@oxagen/cli 2.1.4-494 → 2.1.4-498

package/oxagen.mjs

@@ -4839,6 +4839,16 @@ var init_envelope = __esm({
       ]).optional(),
       tool_denial_kind: short.optional(),
       tool_result_tokens: u32.optional(),
+      /**
+       * How `tool_result_tokens` is known (#5339). `reported` is Claude Code's own
+       * count on its OTel tool span. `estimated` is the recorder's: the result's
+       * UTF-8 bytes over four, from the hook that saw the result. A count with no
+       * basis was written before the member existed, and only the OTel span wrote
+       * a count then, so a reader takes it as `reported`. The recorder writes the
+       * two together. The envelope does not require the pair, so a frame from a
+       * host older than the member still parses.
+       */
+      tool_result_tokens_basis: external_exports.enum(TOKEN_SOURCE_BASES).optional(),
       batch_size: u8.optional(),
       batch_index: u8.optional(),
       attribution_skill: short.optional(),
@@ -11529,7 +11539,10 @@ function normalizeHook(raw, env, options) {
         body2["tool_output_digest"] = digestJcs(
           input.tool_response
         );
-        body2["tool_output_bytes"] = jsonByteLength(input.tool_response);
+        const bytes = jsonByteLength(input.tool_response);
+        body2["tool_output_bytes"] = bytes;
+        body2["tool_result_tokens"] = Math.ceil(bytes / 4);
+        body2["tool_result_tokens_basis"] = "estimated";
       }
       if (failed3) {
         const error = input.error ?? input["error_type"];
@@ -14513,7 +14526,24 @@ function steeringFactsOf(steering) {
     steering_tokens_basis: "estimated"
   };
 }
-var REQUEST_FULL_DIGEST_ATTR, SYSTEM_CONTEXT_PARTS_OMITTED_ATTR, MEMORY_REQUESTS, LISTED_PER_TURN, U32_MAX, NAME_MAX, tokenSourceFacts, SystemContextMemory, SHARED_SYSTEM_CONTEXT_MEMORY, CONVERSATION_FIELDS2, STEERING_ASSEMBLY_PART, SystemContextTracker;
+function contextFactsOf(tokens) {
+  if (tokens <= 0) return {};
+  return {
+    context_frame_tokens: tokens,
+    context_frame_tokens_basis: "estimated"
+  };
+}
+function injectedMarksOf(value) {
+  if (!Array.isArray(value)) return [];
+  return value.flatMap((mark) => {
+    if (typeof mark !== "object" || mark === null) return [];
+    const { at, tokens } = mark;
+    if (typeof at !== "string" || typeof tokens !== "number") return [];
+    if (!Number.isFinite(tokens)) return [];
+    return [{ at, tokens: clampU32(tokens) }];
+  });
+}
+var REQUEST_FULL_DIGEST_ATTR, SYSTEM_CONTEXT_PARTS_OMITTED_ATTR, MEMORY_REQUESTS, LISTED_PER_TURN, INJECTED_MARKS_MAX, U32_MAX, NAME_MAX, tokenSourceFacts, SystemContextMemory, SHARED_SYSTEM_CONTEXT_MEMORY, CONVERSATION_FIELDS2, STEERING_ASSEMBLY_PART, SystemContextTracker;
 var init_system_context = __esm({
   "../../packages/tacho/src/claude-code/system-context.ts"() {
     "use strict";
@@ -14525,6 +14555,7 @@ var init_system_context = __esm({
     SYSTEM_CONTEXT_PARTS_OMITTED_ATTR = "oxagen.system_context_parts_omitted";
     MEMORY_REQUESTS = 512;
     LISTED_PER_TURN = 16;
+    INJECTED_MARKS_MAX = 32;
     U32_MAX = 4294967295;
     NAME_MAX = 512;
     tokenSourceFacts = modelFacts.pick({
@@ -14581,11 +14612,16 @@ var init_system_context = __esm({
         this.steering = parts === void 0 ? void 0 : { parts, tokens: state?.steeringTokens ?? sum(parts) };
         this.listedTurn = state?.listedTurn;
         this.listed = [...state?.listed ?? []];
+        this.injected = injectedMarksOf(state?.injected);
+        const floor = state?.injectedFloor;
+        this.injectedFloor = typeof floor === "number" && Number.isFinite(floor) ? clampU32(floor) : 0;
       }
       memory;
       steering;
       listedTurn;
       listed;
+      injected;
+      injectedFloor;
       state() {
         const state = {};
         if (this.steering !== void 0) {
@@ -14594,6 +14630,9 @@ var init_system_context = __esm({
         }
         if (this.listedTurn !== void 0) state.listedTurn = this.listedTurn;
         if (this.listed.length > 0) state.listed = [...this.listed];
+        if (this.injected.length > 0)
+          state.injected = this.injected.map((mark) => ({ ...mark }));
+        if (this.injectedFloor > 0) state.injectedFloor = this.injectedFloor;
         return state;
       }
       /** Take the steering a sealed `steering.manifest` frame names. */
@@ -14601,21 +14640,84 @@ var init_system_context = __esm({
         const steering = steeringContext(body2);
         if (steering !== void 0) this.steering = steering;
       }
+      /**
+       * Note text an Oxagen hook answer handed the agent after its start, at
+       * `at`, when the answer left (#5339). The text joins the conversation, so
+       * every call made from `at` on carries it. Empty text notes nothing.
+       *
+       * The caller passes only text the session's own conversation reads. The
+       * start's text is the steering manifest's to count, and a subagent's text
+       * stays in the subagent's conversation.
+       */
+      noteInjectedContext(text4, at) {
+        const tokens = clampU32(budgetTokens(text4));
+        if (tokens === 0) return;
+        this.markInjected(at, this.injectedTotal() + tokens);
+      }
+      /**
+       * Empty the running total at `at`: a compaction or a `/clear` took the
+       * text out of the conversation. A call made before `at` still reads the
+       * total it was made with.
+       */
+      clearInjectedContext(at) {
+        if (this.injected.length === 0 && this.injectedFloor === 0) return;
+        this.markInjected(at, 0);
+      }
+      injectedTotal() {
+        return this.injected.at(-1)?.tokens ?? this.injectedFloor;
+      }
+      markInjected(at, tokens) {
+        this.injected.push({ at, tokens: clampU32(tokens) });
+        while (this.injected.length > INJECTED_MARKS_MAX) {
+          const oldest = this.injected.shift();
+          if (oldest !== void 0) this.injectedFloor = oldest.tokens;
+        }
+      }
+      /**
+       * The context Oxagen's hooks had handed the session when a call was made
+       * at `at`: the latest mark at or before it. A record can arrive after
+       * later answers left, since OTel exports in batches and the transcript is
+       * read behind the session. A call with no readable time reads the latest
+       * total. A call older than every mark kept reads the total before them.
+       */
+      injectedAt(at) {
+        const when = at === void 0 ? Number.NaN : Date.parse(at);
+        if (Number.isNaN(when)) return this.injectedTotal();
+        for (let index = this.injected.length - 1; index >= 0; index -= 1) {
+          const mark = this.injected[index];
+          if (mark === void 0) continue;
+          const markAt = Date.parse(mark.at);
+          if (Number.isNaN(markAt) || markAt <= when) return mark.tokens;
+        }
+        return this.injectedFloor;
+      }
       /**
        * Measure one model call. `content` is the proxy's exchange body, `attrs`
-       * the frame's attrs, and `turn` a key that changes when the turn does.
+       * the frame's attrs, `turn` a key that changes when the turn does, and
+       * `at` when the call was made.
        */
-      measure(content, attrs2, turn) {
+      measure(content, attrs2, turn, at) {
         const recorded = requestOf(content);
         const resolved = recorded === void 0 ? void 0 : resolveRequest(recorded.request, this.memory);
         if (recorded === void 0 || resolved === void 0)
-          return this.checked(steeringFactsOf(this.steering), {}, () => {
-          });
+          return this.checked(
+            {
+              ...steeringFactsOf(this.steering),
+              ...contextFactsOf(this.injectedAt(at))
+            },
+            {},
+            () => {
+            }
+          );
         const fullDigest = attrs2?.[REQUEST_FULL_DIGEST_ATTR];
         const key = isSha256Digest(fullDigest) ? fullDigest : recorded.request[PRIOR_MEMBER] === void 0 ? digestBytes(recorded.text) : void 0;
         if (key !== void 0) this.memory.set(key, resolved);
-        const steering = resolved.tools.length > 0 ? this.steering : void 0;
-        const facts = steeringFactsOf(steering);
+        const main3 = resolved.tools.length > 0;
+        const steering = main3 ? this.steering : void 0;
+        const facts = {
+          ...steeringFactsOf(steering),
+          ...main3 ? contextFactsOf(this.injectedAt(at)) : {}
+        };
         const parts = [
           ...resolved.system,
           ...resolved.instructions,
@@ -14648,15 +14750,23 @@ var init_system_context = __esm({
       /**
        * The token sources of a model call whose request the recorder never saw:
        * the counted OTel or transcript row of a call the proxy did not carry
-       * (#4493). Only the steering count is known without the request. The tool
-       * definition count, the context frame count, and the system context stay
-       * absent, never zero. Nothing changes on the tracker, so there is nothing
-       * to commit. The caller asks only for a call that carries the session's
-       * conversation, since a side call carries no steering.
-       */
-      measureUnseen() {
-        return this.checked(steeringFactsOf(this.steering), {}, () => {
-        }).facts;
+       * (#4493). Two counts are known without the request: the steering, from
+       * the manifest, and the context frames, from the hook answers noted before
+       * `at`, when the call was made (#5339). The tool definition count and the
+       * system context stay absent, never zero. Nothing changes on the tracker,
+       * so there is nothing to commit. The caller asks only for a call that
+       * carries the session's conversation, since a side call carries neither.
+       */
+      measureUnseen(at) {
+        return this.checked(
+          {
+            ...steeringFactsOf(this.steering),
+            ...contextFactsOf(this.injectedAt(at))
+          },
+          {},
+          () => {
+          }
+        ).facts;
       }
       /**
        * Hand back facts the envelope accepts, or none. A body the envelope
@@ -15296,7 +15406,12 @@ function spanDraft(span, resource) {
         ...s(attrs2, "tool_name") !== void 0 ? { tool_name: s(attrs2, "tool_name") } : {},
         ...s(attrs2, "tool_use_id") !== void 0 ? { tool_use_id: s(attrs2, "tool_use_id") } : {},
         ...n(attrs2, "duration_ms") !== void 0 ? { tool_duration_ms: n(attrs2, "duration_ms") } : {},
-        ...n(attrs2, "result_tokens") !== void 0 ? { tool_result_tokens: n(attrs2, "result_tokens") } : {},
+        // Claude Code's own count, so it is `reported`. A reader prefers it
+        // to the hook's estimate of the same call (#5339).
+        ...n(attrs2, "result_tokens") !== void 0 ? {
+          tool_result_tokens: n(attrs2, "result_tokens"),
+          tool_result_tokens_basis: "reported"
+        } : {},
         ...s(attrs2, "file_path") !== void 0 ? { tool_target: s(attrs2, "file_path") } : {},
         ...s(attrs2, "full_command") !== void 0 ? { tool_target: (s(attrs2, "full_command") ?? "").slice(0, 512) } : {},
         ...s(attrs2, "skill_name") !== void 0 ? { attribution_skill: s(attrs2, "skill_name") } : {},
@@ -16544,16 +16659,18 @@ var init_recorder = __esm({
         if (kind === "agent_stop") this.turnOpen = false;
         const sighting = kind === "llm_call" ? this.llmCallSighting(body2, fields.source ?? "collector") : NO_SIGHTING;
         const duplicate = sighting.attrs ?? {};
+        const ts2 = fields.ts ?? this.now();
         const measured = kind === "llm_call" ? this.systemContext.measure(
           fields.content,
           fields.attrs,
-          this.turnOpen ? `turn:${this.turnSeq}` : `idle:${this.turnSeq}`
+          this.turnOpen ? `turn:${this.turnSeq}` : `idle:${this.turnSeq}`,
+          ts2
         ) : void 0;
         const event = this.seal(
           kind,
           measured !== void 0 ? { ...measured.facts, ...body2 } : body2,
           {
-            ts: fields.ts ?? this.now(),
+            ts: ts2,
             source: fields.source ?? "collector",
             ...fields.hook_event_name !== void 0 ? { hook_event_name: fields.hook_event_name } : {},
             attrs: { ...fields.attrs, ...measured?.attrs, ...duplicate },
@@ -16967,19 +17084,32 @@ var init_recorder = __esm({
        * recorder knows without the call's request (#4493). Only the counted row,
        * the first sighting of its call, takes them. A stamped duplicate is never
        * summed, and the rollup joins a stamped proxy row back for the members the
-       * counted row lacks. The steering count is the one source known without
-       * the request, so the tool definitions and the system context stay absent
-       * (ADR-062, amendment of 2026-10-02). A member the producer set itself
-       * wins, as on a proxied call. Only a call on the session's own
-       * conversation comes here: an OTel record whose `query_source` says so
-       * ({@link onSessionConversation}), or a transcript `assistant` record,
-       * since Claude Code writes no side call to the transcript.
-       */
-      withUnseenRequestSources(body2, attrs2) {
+       * counted row lacks. Two sources are known without the request: the
+       * steering count, and the context frames Oxagen's hooks handed the session
+       * before `ts`, when the call was made (#5339). The tool definitions and the
+       * system context stay absent (ADR-062, amendments of 2026-10-02 and
+       * 2026-10-03). A member the producer set itself wins, as on a proxied call.
+       * Only a call on the session's own conversation comes here: an OTel record
+       * whose `query_source` says so ({@link onSessionConversation}), or a
+       * transcript `assistant` record, since Claude Code writes no side call to
+       * the transcript.
+       */
+      withUnseenRequestSources(body2, attrs2, ts2) {
         if (attrs2[LLM_CALL_DUPLICATE_OF_ATTR] !== void 0) return body2;
-        const facts = this.systemContext.measureUnseen();
+        const facts = this.systemContext.measureUnseen(ts2);
         return Object.keys(facts).length === 0 ? body2 : { ...facts, ...body2 };
       }
+      /**
+       * Note text an Oxagen hook answer handed this session's agent after its
+       * start, at `at`, when the answer left (#5339). Every model call on the
+       * session's own conversation from then on carries the running total as
+       * `context_frame_tokens`. The hook handler calls this only for a live
+       * answer, since a replay's answer reaches no harness. A rollback of the
+       * chain takes the note back with everything else the hook changed.
+       */
+      noteInjectedContext(text4, at) {
+        this.systemContext.noteInjectedContext(text4, at);
+      }
       /**
        * The attrs a `tool_call` carries, or undefined when the chain already
        * holds this call and the row must not be sealed at all.
@@ -17113,6 +17243,9 @@ var init_recorder = __esm({
           this.started = true;
           this.stopped = false;
           body2 = { ...body2, env_snapshot: this.envSnapshot };
+          const source = body2["session_start_source"];
+          if (source === "compact" || source === "clear")
+            this.systemContext.clearInjectedContext(ts2);
         }
         if (draft.kind === "turn_start") {
           if (this.turnOpen) {
@@ -17389,7 +17522,7 @@ var init_recorder = __esm({
         }
         const event = this.seal(
           draft.kind,
-          draft.kind === "llm_call" && draft.source === "otel_log" && onSessionConversation(draft.standard.context.query_source) ? this.withUnseenRequestSources(draft.body, duplicate) : draft.body,
+          draft.kind === "llm_call" && draft.source === "otel_log" && onSessionConversation(draft.standard.context.query_source) ? this.withUnseenRequestSources(draft.body, duplicate, draft.ts) : draft.body,
           {
             ts: draft.ts,
             source: draft.source,
@@ -17472,7 +17605,7 @@ var init_recorder = __esm({
             duplicate = { [LLM_CALL_DUPLICATE_OF_ATTR]: "transcript" };
           }
           if (draft.kind === "llm_call")
-            body2 = this.withUnseenRequestSources(body2, duplicate);
+            body2 = this.withUnseenRequestSources(body2, duplicate, draft.ts);
           let content = draft.content;
           if (this.isSealedPrompt(draft.kind, body2)) {
             content = void 0;
@@ -23364,7 +23497,7 @@ import { readFileSync as readFileSync11 } from "node:fs";
 import { dirname as dirname8, resolve as resolve2 } from "node:path";
 import { fileURLToPath } from "node:url";
 function readVersion() {
-  if (true) return "2.1.4-494";
+  if (true) return "2.1.4-498";
   try {
     const here = dirname8(fileURLToPath(import.meta.url));
     const pkg = JSON.parse(
@@ -38458,6 +38591,11 @@ function drainMidTurn(input, record, deps, events, replay, used = 0) {
   if (continuation !== void 0) texts.push(continuation);
   return texts;
 }
+function noteHandedContext(record, input, replay, text4, at) {
+  if (replay !== void 0 || input.agent_id !== void 0) return;
+  if (text4.length === 0) return;
+  record.recorder.noteInjectedContext(text4, at);
+}
 function hookLedgerKey(raw, harness, hookId) {
   if (hookId !== void 0) return hookId;
   if (harness === "stella" || typeof raw !== "object" || raw === null)
@@ -38760,6 +38898,11 @@ async function routeHook(raw, env, deps, replay, harness, agent, hookId) {
       ) : void 0;
       if (block === void 0) record.control.resumeOwed = void 0;
       if (block === void 0) notePrompt(record, promptText(input));
+      const handed = [
+        ...delivered,
+        ...recalled !== void 0 ? [recalled.text] : []
+      ];
+      const context = handed.join(CONTEXT_JOINER);
       events.push(
         ...record.recorder.ingestHook(
           payload,
@@ -38767,11 +38910,16 @@ async function routeHook(raw, env, deps, replay, harness, agent, hookId) {
           at,
           (draft) => withReplay({
             ...draft,
-            ...recalled !== void 0 ? {
+            ...recalled !== void 0 || handed.length > 0 ? {
               attrs: {
                 ...draft.attrs,
-                "oxagen.recall_digest": digestText(recalled.text),
-                "oxagen.recalled_memories": String(recalled.count)
+                ...recalled !== void 0 ? {
+                  "oxagen.recall_digest": digestText(recalled.text),
+                  "oxagen.recalled_memories": String(recalled.count)
+                } : {},
+                // How much text the answer hands the agent, as the start
+                // records it (#5339).
+                ...handed.length > 0 ? { "oxagen.delivered_chars": String(context.length) } : {}
               }
             } : {},
             body: {
@@ -38792,16 +38940,13 @@ async function routeHook(raw, env, deps, replay, harness, agent, hookId) {
           record
         };
       }
-      const context = [
-        ...delivered,
-        ...recalled !== void 0 ? [recalled.text] : []
-      ];
+      noteHandedContext(record, input, replay, context, at);
       return {
         events,
-        response: context.length > 0 ? {
+        response: handed.length > 0 ? {
           hookSpecificOutput: {
             hookEventName: "UserPromptSubmit",
-            additionalContext: context.join(CONTEXT_JOINER)
+            additionalContext: context
           }
         } : {},
         record
@@ -39070,6 +39215,7 @@ async function routeHook(raw, env, deps, replay, harness, agent, hookId) {
       if (operatorBlock(view, record) !== void 0)
         return { events, response: {}, record };
       const texts = drainMidTurn(input, record, deps, events, replay);
+      noteHandedContext(record, input, replay, texts.join(CONTEXT_JOINER), at);
       return {
         events,
         response: texts.length > 0 ? {
@@ -39104,6 +39250,7 @@ async function routeHook(raw, env, deps, replay, harness, agent, hookId) {
         ...drainMidTurn(input, record, deps, events, replay, ask?.length ?? 0),
         ...ask !== void 0 ? [ask] : []
       ];
+      noteHandedContext(record, input, replay, texts.join(CONTEXT_JOINER), at);
       return {
         events,
         response: texts.length > 0 ? { decision: "block", reason: texts.join(CONTEXT_JOINER) } : {},
@@ -52121,7 +52268,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "@oxagen/cli",
-      version: "2.1.4-494",
+      version: "2.1.4-498",
       description: "Oxagen governance CLI \u2014 spend, lineage, grounding, credentials, and audit (agentic coding lives in the stella CLI)",
       type: "module",
       main: "./dist/index.js",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oxagen/cli",
-  "version": "2.1.4-494",
+  "version": "2.1.4-498",
   "description": "Oxagen governance CLI — spend, lineage, grounding, credentials, and audit (agentic coding lives in the stella CLI)",
   "type": "module",
   "bin": {