.resets_at * 1000;
+  } else if (typeof errorObj.resets_in_seconds === "number" && errorObj.resets_in_seconds > 0) {
+    resetsAt = Date.now() + errorObj.resets_in_seconds * 1000;
+  }
+  return { planType, resetsAt };
+}
+function isChatGPTByokModel(model) {
+  return model.providerType === CHATGPT_OAUTH_PROVIDER_TYPE2 && model.providerCategory === BYOK_PROVIDER_CATEGORY;
+}
+function selectChatGPTQuotaFailoverHandle(params) {
+  const { currentHandle, models: models3, exhaustedProviders } = params;
+  const random = params.random ?? Math.random;
+  const slashIndex = currentHandle.indexOf("/");
+  if (slashIndex <= 0)
+    return null;
+  const currentProvider = currentHandle.slice(0, slashIndex);
+  const modelSuffix = currentHandle.slice(slashIndex + 1);
+  if (!modelSuffix)
+    return null;
+  const currentEntry = models3.find((m2) => m2.handle === currentHandle);
+  if (!currentEntry || !isChatGPTByokModel(currentEntry))
+    return null;
+  const candidates2 = models3.filter((m2) => {
+    if (!isChatGPTByokModel(m2))
+      return false;
+    const idx = m2.handle.indexOf("/");
+    if (idx <= 0)
+      return false;
+    const provider = m2.handle.slice(0, idx);
+    if (provider === currentProvider)
+      return false;
+    if (exhaustedProviders.has(provider))
+      return false;
+    return m2.handle.slice(idx + 1) === modelSuffix;
+  });
+  if (candidates2.length === 0)
+    return null;
+  const index = Math.min(Math.floor(random() * candidates2.length), candidates2.length - 1);
+  return candidates2[index]?.handle ?? null;
+}
+var INVALID_TOOL_CALL_IDS_FRAGMENT = "invalid tool call ids", APPROVAL_PENDING_DETAIL_FRAGMENT = "waiting for approval", CONVERSATION_BUSY_DETAIL_FRAGMENTS, CONVERSATION_BUSY_RUN_ID_PATTERN, EMPTY_RESPONSE_DETAIL_FRAGMENT = "empty content in", RETRYABLE_PROVIDER_DETAIL_PATTERNS, NON_RETRYABLE_PROVIDER_DETAIL_PATTERNS, NON_RETRYABLE_RUN_ERROR_TYPES, NON_RETRYABLE_429_REASONS, NON_RETRYABLE_QUOTA_DETAIL_PATTERNS, NON_RETRYABLE_4XX_PATTERN, RETRYABLE_429_PATTERN, DEFAULT_TRANSIENT_RETRY_BASE_DELAY_MS = 1000, CLOUDFLARE_EDGE_52X_RETRY_BASE_DELAY_MS = 5000, CONVERSATION_BUSY_RETRY_BASE_DELAY_MS = 1e4, EMPTY_RESPONSE_RETRY_BASE_DELAY_MS = 500, STALE_APPROVAL_RECOVERY_DENIAL_REASON = "The agent harness automatically closed this stale pending tool call to recover from a client/server state desync: the server was still waiting on a result for it, but the harness had no matching tool execution in flight and no result was ever recorded. It was not denied by the user or a permissions policy. Re-issue the tool call if you still need it.", CHATGPT_USAGE_LIMIT_FRAGMENT = "usage_limit_reached", CHATGPT_OAUTH_PROVIDER_TYPE2 = "chatgpt_oauth", BYOK_PROVIDER_CATEGORY = "byok";
 var init_turn_recovery_policy = __esm(() => {
   init_error_formatter();
   init_zai_errors();
@@ -259720,9 +259906,11 @@ async function runLocalOAuthConnectFlow(provider, callbacks) {
     baseURL: callbacks.baseURL,
     timeout: callbacks.timeout
   });
+  clearAvailableModelsCache();
   return { providerName: provider.providerName };
 }
 var init_connect_local_oauth = __esm(() => {
+  init_available_models();
   init_pi_oauth();
   init_local_provider_auth_store();
   init_connect_oauth_core();
@@ -259775,7 +259963,11 @@ function listConnectProvidersForHelp(target2 = defaultProviderStorageTarget()) {
   return providers;
 }
 function listConnectProviderTokens(target2 = defaultProviderStorageTarget()) {
-  return [...listConnectProvidersForHelp(target2), "codex"];
+  return [
+    ...listConnectProvidersForHelp(target2),
+    "codex",
+    ...target2 === "local" ? ["grok"] : []
+  ];
 }
 function isConnectOAuthProvider(provider) {
   return provider.byokProvider.isOAuth === true;
@@ -259811,6 +260003,7 @@ var init_connect_normalize = __esm(() => {
   };
   LOCAL_ALIAS_TO_CANONICAL = {
     gemini: "google",
+    grok: "xai",
     "kimi-code": "kimi-coding",
     moonshot: "moonshotai",
     bedrock: "amazon-bedrock"
@@ -267863,6 +268056,17 @@ function buildQueuedTurnMessage(runtime, batch) {
     messages
   };
 }
+function getDequeuedClientMessageIds(runtime, batch) {
+  const clientMessageIds = new Set;
+  for (const item of batch.items) {
+    const queuedMessage = runtime.queuedMessagesByItemId.get(item.id);
+    const inboundClientMessageIds = queuedMessage ? getInboundClientMessageIds(queuedMessage) : [];
+    for (const clientMessageId of inboundClientMessageIds.length > 0 ? inboundClientMessageIds : item.clientMessageId ? [item.clientMessageId] : []) {
+      clientMessageIds.add(clientMessageId);
+    }
+  }
+  return [...clientMessageIds];
+}
 function shouldQueueInboundMessage(parsed) {
   return parsed.messages.some((payload) => ("content" in payload));
 }
@@ -267933,10 +268137,14 @@ function consumeQueuedTurn(runtime) {
   if (!dequeuedBatch) {
     return null;
   }
+  const clientMessageIds = getDequeuedClientMessageIds(runtime, dequeuedBatch);
   const queuedTurn = buildQueuedTurnMessage(runtime, dequeuedBatch);
   if (!queuedTurn) {
     return null;
   }
+  if (clientMessageIds.length > 0) {
+    runtime.dequeuedClientMessageIdsByBatchId.set(dequeuedBatch.batchId, clientMessageIds);
+  }
   return {
     dequeuedBatch,
     queuedTurn
@@ -267949,6 +268157,17 @@ function computeListenerQueueBlockedReason(runtime) {
   });
   return getListenerBlockedReason(runtime.turnLifecycle.snapshot(), activeScope ? getPendingControlRequestCount(runtime.listener, activeScope) : 0);
 }
+function emitTurnBoundaryStatus(runtime, socket) {
+  if (!isListenerTransportOpen(socket)) {
+    return;
+  }
+  const scope = {
+    agent_id: runtime.agentId,
+    conversation_id: runtime.conversationId
+  };
+  emitQueueUpdate(socket, runtime, scope);
+  emitLoopStatusUpdate(socket, runtime, scope);
+}
 async function drainQueuedMessages(runtime, socket, opts, processQueuedTurn) {
   if (runtime.queuePumpActive) {
     return;
@@ -267970,6 +268189,7 @@ async function drainQueuedMessages(runtime, socket, opts, processQueuedTurn) {
       }
       const { dequeuedBatch, queuedTurn } = consumedQueuedTurn;
       emitDequeuedUserMessage(socket, runtime, queuedTurn, dequeuedBatch);
+      emitTurnBoundaryStatus(runtime, socket);
       const preTurnStatus = getListenerStatus(runtime.listener) === "processing" ? "processing" : "receiving";
       if (opts.connectionId && runtime.listener.lastEmittedStatus !== preTurnStatus) {
         runtime.listener.lastEmittedStatus = preTurnStatus;
@@ -267977,6 +268197,7 @@ async function drainQueuedMessages(runtime, socket, opts, processQueuedTurn) {
       }
       await processQueuedTurn(queuedTurn, dequeuedBatch);
       emitListenerStatus(runtime.listener, opts.onStatusChange, opts.connectionId);
+      emitTurnBoundaryStatus(runtime, socket);
       evictConversationRuntimeIfIdle(runtime);
     }
   } finally {
@@ -272546,9 +272767,9 @@ import {
 import { join as join59 } from "node:path";
 function buildReflectionSubagentPrompt(input) {
   const lines = [];
-  lines.push('Review the conversation transcript payload and update memory files. The payload path is available as the `$TRANSCRIPT_PATH` env var — read it via Bash (e.g. `wc -c "$TRANSCRIPT_PATH"`). Note: `$TRANSCRIPT_PATH` only expands in shell commands; Edit `file_path` is literal and does NOT expand env vars.', "", 'The payload may be either a JSON message array for one conversation or a `multi_transcript_reflection_payload` manifest. If it is a manifest, read each `payload_path` listed in `transcripts` and synthesize across all conversations. Entries with `mode: "replay"` were already reflected before and are included intentionally for re-review/deduplication; do not ignore them just because they are replay slices.', "When reviewing multiple transcripts, prefer durable patterns and latest evidence across sessions. Resolve contradictions by updating stale memory at the source, deduplicate repeated facts, and avoid storing one-off task state.", "", "The primary agent's memory filesystem is available through the `$MEMORY_DIR` environment variable.", "Run git add or git commit commands only from $MEMORY_DIR; the harness handles integration after your commit. If these fail, stop reflecting and report the failure. All other git commands are out of your purview.", 'When using Edit, first resolve the absolute file path from `$MEMORY_DIR` with Bash (for example: `printf "%s/system/persona.md\\n" "$MEMORY_DIR"`) and use the printed path. Do not hardcode memory paths from the prompt.', "In-context memory (in the parent agent's system prompt) is stored in the `system/` folder and are rendered in <memory> tags below. Modification to files in `system/` will edit the parent agent's system prompt.", "Additional memory files (such as skills and external memory) may also be read and modified.", "");
+  lines.push('Review the conversation transcript payload and update memory files. The payload path is available as the `$TRANSCRIPT_PATH` env var — read it via Bash (e.g. `wc -c "$TRANSCRIPT_PATH"`). Note: `$TRANSCRIPT_PATH` only expands in shell commands; Edit `file_path` is literal and does NOT expand env vars.', "", 'The payload may be either a JSON message array for one conversation or a `multi_transcript_reflection_payload` manifest. If it is a manifest, read each `payload_path` listed in `transcripts` and synthesize across all conversations. Entries with `mode: "replay"` were already reflected before and are included intentionally for re-review/deduplication; do not ignore them just because they are replay slices.', "When reviewing multiple transcripts, prefer recurring patterns and the latest evidence across sessions. Resolve contradictions by updating stale memory at the source, deduplicate repeated facts, and avoid storing one-off task state.", "", "The primary agent's memory filesystem is available through the `$MEMORY_DIR` environment variable.", "Run git add or git commit commands only from $MEMORY_DIR; the harness handles integration after your commit. If these fail, stop reflecting and report the failure. All other git commands are out of your purview.", 'When using Edit, first resolve the absolute file path from `$MEMORY_DIR` with Bash (for example: `printf "%s/system/persona.md\\n" "$MEMORY_DIR"`) and use the printed path. Do not hardcode memory paths from the prompt.', "In-context memory (in the parent agent's system prompt) is stored in the `system/` folder and are rendered in <memory> tags below. Modification to files in `system/` will edit the parent agent's system prompt.", "Additional memory files (such as skills and external memory) may also be read and modified.", "");
   if (input.instruction?.trim()) {
-    lines.push("Additional user-provided reflection instruction:", input.instruction.trim(), "", "Use this instruction to focus what you look for, but still only persist durable memory-worthy learnings and do not store transient task state.", "");
+    lines.push("Additional user-provided reflection instruction:", input.instruction.trim(), "", "Use this instruction to focus what you look for, but still only persist memory-worthy learnings and do not store transient task state.", "");
   }
   if (input.parentMemory) {
     lines.push(input.parentMemory);
@@ -272566,7 +272787,7 @@ function buildReflectionSelectorPrompt(options3) {
   if (options3?.instruction?.trim()) {
     lines.push("Additional user-provided reflection instruction:", options3.instruction.trim(), "", "Prefer transcript candidates that help satisfy this instruction, while still avoiding transient or low-signal conversations.", "");
   }
-  lines.push("If the candidates payload includes `user_instruction`, use it as the requested focus for selection.", "", "Select up to `max_selected` conversations. Prefer candidates likely to contain durable memory updates: explicit user corrections, repeated preferences, coding/review/commit style preferences, repo or workflow gotchas, durable facts about people/projects, contradictions with current memory, or repeated agent failures.", "Avoid one-off debugging, transient task status, duplicated/redundant candidates, and conversations already fully reflected unless they are useful for deduplication or contradiction resolution.", "Treat summaries/descriptions as weak internal metadata, not confirmed facts. The final reflection pass will verify against the actual transcript before writing memory.", "", "Return strict JSON as your final response with this shape:", '{"selected_conversations":[{"conversation_id":"conv-...","reason":"durable reason for selecting this transcript","priority":"high"}]}', 'Use priority values `high`, `medium`, or `low`. If nothing looks memory-worthy, write `{"selected_conversations":[]}`.');
+  lines.push("If the candidates payload includes `user_instruction`, use it as the requested focus for selection.", "", "Select up to `max_selected` conversations. Prefer candidates likely to contain useful memory updates: explicit user corrections, repeated preferences, coding/review/commit style preferences, repo or workflow gotchas, facts about people/projects that will matter later, contradictions with current memory, or repeated agent failures.", "Avoid one-off debugging, transient task status, duplicated/redundant candidates, and conversations already fully reflected unless they are useful for deduplication or contradiction resolution.", "Treat summaries/descriptions as weak internal metadata, not confirmed facts. The final reflection pass will verify against the actual transcript before writing memory.", "", "Return strict JSON as your final response with this shape:", '{"selected_conversations":[{"conversation_id":"conv-...","reason":"reason for selecting this transcript","priority":"high"}]}', 'Use priority values `high`, `medium`, or `low`. If nothing looks memory-worthy, write `{"selected_conversations":[]}`.');
   return lines.join(`
 `);
 }
@@ -273351,7 +273572,7 @@ async function buildReflectionAutoPayload(options3) {
     created_at: new Date().toISOString(),
     max_selected: maxSelected,
     user_instruction: instruction?.trim() || undefined,
-    instructions: "Choose conversations likely to contain durable memory updates. Prefer explicit corrections, repeated preferences, project conventions, and contradictions; avoid one-off debugging and transient task state.",
+    instructions: "Choose conversations likely to contain useful memory updates. Prefer explicit corrections, repeated preferences, project conventions, and contradictions; avoid one-off debugging and transient task state.",
     candidates: sortedCandidates
   };
   const candidatesPath = buildPayloadPath(payloadRoot, "candidates");
@@ -273661,7 +273882,7 @@ var init_reflection_transcript = __esm(() => {
   REFLECTION_AUTO_QUERIES = [
     {
       id: "user-corrections",
-      query: "user corrections and preferences repeated mistakes durable feedback"
+      query: "user corrections and preferences repeated mistakes recurring feedback"
     },
     {
       id: "coding-style",
@@ -273669,11 +273890,11 @@ var init_reflection_transcript = __esm(() => {
     },
     {
       id: "collaboration",
-      query: "collaboration communication style team preferences durable workflow"
+      query: "collaboration communication style team preferences recurring workflow"
     },
     {
       id: "repo-gotchas",
-      query: "repo conventions project gotchas durable implementation details"
+      query: "repo conventions project gotchas important implementation details"
     },
     {
       id: "long-term-facts",
@@ -273937,7 +274158,7 @@ function buildTargetInstruction(target2) {
     "(--- ... ---) at the top intact and edit only the body below it.",
     "",
     "If it does not exist, create it ONLY when the new experience yields",
-    "durable, forward-looking guidance worth recording, and begin the file with",
+    "forward-looking guidance worth recording, and begin the file with",
     "a YAML frontmatter block (a --- ... --- header with a short `description:`).",
     "If there is nothing substantive to record yet, do NOT create the file —",
     "leave it absent rather than writing a placeholder that says nothing was",
@@ -274022,7 +274243,7 @@ var init_dream_targets = __esm(() => {
   };
   AGENTS_MD_GUIDANCE = [
     "This file is an AGENTS.md — a README *for coding agents* (the agents.md",
-    "open standard). Maintain it as durable, repo-level guidance an agent needs:",
+    "open standard). Maintain it as repo-level guidance an agent needs over time:",
     "build/test/setup COMMANDS (put these early), code-style conventions,",
     "testing instructions, security considerations, and commit/PR rules. Use",
     "plain Markdown with any headings. Do NOT duplicate human-README content",
@@ -274035,7 +274256,7 @@ var init_dream_targets = __esm(() => {
   ].join(`
 `);
   GENERIC_GUIDANCE = [
-    "Maintain this markdown document as durable, forward-looking guidance",
+    "Maintain this markdown document as forward-looking guidance",
     "distilled from the conversation(s). Prefer editing/merging existing",
     "sections over appending; remove content that new evidence contradicts."
   ].join(`
@@ -274841,7 +275062,7 @@ function getReflectionCompletionMessage(integration) {
     case "merged":
       return;
     case "no_changes":
-      return ({ action: action3 }) => `${action3}; no durable memory changes were needed.`;
+      return ({ action: action3 }) => `${action3}; no memory changes were needed.`;
     case "parent_dirty":
       return "Tried to reflect, but parent memory had uncommitted changes; will retry later.";
     case "merge_conflict":
@@ -457337,9 +457558,13 @@ async function isRetriablePostStopError(stopReason, lastRunId, fallbackDetail) {
 async function drainRecoveryStreamWithEmission(recoveryStream, socket, runtime, params) {
   let recoveryRunIdSent = false;
   return drainStreamWithResume(recoveryStream, createBuffers(params.agentId || ""), () => {}, params.turnLease.signal, undefined, ({ chunk: chunk2, shouldOutput, errorInfo }) => {
+    if (!runtime.turnLifecycle.isCurrent(params.turnLease)) {
+      return;
+    }
     const maybeRunId = chunk2.run_id;
     if (typeof maybeRunId === "string") {
       runtime.turnLifecycle.setRunId(params.turnLease, maybeRunId);
+      params.turnCorrelation?.observeRun(maybeRunId);
       if (!recoveryRunIdSent) {
         recoveryRunIdSent = true;
         emitLoopStatusUpdate(socket, runtime, {
@@ -457686,15 +457911,23 @@ async function resolveRecoveredApprovalResponse(runtime, socket, response, proce
     ]);
     let continuationBatchId = `batch-recovered-${crypto.randomUUID()}`;
     let continuationActingUserId;
+    let continuationCorrelation;
     const consumedQueuedTurn = consumeQueuedTurn(runtime);
     if (consumedQueuedTurn) {
       const { dequeuedBatch, queuedTurn } = consumedQueuedTurn;
       continuationBatchId = dequeuedBatch.batchId;
       continuationActingUserId = queuedTurn.actingUserId;
       continuationInput = appendQueuedTurnToInput(continuationInput, queuedTurn);
+      continuationCorrelation = createTurnCorrelation(runtime, {
+        type: "message",
+        agentId: recovered.agentId,
+        conversationId: recovered.conversationId,
+        messages: continuationInput.messages
+      }, continuationBatchId);
       emitDequeuedUserMessage(socket, runtime, queuedTurn, dequeuedBatch);
     }
     if (!runtime.turnLifecycle.isCurrent(recoveryLease)) {
+      runtime.dequeuedClientMessageIdsByBatchId.delete(continuationBatchId);
       return true;
     }
     await processTurn({
@@ -457703,7 +457936,7 @@ async function resolveRecoveredApprovalResponse(runtime, socket, response, proce
       conversationId: recovered.conversationId,
       ...continuationActingUserId ? { actingUserId: continuationActingUserId } : {},
       messages: continuationInput.messages
-    }, socket, runtime, opts?.onStatusChange, opts?.connectionId, continuationBatchId, recoveryLease);
+    }, socket, runtime, opts?.onStatusChange, opts?.connectionId, continuationBatchId, recoveryLease, continuationCorrelation);
     if (runtime.turnLifecycle.isCurrent(recoveryLease)) {
       throw new Error("Recovered continuation returned without finalizing");
     }
@@ -457756,6 +457989,7 @@ var init_recovery = __esm(async () => {
   init_recoverable_notices();
   init_runtime();
   init_secrets_sync();
+  init_turn_correlation();
   init_turn_input_state();
   init_turn_status();
   init_turn_terminal();
@@ -457811,6 +458045,96 @@ var init_approval_recovery = __esm(() => {
   init_turn_recovery_policy();
 });
 
+// src/agent/chatgpt-plan-rotation.ts
+function providerFromHandle(handle2) {
+  const idx = handle2.indexOf("/");
+  return idx > 0 ? handle2.slice(0, idx) : null;
+}
+function isChatGPTByokHandleInModels(handle2, models3) {
+  const entry = models3.find((m4) => m4.handle === handle2);
+  return entry?.providerType === "chatgpt_oauth" && entry?.providerCategory === "byok";
+}
+async function resolveAgentModelHandle2(agentId) {
+  try {
+    const agent2 = await getBackend().retrieveAgent(agentId);
+    const record5 = agent2;
+    if (typeof record5.model === "string" && record5.model.length > 0) {
+      return record5.model;
+    }
+    return resolveModelHandleFromLlmConfig(record5.llm_config);
+  } catch {
+    return null;
+  }
+}
+async function rotateChatGPTPlanOnQuotaLimit(params) {
+  const { agentId, detail } = params;
+  const parsedDetail = parseChatGPTUsageLimitDetail(detail);
+  if (!parsedDetail)
+    return null;
+  let models3 = getCachedAvailableModels();
+  if (!models3) {
+    try {
+      await getAvailableModelHandles();
+      models3 = getCachedAvailableModels();
+    } catch {
+      return null;
+    }
+  }
+  if (!models3)
+    return null;
+  let currentHandle = params.currentHandle;
+  if (!currentHandle || !isChatGPTByokHandleInModels(currentHandle, models3)) {
+    currentHandle = await resolveAgentModelHandle2(agentId);
+  }
+  if (!currentHandle || !isChatGPTByokHandleInModels(currentHandle, models3)) {
+    return null;
+  }
+  const fromProvider = providerFromHandle(currentHandle);
+  if (!fromProvider)
+    return null;
+  exhaustedProviders.add(fromProvider);
+  const toHandle = selectChatGPTQuotaFailoverHandle({
+    currentHandle,
+    models: models3,
+    exhaustedProviders
+  });
+  if (!toHandle)
+    return null;
+  const toProvider = providerFromHandle(toHandle);
+  if (!toProvider)
+    return null;
+  try {
+    await updateAgentLLMConfig(agentId, toHandle, {
+      provider_type: "chatgpt_oauth"
+    });
+  } catch {
+    return null;
+  }
+  return {
+    fromProvider,
+    toProvider,
+    toHandle,
+    resetsAt: parsedDetail.resetsAt
+  };
+}
+function formatPlanRotationNotice(params) {
+  const { fromProvider, toProvider, resetsAt } = params;
+  const resetSuffix = resetsAt !== null ? ` (resets ${new Date(resetsAt).toLocaleTimeString([], {
+    hour: "numeric",
+    minute: "2-digit"
+  })})` : "";
+  return `${fromProvider} hit its usage limit${resetSuffix} — switched to ${toProvider}`;
+}
+var CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN = 3, exhaustedProviders;
+var init_chatgpt_plan_rotation = __esm(() => {
+  init_available_models();
+  init_model_handles();
+  init_modify();
+  init_turn_recovery_policy();
+  init_backend2();
+  exhaustedProviders = new Set;
+});
+
 // src/websocket/listener/skill-injection.ts
 function injectQueuedSkillContent(messages, context3) {
   const skillContents = consumeQueuedSkillContent();
@@ -458055,9 +458379,11 @@ async function resolveStaleApprovals(runtime, socket, turnLease, deps = {}) {
         }
       ]);
       let continuationActingUserId;
+      let recoveryTurnCorrelation;
       const consumedQueuedTurn = consumeQueuedTurn(runtime);
       if (consumedQueuedTurn) {
         const { dequeuedBatch, queuedTurn } = consumedQueuedTurn;
+        recoveryTurnCorrelation = createTurnCorrelation(runtime, queuedTurn, dequeuedBatch.batchId);
         continuationActingUserId = queuedTurn.actingUserId;
         continuationInput = appendQueuedTurnToInput(continuationInput, queuedTurn);
         emitDequeuedUserMessage(socket, runtime, queuedTurn, dequeuedBatch);
@@ -458088,7 +458414,8 @@ async function resolveStaleApprovals(runtime, socket, turnLease, deps = {}) {
       const drainResult = await drainRecoveryStream(recoveryStream, socket, runtime, {
         agentId: runtime.agentId ?? undefined,
         conversationId: recoveryConversationId,
-        turnLease
+        turnLease,
+        turnCorrelation: recoveryTurnCorrelation
       });
       assertCurrentTurnLease();
       if (drainResult.stopReason === "error") {
@@ -458383,6 +458710,7 @@ var init_send = __esm(async () => {
   init_permission_mode();
   init_protocol_outbound();
   init_skill_injection();
+  init_turn_correlation();
   init_turn_input_state();
   init_turn_status();
   await __promiseAll([
@@ -458398,6 +458726,20 @@ var init_send = __esm(async () => {
 });
 
 // src/websocket/listener/teleport.ts
+function buildTeleportContinuationMessages(params) {
+  return [
+    {
+      type: "approval",
+      approvals: params.approvals,
+      otid: params.teleportId
+    },
+    {
+      role: "system",
+      content: "<system-reminder>Teleportation to this environment is complete. Continue the existing task from this environment now.</system-reminder>",
+      otid: `${params.teleportId}:continue`
+    }
+  ];
+}
 function getPendingTeleports(runtime) {
   runtime.pendingTeleports ??= new Map;
   return runtime.pendingTeleports;
@@ -460708,6 +461050,7 @@ async function handleApprovalStop(params) {
     turnInput,
     turnToolContextId,
     turnLease,
+    turnCorrelation,
     processOwnedTurn = false,
     buildSendOptions,
     dependencies: dependencies4
@@ -461010,6 +461353,7 @@ async function handleApprovalStop(params) {
   const consumedQueuedTurn = consumeQueuedTurn(runtime);
   if (consumedQueuedTurn) {
     const { dequeuedBatch, queuedTurn } = consumedQueuedTurn;
+    turnCorrelation?.appendDequeuedBatch(dequeuedBatch.batchId);
     continuationBatchId = dequeuedBatch.batchId;
     continuationActingUserId = queuedTurn.actingUserId;
     nextTurnInput = appendQueuedTurnToInput(nextTurnInput, queuedTurn);
@@ -462795,23 +463139,24 @@ var init_turn_setup = __esm(async () => {
 });
 
 // src/websocket/listener/turn.ts
-async function handleIncomingMessage(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId = `batch-direct-${crypto.randomUUID()}`, existingTurnLease) {
+async function handleIncomingMessage(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId = `batch-direct-${crypto.randomUUID()}`, existingTurnLease, existingTurnCorrelation) {
   notifyTurnStarted(msg);
   try {
-    await handleIncomingMessageInner(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId, existingTurnLease);
+    await handleIncomingMessageInner(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId, existingTurnLease, existingTurnCorrelation);
   } finally {
     notifyTurnFinished(msg);
     finishPendingTeleport(runtime);
   }
 }
-async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId = `batch-direct-${crypto.randomUUID()}`, existingTurnLease) {
+async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange, connectionId, dequeuedBatchId = `batch-direct-${crypto.randomUUID()}`, existingTurnLease, existingTurnCorrelation) {
   const agentId = msg.agentId;
   const requestedConversationId = msg.conversationId || undefined;
   const conversationId = requestedConversationId ?? "default";
   const normalizedAgentId = normalizeCwdAgentId(agentId);
   const turnWorkingDirectory = getConversationWorkingDirectory(runtime.listener, normalizedAgentId, conversationId);
   const turnPermissionModeState = getOrCreateConversationPermissionModeStateRef(runtime.listener, normalizedAgentId, conversationId);
-  let postStopApprovalRecoveryRetries = 0, llmApiErrorRetries = 0, emptyResponseRetries = 0, lastApprovalContinuationAccepted = false, activeDequeuedBatchId = dequeuedBatchId;
+  let postStopApprovalRecoveryRetries = 0, llmApiErrorRetries = 0, emptyResponseRetries = 0, chatgptPlanSwaps = 0, lastApprovalContinuationAccepted = false, activeDequeuedBatchId = dequeuedBatchId;
+  const turnCorrelation = existingTurnCorrelation ?? createTurnCorrelation(runtime, msg, activeDequeuedBatchId);
   const msgRunIds = [];
   let lastExecutionResults = null;
   let lastExecutingToolCallIds = [];
@@ -462980,6 +463325,7 @@ async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange,
         if (typeof maybeRunId === "string") {
           runId = maybeRunId;
           runtime.turnLifecycle.setRunId(turnLease, maybeRunId);
+          turnCorrelation.observeRun(maybeRunId);
           if (!runIdSent) {
             runIdSent = true;
             msgRunIds.push(maybeRunId);
@@ -463181,6 +463527,51 @@ async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange,
           turnToolContextId = getStreamToolContextId(stream12);
           continue;
         }
+        if (chatgptPlanSwaps < CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN) {
+          const rotation = await rotateChatGPTPlanOnQuotaLimit({
+            agentId,
+            currentHandle: null,
+            detail: errorDetail2
+          });
+          if (rotation) {
+            chatgptPlanSwaps += 1;
+            emitRecoverableRetryNotice(socket, runtime, {
+              kind: "transient_provider_retry",
+              message: formatPlanRotationNotice(rotation),
+              reason: "llm_api_error",
+              attempt: chatgptPlanSwaps,
+              maxAttempts: CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN,
+              delayMs: 0,
+              runId: lastRunId || undefined,
+              agentId,
+              conversationId
+            });
+            if (turnAbortSignal.aborted) {
+              throw new Error("Cancelled by user");
+            }
+            turnInput = refreshTurnInputOtidsForNewRequest(turnInput);
+            setTurnLoopStatus(runtime, turnLease, "SENDING_API_REQUEST", {
+              agent_id: agentId,
+              conversation_id: conversationId
+            });
+            const retryInputWithSkillContent = injectQueuedSkillContent(turnInput.messages, { socket, runtime, agentId, conversationId });
+            const retrySendResult = await turnInputSender.send(retryInputWithSkillContent);
+            turnInput = updateTurnInputMessagesPreservingOtids(turnInput, retryInputWithSkillContent);
+            const retryStream = turnInputSender.accept(retrySendResult);
+            if (!retryStream) {
+              return;
+            }
+            stream12 = retryStream;
+            pendingNormalizationInterruptedToolCallIds = [];
+            markAwaitingAcceptedApprovalContinuationRunId(runtime, turnLease, turnInput.messages);
+            setTurnLoopStatus(runtime, turnLease, "PROCESSING_API_RESPONSE", {
+              agent_id: agentId,
+              conversation_id: conversationId
+            });
+            turnToolContextId = getStreamToolContextId(stream12);
+            continue;
+          }
+        }
         const retriable = await isRetriablePostStopError(stopReason || "error", lastRunId, errorDetail2);
         if (finishIfInterrupted(lastRunId || runtime.activeRunId)) {
           break;
@@ -463287,6 +463678,7 @@ async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange,
         pendingNormalizationInterruptedToolCallIds,
         turnToolContextId,
         turnLease,
+        turnCorrelation,
         processOwnedTurn: msg.processOwnedTurn === true,
         buildSendOptions
       });
@@ -463464,6 +463856,7 @@ async function handleIncomingMessageInner(msg, socket, runtime, onStatusChange,
 }
 var init_turn = __esm(async () => {
   init_approval_recovery();
+  init_chatgpt_plan_rotation();
   init_check_approval();
   init_turn_recovery_policy();
   init_backend2();
@@ -463483,6 +463876,7 @@ var init_turn = __esm(async () => {
   init_teleport();
   init_turn_cleanup();
   init_turn_context();
+  init_turn_correlation();
   init_turn_input_state();
   init_turn_observers();
   init_turn_status();
@@ -464079,6 +464473,7 @@ function createConnectionTurnProcessor(runtime) {
     }
     const connection = runtime.connections.get(queuedTurn.connectionId);
     if (!connection || connection.cancellation.signal.aborted) {
+      scopedRuntime.dequeuedClientMessageIdsByBatchId.delete(dequeuedBatch.batchId);
       return;
     }
     await handleIncomingMessage(queuedTurn, getOrCreateProcessTransport(runtime), scopedRuntime, connection.options.onStatusChange, connection.id, dequeuedBatch.batchId);
@@ -467726,32 +468121,6 @@ var init_git_branches = __esm(() => {
   init_protocol_outbound();
 });
 
-// src/websocket/listener/inbound-queue.ts
-function getInboundClientMessageId(incoming) {
-  return incoming.messages.find((payload) => ("content" in payload))?.client_message_id;
-}
-function enqueueInboundUserMessage(runtime, incoming, actingUserId) {
-  const firstUserPayload = incoming.messages.find((payload) => ("content" in payload));
-  if (!firstUserPayload) {
-    return false;
-  }
-  const enqueuedItem = runtime.queueRuntime.enqueue({
-    kind: "message",
-    source: "user",
-    content: firstUserPayload.content,
-    clientMessageId: firstUserPayload.client_message_id ?? `cm-submit-${crypto.randomUUID()}`,
-    agentId: incoming.agentId,
-    conversationId: incoming.conversationId || "default",
-    ...incoming.noCoalesce ? { noCoalesce: true } : {},
-    actingUserId
-  });
-  if (!enqueuedItem) {
-    return false;
-  }
-  runtime.queuedMessagesByItemId.set(enqueuedItem.id, incoming);
-  return true;
-}
-
 // src/websocket/listener/inbound-dispatch.ts
 function getAcceptedInputDisposition(runtime, clientMessageId) {
   if (!clientMessageId)
@@ -468497,13 +468866,10 @@ function createListenerMessageHandler(params) {
               connectionId,
               agentId: parsed.runtime.agent_id,
               conversationId: parsed.runtime.conversation_id,
-              messages: [
-                {
-                  type: "approval",
-                  approvals,
-                  otid: teleportId
-                }
-              ]
+              messages: buildTeleportContinuationMessages({
+                teleportId,
+                approvals
+              })
             }, socket, scopedRuntime2, opts.onStatusChange, connectionId);
           });
           return;
@@ -468673,12 +469039,10 @@ function createListenerMessageHandler(params) {
           success: removed !== null,
           item_id: parsed.item_id
         }, "remove_queue_item_response", "remove_queue_item");
-        if (removed !== null) {
-          emitQueueUpdateIfOpen(runtime, {
-            agent_id: parsed.runtime.agent_id,
-            conversation_id: parsed.runtime.conversation_id
-          });
-        }
+        emitQueueUpdateIfOpen(runtime, {
+          agent_id: parsed.runtime.agent_id,
+          conversation_id: parsed.runtime.conversation_id
+        });
         return;
       }
       if (fileCommandSession.handle(parsed)) {
@@ -485848,6 +486212,9 @@ function isAssistantMessage2(message) {
 function isUserMessage(message) {
   return message.message_type === "user_message";
 }
+function isTaskNotificationMessage(message) {
+  return extractMessageText(message).trimStart().startsWith("<task-notification>");
+}
 function messageRunId(message) {
   if (!message)
     return null;
@@ -485884,8 +486251,9 @@ async function waitForEnvironmentAssistantMessage(params) {
         return sequenceId !== null && sequenceId > anchorSequenceId;
       });
       const nextUserSequenceId = newerMessages.reduce((closest, message) => {
-        if (!isUserMessage(message))
+        if (!isUserMessage(message) || isTaskNotificationMessage(message)) {
           return closest;
+        }
         const sequenceId = messageSequenceId(message);
         if (sequenceId === null)
           return closest;
@@ -497557,7 +497925,7 @@ var init_mcp_client = __esm(() => {
   init_streamableHttp();
   DEFAULT_CLIENT_INFO = {
     name: "letta-code",
-    version: "0.30.26"
+    version: "0.30.28"
   };
 });
 
@@ -499949,6 +500317,7 @@ ${loadedContents.join(`
   let llmApiErrorRetries = 0;
   let emptyResponseRetries = 0;
   let conversationBusyRetries = 0;
+  let chatgptPlanSwaps = 0;
   markMilestone("HEADLESS_FIRST_STREAM_START");
   measureSinceMilestone("headless-setup-total", "HEADLESS_CLIENT_READY");
   const checkMaxTurns = async () => {
@@ -500215,6 +500584,7 @@ ${loadedContents.join(`
         llmApiErrorRetries = 0;
         emptyResponseRetries = 0;
         conversationBusyRetries = 0;
+        chatgptPlanSwaps = 0;
         const continueMessage = await emitHeadlessTurnEnd({
           agent: agent2,
           conversationId,
@@ -500315,6 +500685,34 @@ ${loadedContents.join(`
         }
       }
       const detailFromRun = await fetchRunErrorDetail(lastRunId);
+      if (chatgptPlanSwaps < CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN) {
+        const rotation = await rotateChatGPTPlanOnQuotaLimit({
+          agentId: agent2.id,
+          currentHandle: null,
+          detail: detailFromRun ?? latestErrorText
+        });
+        if (rotation) {
+          chatgptPlanSwaps += 1;
+          const rotationMessage = formatPlanRotationNotice(rotation);
+          if (outputFormat === "stream-json") {
+            const retryMsg = {
+              type: "retry",
+              reason: "llm_api_error",
+              attempt: chatgptPlanSwaps,
+              max_attempts: CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN,
+              delay_ms: 0,
+              run_id: lastRunId ?? undefined,
+              session_id: sessionId,
+              uuid: `retry-${lastRunId || randomUUID35()}`
+            };
+            writeWireMessage(retryMsg);
+          } else {
+            console.error(rotationMessage);
+          }
+          currentInput = refreshInputOtidsForNewRequest(currentInput);
+          continue;
+        }
+      }
       if (stopReason === "llm_api_error") {
         if (llmApiErrorRetries < LLM_API_ERROR_MAX_RETRIES2) {
           const attempt2 = llmApiErrorRetries + 1;
@@ -501686,6 +502084,7 @@ var init_headless = __esm(async () => {
   init_subagent_stdout_failure();
   init_approval_recovery();
   init_bootstrap_handler();
+  init_chatgpt_plan_rotation();
   init_client_skills();
   init_context();
   init_create5();
@@ -547536,6 +547935,7 @@ function useConversationLoop(ctx) {
     buffersRef,
     clearApprovalToolContext,
     closeTrajectorySegment,
+    chatgptPlanSwapsRef,
     consumeQueuedMessages,
     queueModeRef,
     contextTrackerRef,
@@ -547767,6 +548167,7 @@ function useConversationLoop(ctx) {
       emptyResponseRetriesRef.current = 0;
       conversationBusyRetriesRef.current = 0;
       quotaAutoSwapAttemptedRef.current = false;
+      chatgptPlanSwapsRef.current = 0;
     }
     let currentRunId;
     let preserveTranscriptStartForApproval = false;
@@ -548731,6 +549132,27 @@ ${feedback}
           buffersRef.current.interrupted = false;
           continue;
         }
+        if (chatgptPlanSwapsRef.current < CHATGPT_PLAN_ROTATION_MAX_SWAPS_PER_TURN) {
+          const rotation = await rotateChatGPTPlanOnQuotaLimit({
+            agentId: agentIdRef.current,
+            currentHandle: currentModelId,
+            detail: detailFromRun ?? fallbackError
+          });
+          if (rotation) {
+            chatgptPlanSwapsRef.current += 1;
+            const statusId = uid("status");
+            buffersRef.current.byId.set(statusId, {
+              kind: "status",
+              id: statusId,
+              lines: [formatPlanRotationNotice(rotation)]
+            });
+            buffersRef.current.order.push(statusId);
+            refreshDerived();
+            currentInput = refreshInputOtidsForNewRequest(currentInput);
+            buffersRef.current.interrupted = false;
+            continue;
+          }
+        }
         const autoSwapOnQuotaLimitEnabled = settingsManager.getSetting("autoSwapOnQuotaLimit") !== false;
         const supportsHostedAutoQuotaFallback = !getBackend().capabilities.localModelCatalog;
         const isQuotaLimit = isQuotaLimitErrorDetail(detailFromRun ?? fallbackError);
@@ -549012,6 +549434,7 @@ var init_use_conversation_loop = __esm(async () => {
   init_error();
   init_approval_recovery();
   init_available_models();
+  init_chatgpt_plan_rotation();
   init_check_approval();
   init_model();
   init_prompt_assets();
@@ -557857,6 +558280,7 @@ function App2({
   const llmApiErrorRetriesRef = import_react121.useRef(0);
   const quotaAutoSwapAttemptedRef = import_react121.useRef(false);
   const emptyResponseRetriesRef = import_react121.useRef(0);
+  const chatgptPlanSwapsRef = import_react121.useRef(0);
   const conversationBusyRetriesRef = import_react121.useRef(0);
   const [queueDisplay, setQueueDisplay] = import_react121.useState([]);
   const tuiQueueRef = import_react121.useRef(null);
@@ -559345,6 +559769,7 @@ Memory may be stale. Try running: git -C ${getScopedMemoryFilesystemRoot(agentId
     autoAllowedExecutionRef,
     buffersRef,
     clearApprovalToolContext,
+    chatgptPlanSwapsRef,
     closeTrajectorySegment,
     consumeQueuedMessages,
     queueModeRef,
@@ -564456,4 +564881,4 @@ function registerBunOAuthFlows() {
 registerBunOAuthFlows();
 await init_src5().then(() => exports_src2);
 
-//# debugId=BBC1C9C009AB28B764756E2164756E21
+//# debugId=66D224FF7F960BF164756E2164756E21