@stigmer/runner 3.1.2 → 3.1.4

package/dist/shared/plan-mode-prompt.d.ts

@@ -3,8 +3,9 @@
  *
  * Plan mode's output contract spans three components that must agree:
  * - the model produces the plan as its FINAL message (this directive),
- * - the runner publishes that final message verbatim as a `plan.md` artifact
- *   (`plan-artifact.ts` — `extractFinalPlanText`),
+ * - the runner publishes that final message verbatim as a plan markdown
+ *   artifact, named from the plan's title (`plan-artifact.ts` —
+ *   `extractFinalPlanText`, `planArtifactName`),
  * - the SDK promotes the same message to a first-class plan document in the
  *   thread and offers "Build from plan".
  *
@@ -24,7 +25,17 @@
 /**
  * Directive body injected into the system prompt of every Plan-mode
  * execution. Deliberately explicit about the deliverable's shape: the final
- * message is published verbatim as `plan.md`, so a fenced or chat-suffixed
- * plan degrades the reviewable document the user sees.
+ * message is published verbatim as a plan document (whose filename is derived
+ * from the leading `#` title), so a fenced or chat-suffixed plan degrades the
+ * reviewable document the user sees.
+ *
+ * Fence hygiene inside the document is part of the same contract. A plan
+ * frequently quotes file content that itself contains fenced code blocks
+ * ("insert this section into README.md"); with a same-length outer fence the
+ * inner block's closer terminates the outer fence early and corrupts the
+ * rendered document — on every client (the react SDK and the terminal
+ * renderer alike). And because the plan viewers render top-level ```mermaid
+ * fences as diagrams, the directive steers diagrams into the plan body proper
+ * rather than leaving them buried, unrendered, inside quoted file content.
  */
 export declare const PLAN_MODE_DIRECTIVE: string;

package/dist/shared/plan-mode-prompt.js

@@ -3,8 +3,9 @@
  *
  * Plan mode's output contract spans three components that must agree:
  * - the model produces the plan as its FINAL message (this directive),
- * - the runner publishes that final message verbatim as a `plan.md` artifact
- *   (`plan-artifact.ts` — `extractFinalPlanText`),
+ * - the runner publishes that final message verbatim as a plan markdown
+ *   artifact, named from the plan's title (`plan-artifact.ts` —
+ *   `extractFinalPlanText`, `planArtifactName`),
  * - the SDK promotes the same message to a first-class plan document in the
  *   thread and offers "Build from plan".
  *
@@ -24,8 +25,18 @@
 /**
  * Directive body injected into the system prompt of every Plan-mode
  * execution. Deliberately explicit about the deliverable's shape: the final
- * message is published verbatim as `plan.md`, so a fenced or chat-suffixed
- * plan degrades the reviewable document the user sees.
+ * message is published verbatim as a plan document (whose filename is derived
+ * from the leading `#` title), so a fenced or chat-suffixed plan degrades the
+ * reviewable document the user sees.
+ *
+ * Fence hygiene inside the document is part of the same contract. A plan
+ * frequently quotes file content that itself contains fenced code blocks
+ * ("insert this section into README.md"); with a same-length outer fence the
+ * inner block's closer terminates the outer fence early and corrupts the
+ * rendered document — on every client (the react SDK and the terminal
+ * renderer alike). And because the plan viewers render top-level ```mermaid
+ * fences as diagrams, the directive steers diagrams into the plan body proper
+ * rather than leaving them buried, unrendered, inside quoted file content.
  */
 export const PLAN_MODE_DIRECTIVE = [
     "IMPORTANT: You are in Plan mode — a read-only analysis turn whose " +
@@ -37,13 +48,25 @@ export const PLAN_MODE_DIRECTIVE = [
     "- Only read, search, and analyze.",
     "",
     "Deliverable — your FINAL message IS the plan. It is published verbatim " +
-        "as a `plan.md` document that the user reviews and builds from, so:",
+        "as a plan document that the user reviews and builds from, so:",
     "- Write it as a complete, well-structured markdown document: start with " +
         "a single `#` title and organize the work under `##` section headings. " +
         "Use lists and tables where they aid scanning.",
+    '- Give the `#` title a concise, descriptive name for the work itself; do ' +
+        'NOT prefix it with "Plan:" (this document is already a plan — the ' +
+        "prefix is redundant and leaks into the plan's filename).",
     "- Reference concrete file paths and describe the specific changes " +
         "planned for each.",
     "- Do NOT wrap the document in a code fence.",
+    "- When quoting content that itself contains fenced code blocks (e.g. a " +
+        "proposed file section with a code sample inside), open the outer fence " +
+        "with MORE backticks than any inner fence (four or more) — a same-length " +
+        "inner closer would terminate the outer fence early and corrupt the " +
+        "rendered document.",
+    "- Fenced ```mermaid blocks at the top level of the document render as " +
+        "diagrams in the plan viewer. When a diagram helps communicate the " +
+        "design (architecture, flows), include it directly in the plan body — " +
+        "not only inside quoted file content, where it stays unrendered source.",
     '- Do NOT end with conversational closers ("Let me know...", "Shall I ' +
         'proceed?") — the next step is the user\'s Build action, and trailing ' +
         "chat would be published as part of the document.",

package/dist/shared/plan-mode-prompt.js.map

@@ -1 +1 @@
-{"version":3,"file":"plan-mode-prompt.js","sourceRoot":"","sources":["../../src/shared/plan-mode-prompt.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH;;;;;GAKG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG;IACjC,oEAAoE;QAClE,wCAAwC;IAC1C,EAAE;IACF,cAAc;IACd,6CAA6C;IAC7C,yEAAyE;IACzE,mCAAmC;IACnC,EAAE;IACF,yEAAyE;QACvE,oEAAoE;IACtE,0EAA0E;QACxE,wEAAwE;QACxE,+CAA+C;IACjD,oEAAoE;QAClE,mBAAmB;IACrB,6CAA6C;IAC7C,uEAAuE;QACrE,uEAAuE;QACvE,kDAAkD;CACrD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC"}
+{"version":3,"file":"plan-mode-prompt.js","sourceRoot":"","sources":["../../src/shared/plan-mode-prompt.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH;;;;;;;;;;;;;;;GAeG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG;IACjC,oEAAoE;QAClE,wCAAwC;IAC1C,EAAE;IACF,cAAc;IACd,6CAA6C;IAC7C,yEAAyE;IACzE,mCAAmC;IACnC,EAAE;IACF,yEAAyE;QACvE,+DAA+D;IACjE,0EAA0E;QACxE,wEAAwE;QACxE,+CAA+C;IACjD,2EAA2E;QACzE,oEAAoE;QACpE,0DAA0D;IAC5D,oEAAoE;QAClE,mBAAmB;IACrB,6CAA6C;IAC7C,yEAAyE;QACvE,yEAAyE;QACzE,0EAA0E;QAC1E,qEAAqE;QACrE,oBAAoB;IACtB,wEAAwE;QACtE,oEAAoE;QACpE,uEAAuE;QACvE,wEAAwE;IAC1E,uEAAuE;QACrE,uEAAuE;QACvE,kDAAkD;CACrD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC"}

package/dist/shared/todos.d.ts

@@ -0,0 +1,56 @@
+/**
+ * Todo-list projection shared by every harness's status writer.
+ *
+ * All harnesses expose an agent to-do tool — the Cursor SDK emits `TodoWrite`
+ * (legacy) / `updateTodos` (current), the native deepagents harness emits
+ * `write_todos` — and each writes the SAME `AgentExecutionStatus.todos` proto
+ * map that the clients (React `TodoCard`, CLI) render. This module is the single
+ * place that maps a raw tool payload into that map, so the Cursor `TodoTracker`
+ * and the native v2/v3 status builders cannot drift in how they build todos.
+ *
+ * The mapping is a superset of what any one harness emits: the Cursor SDK sends
+ * per-item `id`/`created_at` and can send a `merge` flag and camelCase statuses;
+ * deepagents sends only `{ content, status }` (three statuses, no id, no merge).
+ * One helper handling the superset serves all three writers.
+ */
+import type { TodoItem } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/todo_pb";
+/**
+ * A single raw todo as it arrives in a to-do tool call's arguments. Every field
+ * is optional and defensively typed: the Cursor `updateTodos` schema omits
+ * per-item `id`, deepagents omits `id`/`created_at`, and the values reach us as
+ * untyped JSON (a Cursor `args` string or a protobuf `JsonObject` value).
+ */
+export interface RawTodoItem {
+    id?: string;
+    content?: string;
+    status?: string;
+    created_at?: string;
+}
+/**
+ * Project a to-do tool call's payload into a `status.todos` map, in place.
+ * Returns whether the map changed — the caller's "dirty" / force-persist signal.
+ *
+ * Semantics (identical across all three harness writers):
+ *   - Empty or non-array `rawTodos`: a full-replace (`!merge`) clears the map and
+ *     is a change; a merge is a no-op (nothing to merge in) and is not.
+ *   - Otherwise a full-replace clears first, then every item is (re)written.
+ *   - `id` is the provided id or a stable index fallback (`todo-<i>`); order is
+ *     the array order, so index ids are stable across full-replace updates.
+ *   - `created_at` is preserved from the prior same-id entry ONLY on merge; a
+ *     full-replace clears first, so it becomes `raw.created_at || now`. (The
+ *     native harness is always full-replace and sends no `created_at`, so a
+ *     native todo's `created_at` is `now` on each write — deliberately identical
+ *     to the Cursor non-merge path.)
+ *   - `updated_at` is always `now`.
+ *
+ * Note on clearing: an empty full-replace clears the in-memory map, but the Go
+ * `update_status` activity replaces persisted todos only when the incoming map
+ * is non-empty, so a settled list survives in history — the clear never reaches
+ * the record. This is intentional and relied upon by both harnesses.
+ */
+export declare function applyTodoUpdate(target: {
+    [key: string]: TodoItem;
+}, rawTodos: unknown, opts: {
+    merge: boolean;
+    now?: string;
+}): boolean;

package/dist/shared/todos.js

@@ -0,0 +1,98 @@
+/**
+ * Todo-list projection shared by every harness's status writer.
+ *
+ * All harnesses expose an agent to-do tool — the Cursor SDK emits `TodoWrite`
+ * (legacy) / `updateTodos` (current), the native deepagents harness emits
+ * `write_todos` — and each writes the SAME `AgentExecutionStatus.todos` proto
+ * map that the clients (React `TodoCard`, CLI) render. This module is the single
+ * place that maps a raw tool payload into that map, so the Cursor `TodoTracker`
+ * and the native v2/v3 status builders cannot drift in how they build todos.
+ *
+ * The mapping is a superset of what any one harness emits: the Cursor SDK sends
+ * per-item `id`/`created_at` and can send a `merge` flag and camelCase statuses;
+ * deepagents sends only `{ content, status }` (three statuses, no id, no merge).
+ * One helper handling the superset serves all three writers.
+ */
+import { create } from "@bufbuild/protobuf";
+import { TodoItemSchema } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/todo_pb";
+import { TodoStatus } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
+import { utcTimestamp } from "./status.js";
+/**
+ * Raw status string -> proto `TodoStatus`. Keyed on the lowercased status so
+ * the current SDK's camelCase `inProgress` and the legacy snake_case
+ * `in_progress` both resolve. `cancelled` is only ever emitted by the Cursor
+ * SDK — deepagents' three-status schema never sends it — but it is kept here so
+ * the one shared mapping is faithful to every producer. Unknown values fall back
+ * to `TODO_PENDING` at the call site.
+ */
+const STATUS_MAP = {
+    pending: TodoStatus.TODO_PENDING,
+    in_progress: TodoStatus.TODO_IN_PROGRESS,
+    inprogress: TodoStatus.TODO_IN_PROGRESS,
+    completed: TodoStatus.TODO_COMPLETED,
+    cancelled: TodoStatus.TODO_CANCELLED,
+};
+/**
+ * Project a to-do tool call's payload into a `status.todos` map, in place.
+ * Returns whether the map changed — the caller's "dirty" / force-persist signal.
+ *
+ * Semantics (identical across all three harness writers):
+ *   - Empty or non-array `rawTodos`: a full-replace (`!merge`) clears the map and
+ *     is a change; a merge is a no-op (nothing to merge in) and is not.
+ *   - Otherwise a full-replace clears first, then every item is (re)written.
+ *   - `id` is the provided id or a stable index fallback (`todo-<i>`); order is
+ *     the array order, so index ids are stable across full-replace updates.
+ *   - `created_at` is preserved from the prior same-id entry ONLY on merge; a
+ *     full-replace clears first, so it becomes `raw.created_at || now`. (The
+ *     native harness is always full-replace and sends no `created_at`, so a
+ *     native todo's `created_at` is `now` on each write — deliberately identical
+ *     to the Cursor non-merge path.)
+ *   - `updated_at` is always `now`.
+ *
+ * Note on clearing: an empty full-replace clears the in-memory map, but the Go
+ * `update_status` activity replaces persisted todos only when the incoming map
+ * is non-empty, so a settled list survives in history — the clear never reaches
+ * the record. This is intentional and relied upon by both harnesses.
+ */
+export function applyTodoUpdate(target, rawTodos, opts) {
+    const { merge } = opts;
+    const now = opts.now ?? utcTimestamp();
+    if (!Array.isArray(rawTodos) || rawTodos.length === 0) {
+        if (!merge) {
+            clearMap(target);
+            return true;
+        }
+        return false;
+    }
+    if (!merge) {
+        clearMap(target);
+    }
+    for (let i = 0; i < rawTodos.length; i++) {
+        const raw = coerceRawTodo(rawTodos[i]);
+        const id = raw.id || `todo-${i}`;
+        const statusStr = (raw.status ?? "pending").toLowerCase();
+        const status = STATUS_MAP[statusStr] ?? TodoStatus.TODO_PENDING;
+        // Preserve the original creation time only when merging into an existing
+        // entry; a full-replace cleared the map above, so `existing` is undefined.
+        const existing = merge ? target[id] : undefined;
+        target[id] = create(TodoItemSchema, {
+            id,
+            content: raw.content ?? "",
+            status,
+            createdAt: existing?.createdAt || raw.created_at || now,
+            updatedAt: now,
+        });
+    }
+    return true;
+}
+/** Narrow an untyped array element to a {@link RawTodoItem}; non-objects become empty. */
+function coerceRawTodo(value) {
+    return typeof value === "object" && value !== null ? value : {};
+}
+/** Delete every key from the map in place (proto map fields have no `.clear()`). */
+function clearMap(target) {
+    for (const key of Object.keys(target)) {
+        delete target[key];
+    }
+}
+//# sourceMappingURL=todos.js.map

package/dist/shared/todos.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"todos.js","sourceRoot":"","sources":["../../src/shared/todos.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AAC5C,OAAO,EAAE,cAAc,EAAE,MAAM,8DAA8D,CAAC;AAE9F,OAAO,EAAE,UAAU,EAAE,MAAM,8DAA8D,CAAC;AAC1F,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAe3C;;;;;;;GAOG;AACH,MAAM,UAAU,GAA+B;IAC7C,OAAO,EAAE,UAAU,CAAC,YAAY;IAChC,WAAW,EAAE,UAAU,CAAC,gBAAgB;IACxC,UAAU,EAAE,UAAU,CAAC,gBAAgB;IACvC,SAAS,EAAE,UAAU,CAAC,cAAc;IACpC,SAAS,EAAE,UAAU,CAAC,cAAc;CACrC,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,UAAU,eAAe,CAC7B,MAAmC,EACnC,QAAiB,EACjB,IAAsC;IAEtC,MAAM,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC;IACvB,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,YAAY,EAAE,CAAC;IAEvC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,QAAQ,CAAC,MAAM,CAAC,CAAC;YACjB,OAAO,IAAI,CAAC;QACd,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IAED,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,QAAQ,CAAC,MAAM,CAAC,CAAC;IACnB,CAAC;IAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACzC,MAAM,GAAG,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,EAAE,GAAG,GAAG,CAAC,EAAE,IAAI,QAAQ,CAAC,EAAE,CAAC;QACjC,MAAM,SAAS,GAAG,CAAC,GAAG,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;QAC1D,MAAM,MAAM,GAAG,UAAU,CAAC,SAAS,CAAC,IAAI,UAAU,CAAC,YAAY,CAAC;QAEhE,yEAAyE;QACzE,2EAA2E;QAC3E,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAEhD,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,cAAc,EAAE;YAClC,EAAE;YACF,OAAO,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE;YAC1B,MAAM;YACN,SAAS,EAAE,QAAQ,EAAE,SAAS,IAAI,GAAG,CAAC,UAAU,IAAI,GAAG;YACvD,SAAS,EAAE,GAAG;SACf,CAAC,CAAC;IACL,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED,0FAA0F;AAC1F,SAAS,aAAa,CAAC,KAAc;IACnC,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,CAAC,CAAC,CAAE,KAAqB,CAAC,CAAC,CAAC,EAAE,CAAC;AACnF,CAAC;AAED,oFAAoF;AACpF,SAAS,QAAQ,CAAC,MAAmC;IACnD,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;QACtC,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;AACH,CAAC"}

package/dist/shared/tool-row.d.ts

@@ -115,3 +115,19 @@ export declare function isToolCallRowHidden(tc: ToolCall): boolean;
  * none; Cursor clones them in) — the snapshot reflects whatever is present.
  */
 export declare function collectSubAgentToolCallIds(subAgents: readonly SubAgentExecution[]): Set<string>;
+/**
+ * Collect the ids of tool-call rows that have already SETTLED (reached a terminal
+ * state: completed, failed, or skipped) in a transcript.
+ *
+ * The deep-agent turn-boundary provenance derivation (DD-28) snapshots this
+ * BEFORE a turn's stream to scope "this turn's tool calls" by identity: a call
+ * whose id is absent from the snapshot is this-turn's — either freshly streamed,
+ * or a prior gate that was WAITING_APPROVAL before the stream and executes now on
+ * resume (StatusBuilder updates the seeded row in place, so it keeps its id).
+ * Unlike the Cursor harness, the deep-agent cannot scope positionally — an
+ * approved command executes at its SEEDED position — so identity is the only
+ * correct scope, mirroring {@link collectSubAgentToolCallIds}. Terminal (not just
+ * completed) so a PRIOR turn's failed/skipped non-shell call is excluded and
+ * cannot spuriously disqualify this turn.
+ */
+export declare function collectSettledToolCallIds(messages: readonly AgentMessage[]): Set<string>;

package/dist/shared/tool-row.js

@@ -173,4 +173,35 @@ export function collectSubAgentToolCallIds(subAgents) {
     }
     return ids;
 }
+/** Terminal tool-call statuses — a row that has finished (however it finished). */
+const TERMINAL_TOOL_CALL_STATUSES = new Set([
+    ToolCallStatus.TOOL_CALL_COMPLETED,
+    ToolCallStatus.TOOL_CALL_FAILED,
+    ToolCallStatus.TOOL_CALL_SKIPPED,
+]);
+/**
+ * Collect the ids of tool-call rows that have already SETTLED (reached a terminal
+ * state: completed, failed, or skipped) in a transcript.
+ *
+ * The deep-agent turn-boundary provenance derivation (DD-28) snapshots this
+ * BEFORE a turn's stream to scope "this turn's tool calls" by identity: a call
+ * whose id is absent from the snapshot is this-turn's — either freshly streamed,
+ * or a prior gate that was WAITING_APPROVAL before the stream and executes now on
+ * resume (StatusBuilder updates the seeded row in place, so it keeps its id).
+ * Unlike the Cursor harness, the deep-agent cannot scope positionally — an
+ * approved command executes at its SEEDED position — so identity is the only
+ * correct scope, mirroring {@link collectSubAgentToolCallIds}. Terminal (not just
+ * completed) so a PRIOR turn's failed/skipped non-shell call is excluded and
+ * cannot spuriously disqualify this turn.
+ */
+export function collectSettledToolCallIds(messages) {
+    const ids = new Set();
+    for (const msg of messages) {
+        for (const tc of msg.toolCalls) {
+            if (TERMINAL_TOOL_CALL_STATUSES.has(tc.status))
+                ids.add(tc.id);
+        }
+    }
+    return ids;
+}
 //# sourceMappingURL=tool-row.js.map

package/dist/shared/tool-row.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-row.js","sourceRoot":"","sources":["../../src/shared/tool-row.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,8DAA8D,CAAC;AAG9F,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,8BAA8B,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AACtD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAE3C;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAY,EAAE,WAAmB;IAChE,IAAI,EAAE,CAAC,eAAe;QAAE,OAAO;IAC/B,EAAE,CAAC,eAAe,GAAG,WAAW,CAAC;IAEjC,IAAI,yBAAyB,CAAC,EAAE,CAAC,EAAE,CAAC;QAClC,0EAA0E;QAC1E,2EAA2E;QAC3E,0EAA0E;QAC1E,0EAA0E;QAC1E,EAAE,CAAC,MAAM,GAAG,EAAE,CAAC;IACjB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,yBAAyB,CAAC,EAAY;IACpD,MAAM,IAAI,GAAG,eAAe,CAAC,CAAC,EAAE,CAAC,IAAI,IAAI,EAAE,CAA4B,CAAC,IAAI,EAAE,CAAC;IAC/E,IAAI,CAAC,gBAAgB,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC1C,EAAE,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IACtC,EAAE,CAAC,WAAW,GAAG,EAAE,CAAC;IACpB,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,iCAAiC,CAC/C,QAAiC,EACjC,SAAwC;IAExC,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,SAAS,EAAE,CAAC;YAC/B,IAAI,oBAAoB,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK,OAAO,EAAE,CAAC;gBAC9C,yBAAyB,CAAC,EAAE,CAAC,CAAC;YAChC,CAAC;QACH,CAAC;IACH,CAAC;IACD,KAAK,MAAM,EAAE,IAAI,SAAS,IAAI,EAAE,EAAE,CAAC;QACjC,iCAAiC,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;IACjD,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,eAAe,CAAC,EAAY;IAC1C,EAAE,CAAC,MAAM,GAAG,cAAc,CAAC,iBAAiB,CAAC;IAC7C,EAAE,CAAC,gBAAgB,GAAG,KAAK,CAAC;IAC5B,EAAE,CAAC,mBAAmB,GAAG,EAAE,CAAC;IAC5B,EAAE,CAAC,eAAe,GAAG,EAAE,CAAC;IACxB,EAAE,CAAC,KAAK,GAAG,EAAE,CAAC;IACd,EAAE,CAAC,MAAM,GAAG,EAAE,CAAC;IACf,EAAE,CAAC,WAAW,GAAG,EAAE,CAAC;IACpB,EAAE,CAAC,IAAI,GAAG,SAAS,CAAC;IACpB,IAAI,CAAC,EAAE,CAAC,WAAW;QAAE,EAAE,CAAC,WAAW,GAAG,YAAY,EAAE,CAAC;AACvD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CAAC,EAAY;IAC9C,OAAO,CACL,EAAE,CAAC,MAAM,KAAK,cAAc,CAAC,iBAAiB;QAC9C,CAAC,EAAE,CAAC,gBAAgB;QACpB,EAAE,CAAC,IAAI,KAAK,SAAS;QACrB,CAAC,EAAE,CAAC,MAAM;QACV,CAAC,EAAE,CAAC,KAAK,CACV,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,UAAU,0BAA0B,CACxC,SAAuC;IAEvC,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAC;IAC9B,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,KAAK,MAAM,GAAG,IAAI,EAAE,CAAC,QAAQ,EAAE,CAAC;YAC9B,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,SAAS;gBAAE,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;QACjD,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}
+{"version":3,"file":"tool-row.js","sourceRoot":"","sources":["../../src/shared/tool-row.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,8DAA8D,CAAC;AAG9F,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,8BAA8B,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AACtD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAE3C;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAY,EAAE,WAAmB;IAChE,IAAI,EAAE,CAAC,eAAe;QAAE,OAAO;IAC/B,EAAE,CAAC,eAAe,GAAG,WAAW,CAAC;IAEjC,IAAI,yBAAyB,CAAC,EAAE,CAAC,EAAE,CAAC;QAClC,0EAA0E;QAC1E,2EAA2E;QAC3E,0EAA0E;QAC1E,0EAA0E;QAC1E,EAAE,CAAC,MAAM,GAAG,EAAE,CAAC;IACjB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,yBAAyB,CAAC,EAAY;IACpD,MAAM,IAAI,GAAG,eAAe,CAAC,CAAC,EAAE,CAAC,IAAI,IAAI,EAAE,CAA4B,CAAC,IAAI,EAAE,CAAC;IAC/E,IAAI,CAAC,gBAAgB,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC1C,EAAE,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IACtC,EAAE,CAAC,WAAW,GAAG,EAAE,CAAC;IACpB,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,iCAAiC,CAC/C,QAAiC,EACjC,SAAwC;IAExC,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,SAAS,EAAE,CAAC;YAC/B,IAAI,oBAAoB,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK,OAAO,EAAE,CAAC;gBAC9C,yBAAyB,CAAC,EAAE,CAAC,CAAC;YAChC,CAAC;QACH,CAAC;IACH,CAAC;IACD,KAAK,MAAM,EAAE,IAAI,SAAS,IAAI,EAAE,EAAE,CAAC;QACjC,iCAAiC,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;IACjD,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,eAAe,CAAC,EAAY;IAC1C,EAAE,CAAC,MAAM,GAAG,cAAc,CAAC,iBAAiB,CAAC;IAC7C,EAAE,CAAC,gBAAgB,GAAG,KAAK,CAAC;IAC5B,EAAE,CAAC,mBAAmB,GAAG,EAAE,CAAC;IAC5B,EAAE,CAAC,eAAe,GAAG,EAAE,CAAC;IACxB,EAAE,CAAC,KAAK,GAAG,EAAE,CAAC;IACd,EAAE,CAAC,MAAM,GAAG,EAAE,CAAC;IACf,EAAE,CAAC,WAAW,GAAG,EAAE,CAAC;IACpB,EAAE,CAAC,IAAI,GAAG,SAAS,CAAC;IACpB,IAAI,CAAC,EAAE,CAAC,WAAW;QAAE,EAAE,CAAC,WAAW,GAAG,YAAY,EAAE,CAAC;AACvD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CAAC,EAAY;IAC9C,OAAO,CACL,EAAE,CAAC,MAAM,KAAK,cAAc,CAAC,iBAAiB;QAC9C,CAAC,EAAE,CAAC,gBAAgB;QACpB,EAAE,CAAC,IAAI,KAAK,SAAS;QACrB,CAAC,EAAE,CAAC,MAAM;QACV,CAAC,EAAE,CAAC,KAAK,CACV,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,UAAU,0BAA0B,CACxC,SAAuC;IAEvC,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAC;IAC9B,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,KAAK,MAAM,GAAG,IAAI,EAAE,CAAC,QAAQ,EAAE,CAAC;YAC9B,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,SAAS;gBAAE,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;QACjD,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,mFAAmF;AACnF,MAAM,2BAA2B,GAAgC,IAAI,GAAG,CAAC;IACvE,cAAc,CAAC,mBAAmB;IAClC,cAAc,CAAC,gBAAgB;IAC/B,cAAc,CAAC,iBAAiB;CACjC,CAAC,CAAC;AAEH;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,yBAAyB,CACvC,QAAiC;IAEjC,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAC;IAC9B,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,SAAS,EAAE,CAAC;YAC/B,IAAI,2BAA2B,CAAC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC;gBAAE,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;QACjE,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stigmer/runner",
-  "version": "3.1.2",
+  "version": "3.1.4",
   "description": "Embeddable Temporal worker for the Stigmer AI agent platform — handles agent execution, workflow orchestration, and MCP server management",
   "license": "Apache-2.0",
   "type": "module",
@@ -85,7 +85,7 @@
     "@opentelemetry/sdk-metrics": "^2.0.0",
     "@opentelemetry/sdk-trace-base": "^2.0.0",
     "@opentelemetry/sdk-trace-node": "^2.0.0",
-    "@stigmer/protos": "3.1.2",
+    "@stigmer/protos": "3.1.4",
     "@temporalio/activity": "^1.11.0",
     "@temporalio/client": "^1.11.0",
     "@temporalio/common": "^1.11.0",
@@ -93,6 +93,7 @@
     "@temporalio/worker": "^1.11.0",
     "@temporalio/workflow": "^1.16.2",
     "deepagents": "^1.10.0",
+    "diff": "^8.0.4",
     "jq-wasm": "^1.1.0-jq-1.8.1",
     "js-yaml": "^4.1.1",
     "proper-lockfile": "^4.1.2",

package/src/__tests__/otel-turn-span.test.ts

@@ -0,0 +1,61 @@
+import { describe, it, expect, vi, beforeEach } from "vitest";
+
+// Spy span/tracer, hoisted so the vi.mock factory (which runs before imports)
+// can close over them without a temporal-dead-zone error.
+const { spanEnd, spanSetAttribute, startSpan } = vi.hoisted(() => {
+  const spanEnd = vi.fn();
+  const spanSetAttribute = vi.fn();
+  const startSpan = vi.fn(() => ({ end: spanEnd, setAttribute: spanSetAttribute }));
+  return { spanEnd, spanSetAttribute, startSpan };
+});
+
+vi.mock("@opentelemetry/api", () => ({
+  trace: { getTracer: () => ({ startSpan }) },
+}));
+
+import {
+  startCursorTurnSpan,
+  ATTR_LLM_INPUT_TOKENS,
+  ATTR_LLM_OUTPUT_TOKENS,
+} from "../otel.js";
+
+const spanAttrs = { model: "gpt-x", mode: "local", sessionId: "sess-1" };
+
+describe("startCursorTurnSpan", () => {
+  beforeEach(() => {
+    spanEnd.mockClear();
+    spanSetAttribute.mockClear();
+    startSpan.mockClear();
+  });
+
+  it("records the input and output token attributes", async () => {
+    const span = await startCursorTurnSpan(spanAttrs);
+    span.setTokens(1200, 340);
+    expect(spanSetAttribute).toHaveBeenCalledWith(ATTR_LLM_INPUT_TOKENS, 1200);
+    expect(spanSetAttribute).toHaveBeenCalledWith(ATTR_LLM_OUTPUT_TOKENS, 340);
+  });
+
+  it("omits zero-valued token attributes (no count = no attribute)", async () => {
+    const span = await startCursorTurnSpan(spanAttrs);
+    span.setTokens(0, 0);
+    expect(spanSetAttribute).not.toHaveBeenCalled();
+  });
+
+  it("ends the span exactly once even when end() is called repeatedly", async () => {
+    // finishTurnTelemetry is invoked from the activity's finally on every exit
+    // path; a second end() (e.g. a defensive double-invoke) must be a no-op so
+    // OTel never warns about ending an already-ended span.
+    const span = await startCursorTurnSpan(spanAttrs);
+    span.end();
+    span.end();
+    span.end();
+    expect(spanEnd).toHaveBeenCalledTimes(1);
+  });
+
+  it("ignores setTokens after the span has ended (no late mutation)", async () => {
+    const span = await startCursorTurnSpan(spanAttrs);
+    span.end();
+    span.setTokens(1200, 340);
+    expect(spanSetAttribute).not.toHaveBeenCalled();
+  });
+});

package/src/activities/execute-cursor/__tests__/build-prompt.test.ts

@@ -332,6 +332,13 @@ describe("formatImplementPlanSection", () => {
     expect(formatImplementPlanSection(undefined, [PLAN_PATH])).toBeUndefined();
   });
 
+  it("carries the plan-derived progress-tracking instruction (Tier 3)", () => {
+    const section = formatImplementPlanSection(true, [PLAN_PATH]);
+
+    expect(section).toContain("to-do list");
+    expect(section).toContain("break the plan into");
+  });
+
   it("injects the directive into a build-from-plan first prompt", () => {
     const prompt = buildPrompt(
       input({

package/src/activities/execute-cursor/__tests__/delta-enricher.test.ts

@@ -240,3 +240,71 @@ describe("DeltaEnricher finalize reconciliation", () => {
     expect(messages[1].toolCalls[0].status).toBe(ToolCallStatus.TOOL_CALL_COMPLETED);
   });
 });
+
+describe("DeltaEnricher recovery-retry stream epilogue", () => {
+  // Regression for the zombie RUNNING row of aex_01kws27q1e2esvkqjpvectttxf:
+  // the recovery retry's bare stream loop buffered the tool-call-completed
+  // delta but never flushed it, so the row lacked the completedAt evidence
+  // finalize's sweep requires and persisted as RUNNING forever. The epilogue
+  // must applyEnrichments (flush the buffered evidence) BEFORE finalize.
+  it("applyEnrichments + finalize promotes a retry's buffered-completion RUNNING row", () => {
+    vi.spyOn(console, "log").mockImplementation(() => {});
+
+    const enricher = new DeltaEnricher();
+    const messages: AgentMessage[] = [
+      makeMessage([{ id: "tc-retry", name: "Shell", status: ToolCallStatus.TOOL_CALL_RUNNING }]),
+    ];
+
+    // The retry's onDelta buffered the completion, but the bare retry loop
+    // never applied it to the transcript.
+    enricher.processDelta({
+      type: "tool-call-completed",
+      callId: "tc-retry",
+      toolCall: { type: "shell" },
+    } as unknown as InteractionUpdate);
+
+    // Without the flush, finalize alone cannot promote (no evidence on the row).
+    const unflushed = makeMessage([
+      { id: "tc-retry", name: "Shell", status: ToolCallStatus.TOOL_CALL_RUNNING },
+    ]);
+    new DeltaEnricher().finalize([unflushed]);
+    expect(unflushed.toolCalls[0].status).toBe(ToolCallStatus.TOOL_CALL_RUNNING);
+
+    // The epilogue sequence: flush the buffered evidence, then finalize.
+    enricher.applyEnrichments(messages);
+    enricher.finalize(messages);
+
+    const tc = messages[0].toolCalls[0];
+    expect(tc.status).toBe(ToolCallStatus.TOOL_CALL_COMPLETED);
+    expect(tc.completedAt).toBeTruthy();
+  });
+
+  it("re-running applyEnrichments and finalize is idempotent", () => {
+    vi.spyOn(console, "log").mockImplementation(() => {});
+
+    const enricher = new DeltaEnricher();
+    const messages: AgentMessage[] = [
+      makeMessage([{ id: "tc-1", name: "Shell", status: ToolCallStatus.TOOL_CALL_RUNNING }]),
+    ];
+
+    enricher.processDelta({
+      type: "tool-call-completed",
+      callId: "tc-1",
+      toolCall: { type: "shell" },
+    } as unknown as InteractionUpdate);
+
+    // Primary epilogue ...
+    enricher.applyEnrichments(messages);
+    enricher.finalize(messages);
+    const stamped = messages[0].toolCalls[0].completedAt;
+
+    // ... and the retry epilogue re-runs both. The row must keep its original
+    // completion evidence and terminal status.
+    enricher.applyEnrichments(messages);
+    enricher.finalize(messages);
+
+    const tc = messages[0].toolCalls[0];
+    expect(tc.status).toBe(ToolCallStatus.TOOL_CALL_COMPLETED);
+    expect(tc.completedAt).toBe(stamped);
+  });
+});