fix(core): auto-compact subagent context to prevent overflow

docs/users/features/code-review.md

@@ -129,15 +129,15 @@ You can review PRs from other repositories by passing the full URL:
 
 This runs in **lightweight mode** — no worktree, no linter, no build/test, no autofix. The review is based on the diff text only (fetched via GitHub API). PR comments can still be posted if you have write access.
 
-| Capability                                       | Same-repo | Cross-repo                    |
-| ------------------------------------------------ | --------- | ----------------------------- |
+| Capability                                                 | Same-repo | Cross-repo                    |
+| ---------------------------------------------------------- | --------- | ----------------------------- |
 | LLM review (Agents 1-6 + verify + iterative reverse audit) | ✅        | ✅                            |
 | Agent 7: Build & test                                      | ✅        | ❌ (no local codebase)        |
-| Deterministic analysis (linter/typecheck)        | ✅        | ❌                            |
-| Cross-file impact analysis                       | ✅        | ❌                            |
-| Autofix                                          | ✅        | ❌                            |
-| PR inline comments                               | ✅        | ✅ (if you have write access) |
-| Incremental review cache                         | ✅        | ❌                            |
+| Deterministic analysis (linter/typecheck)                  | ✅        | ❌                            |
+| Cross-file impact analysis                                 | ✅        | ❌                            |
+| Autofix                                                    | ✅        | ❌                            |
+| PR inline comments                                         | ✅        | ✅ (if you have write access) |
+| Incremental review cache                                   | ✅        | ❌                            |
 
 ## PR Inline Comments
 
@@ -258,13 +258,13 @@ For large diffs (>10 modified symbols), analysis prioritizes functions with sign
 
 The review pipeline uses a bounded number of LLM calls regardless of how many findings are produced:
 
-| Stage                            | LLM calls         | Notes                                                |
-| -------------------------------- | ----------------- | ---------------------------------------------------- |
-| Deterministic analysis (Step 3)  | 0                 | Shell commands only                                  |
-| Review agents (Step 4)           | 9 (or 8)          | Run in parallel; Agent 7 skipped in cross-repo mode  |
-| Batch verification (Step 5)      | 1                 | Single agent verifies all findings at once           |
-| Iterative reverse audit (Step 6) | 1-3               | Loops until "No issues found" or 3-round cap         |
-| **Total**                        | **11-13 (10-12)** | Same-repo: 11-13; cross-repo: 10-12 (no Agent 7)     |
+| Stage                            | LLM calls         | Notes                                               |
+| -------------------------------- | ----------------- | --------------------------------------------------- |
+| Deterministic analysis (Step 3)  | 0                 | Shell commands only                                 |
+| Review agents (Step 4)           | 9 (or 8)          | Run in parallel; Agent 7 skipped in cross-repo mode |
+| Batch verification (Step 5)      | 1                 | Single agent verifies all findings at once          |
+| Iterative reverse audit (Step 6) | 1-3               | Loops until "No issues found" or 3-round cap        |
+| **Total**                        | **11-13 (10-12)** | Same-repo: 11-13; cross-repo: 10-12 (no Agent 7)    |
 
 Most PRs converge to the lower end of the range (1 reverse audit round); the cap prevents runaway cost on pathological cases.
 

packages/cli/src/commands/review/cleanup.ts

@@ -84,9 +84,7 @@ function runCleanup(target: string): void {
       writeStdoutLine(`Removed temp file: ${full}`);
       removedAny = true;
     } catch (err) {
-      writeStderrLine(
-        `Failed to remove ${full}: ${(err as Error).message}`,
-      );
+      writeStderrLine(`Failed to remove ${full}: ${(err as Error).message}`);
     }
   }
 

packages/cli/src/commands/review/deterministic.ts

@@ -35,7 +35,10 @@
 // review output verbatim — they skip Step 5 verification.
 
 import type { CommandModule } from 'yargs';
-import { execFileSync, type ExecFileSyncOptionsWithStringEncoding } from 'node:child_process';
+import {
+  execFileSync,
+  type ExecFileSyncOptionsWithStringEncoding,
+} from 'node:child_process';
 import { existsSync, readFileSync, writeFileSync, mkdirSync } from 'node:fs';
 import { join, dirname, resolve } from 'node:path';
 import { writeStdoutLine } from '../../utils/stdioHelpers.js';
@@ -279,7 +282,11 @@ const eslintTool: ToolDef = {
       ],
       TIMEOUT_LINTER_MS,
     );
-    const findings = parseEslintJson(ex.stdout, ctx.worktree, ctx.changedFilesSet);
+    const findings = parseEslintJson(
+      ex.stdout,
+      ctx.worktree,
+      ctx.changedFilesSet,
+    );
     return { exitCode: ex.exitCode, findings, timedOut: ex.timedOut };
   },
 };
@@ -351,7 +358,8 @@ const ruffTool: ToolDef = {
     if (!hasConfig) {
       return {
         ok: false,
-        reason: 'no ruff config (ruff.toml / .ruff.toml / pyproject [tool.ruff])',
+        reason:
+          'no ruff config (ruff.toml / .ruff.toml / pyproject [tool.ruff])',
       };
     }
     if (!which('ruff')) return { ok: false, reason: 'ruff not in PATH' };
@@ -368,7 +376,11 @@ const ruffTool: ToolDef = {
       ['check', '--output-format=json', ...targets],
       TIMEOUT_LINTER_MS,
     );
-    const findings = parseRuffJson(ex.stdout, ctx.worktree, ctx.changedFilesSet);
+    const findings = parseRuffJson(
+      ex.stdout,
+      ctx.worktree,
+      ctx.changedFilesSet,
+    );
     return { exitCode: ex.exitCode, findings, timedOut: ex.timedOut };
   },
 };
@@ -695,10 +707,7 @@ async function runDeterministic(args: DeterministicArgs): Promise<void> {
   writeFileSync(args.out, JSON.stringify(result, null, 2) + '\n', 'utf8');
 
   const summary = toolsRun
-    .map(
-      (r) =>
-        `${r.tool}=${r.findingsCount}${r.timedOut ? ' (timeout)' : ''}`,
-    )
+    .map((r) => `${r.tool}=${r.findingsCount}${r.timedOut ? ' (timeout)' : ''}`)
     .join(', ');
   writeStdoutLine(
     `Wrote deterministic report to ${args.out}: ${findings.length} findings (${summary || 'no tools applicable'}; skipped ${toolsSkipped.length})`,

packages/cli/src/commands/review/fetch-pr.ts

@@ -26,11 +26,7 @@ import { dirname } from 'node:path';
 import { writeStdoutLine, writeStderrLine } from '../../utils/stdioHelpers.js';
 import { ensureAuthenticated, gh } from './lib/gh.js';
 import { git, refExists } from './lib/git.js';
-import {
-  REVIEW_TMP_DIR,
-  reviewBranch,
-  worktreePath,
-} from './lib/paths.js';
+import { REVIEW_TMP_DIR, reviewBranch, worktreePath } from './lib/paths.js';
 
 interface PrMetadata {
   headRefName: string;
@@ -88,12 +84,7 @@ function cleanStale(prNumber: string): void {
 }
 
 async function runFetchPr(args: FetchPrArgs): Promise<void> {
-  const {
-    pr_number: prNumber,
-    owner_repo: ownerRepo,
-    remote,
-    out,
-  } = args;
+  const { pr_number: prNumber, owner_repo: ownerRepo, remote, out } = args;
 
   if (ownerRepo.indexOf('/') < 0) {
     throw new Error('owner_repo must look like "owner/repo"');

packages/cli/src/commands/review/load-rules.ts

@@ -123,7 +123,9 @@ async function runLoadRules(args: LoadRulesArgs): Promise<void> {
   writeFileSync(out, combined, 'utf8');
 
   if (loaded.length === 0) {
-    writeStdoutLine(`No review rules found on ${baseRef}; wrote empty file to ${out}`);
+    writeStdoutLine(
+      `No review rules found on ${baseRef}; wrote empty file to ${out}`,
+    );
   } else {
     writeStdoutLine(
       `Loaded ${loaded.length} rule source(s) from ${baseRef} → ${out}: ${loaded.join(', ')}`,
@@ -134,7 +136,7 @@ async function runLoadRules(args: LoadRulesArgs): Promise<void> {
 export const loadRulesCommand: CommandModule = {
   command: 'load-rules <base_ref>',
   describe:
-    "Read project review rules from the base branch (.qwen/review-rules.md, .github/copilot-instructions.md, AGENTS.md, QWEN.md) and write a combined Markdown file",
+    'Read project review rules from the base branch (.qwen/review-rules.md, .github/copilot-instructions.md, AGENTS.md, QWEN.md) and write a combined Markdown file',
   builder: (yargs) =>
     yargs
       .positional('base_ref', {
@@ -146,7 +148,8 @@ export const loadRulesCommand: CommandModule = {
       .option('out', {
         type: 'string',
         demandOption: true,
-        describe: 'Output Markdown path (will be overwritten — empty if no rules found)',
+        describe:
+          'Output Markdown path (will be overwritten — empty if no rules found)',
       }),
   handler: async (argv) => {
     await runLoadRules(argv as unknown as LoadRulesArgs);

packages/cli/src/commands/review/pr-context.ts

@@ -67,10 +67,7 @@ function snippet(s: string | undefined, max = 240): string {
  * Walk a comment's `in_reply_to_id` chain up to the root. Defends against
  * cycles (which shouldn't happen on GitHub but cheap to handle).
  */
-function findRootId(
-  startId: number,
-  byId: Map<number, RawComment>,
-): number {
+function findRootId(startId: number, byId: Map<number, RawComment>): number {
   const seen = new Set<number>();
   let cur = startId;
   while (true) {
@@ -141,7 +138,9 @@ function buildMarkdown(
   parts.push(`- **Repo:** ${ownerRepo}`);
   parts.push(`- **Author:** @${meta.author?.login ?? 'unknown'}`);
   parts.push(`- **State:** ${meta.state}`);
-  parts.push(`- **Base → Head:** \`${meta.baseRefName}\` ← \`${meta.headRefName}\``);
+  parts.push(
+    `- **Base → Head:** \`${meta.baseRefName}\` ← \`${meta.headRefName}\``,
+  );
   parts.push(`- **HEAD SHA:** \`${meta.headRefOid}\``);
   parts.push(
     `- **Diff:** ${meta.changedFiles} files, +${meta.additions}/-${meta.deletions}`,
@@ -187,7 +186,9 @@ function buildMarkdown(
   // only root-comment snippets and forced the LLM driver to manually
   // summarise each reply chain in agent prompts.
   if (repliedRoots.length > 0 || issue.length > 0) {
-    parts.push('## Already discussed — do NOT re-report unless the latest reply itself raises a new concern');
+    parts.push(
+      '## Already discussed — do NOT re-report unless the latest reply itself raises a new concern',
+    );
     parts.push('');
     if (repliedRoots.length > 0) {
       parts.push('### Inline-comment threads with replies');
@@ -209,9 +210,7 @@ function buildMarkdown(
         if (replies.length > 0) {
           parts.push('Replies (chronological):');
           for (const r of replies) {
-            parts.push(
-              `- **@${r.user?.login ?? '?'}**: ${snippet(r.body)}`,
-            );
+            parts.push(`- **@${r.user?.login ?? '?'}**: ${snippet(r.body)}`);
           }
           parts.push('');
         }
@@ -221,16 +220,16 @@ function buildMarkdown(
       parts.push('### Issue-level comments (general PR thread)');
       parts.push('');
       for (const c of issue) {
-        parts.push(
-          `- by @${c.user?.login ?? '?'}: ${snippet(c.body)}`,
-        );
+        parts.push(`- by @${c.user?.login ?? '?'}: ${snippet(c.body)}`);
       }
       parts.push('');
     }
   }
 
   if (openRoots.length > 0) {
-    parts.push('## Open inline comments (no replies yet — may still need attention)');
+    parts.push(
+      '## Open inline comments (no replies yet — may still need attention)',
+    );
     parts.push('');
     for (const c of openRoots) {
       parts.push(

packages/cli/src/commands/review/presubmit.ts

@@ -201,9 +201,7 @@ async function runPresubmit(args: PresubmitArgs): Promise<void> {
   if (newFindingsPath) {
     newFindings = JSON.parse(readFileSync(newFindingsPath, 'utf8'));
   }
-  const newFindingKeys = new Set(
-    newFindings.map((f) => `${f.path}:${f.line}`),
-  );
+  const newFindingKeys = new Set(newFindings.map((f) => `${f.path}:${f.line}`));
 
   const buckets = classifyExistingComments(
     qwenComments,
@@ -216,7 +214,9 @@ async function runPresubmit(args: PresubmitArgs): Promise<void> {
   const downgradeReasons: string[] = [];
   if (isSelfPr) downgradeReasons.push('self-PR');
   if (ciStatus.class === 'any_failure') {
-    downgradeReasons.push(`CI failing: ${ciStatus.failedCheckNames.join(', ')}`);
+    downgradeReasons.push(
+      `CI failing: ${ciStatus.failedCheckNames.join(', ')}`,
+    );
   }
   if (ciStatus.class === 'all_pending') {
     downgradeReasons.push('CI still running');

packages/core/src/agents/runtime/agent-core.ts

@@ -317,11 +317,17 @@ export class AgentCore {
     }
 
     try {
-      return new GeminiChat(
+      const chat = new GeminiChat(
         this.runtimeContext,
         generationConfig,
         startHistory,
       );
+      // Seed the per-chat token count so the auto-compaction threshold
+      // gate sees the inherited history's true size on the first send.
+      // Without this, fork subagents start at 0 and the gate NOOPs even
+      // when `startHistory` is already huge — first API call can 400.
+      chat.setLastPromptTokenCount(this.lastPromptTokenCount);
+      return chat;
     } catch (error) {
       await reportError(
         error,
@@ -540,6 +546,18 @@ export class AgentCore {
           continue;
         }
 
+        // GeminiChat already mutated its own history; surface to the debug
+        // log so subagent compactions show up alongside the main session's.
+        if (streamEvent.type === 'compressed') {
+          this.runtimeContext
+            .getDebugLogger()
+            .debug(
+              `[AGENT-COMPACT] subagent=${this.subagentId} round=${turnCounter} ` +
+                `tokens ${streamEvent.info.originalTokenCount} -> ${streamEvent.info.newTokenCount}`,
+            );
+          continue;
+        }
+
         // Handle chunk events
         if (streamEvent.type === 'chunk') {
           const resp = streamEvent.value;

packages/core/src/agents/runtime/agent-headless.test.ts

@@ -280,6 +280,7 @@ describe('subagent.ts', () => {
         () =>
           ({
             sendMessageStream: mockSendMessageStream,
+            setLastPromptTokenCount: vi.fn(),
           }) as unknown as GeminiChat,
       );
 
@@ -958,6 +959,7 @@ describe('subagent.ts', () => {
           () =>
             ({
               sendMessageStream: mockSendMessageStream,
+              setLastPromptTokenCount: vi.fn(),
             }) as unknown as GeminiChat,
         );
 
@@ -997,6 +999,7 @@ describe('subagent.ts', () => {
           () =>
             ({
               sendMessageStream: mockSendMessageStream,
+              setLastPromptTokenCount: vi.fn(),
             }) as unknown as GeminiChat,
         );
 
@@ -1061,6 +1064,7 @@ describe('subagent.ts', () => {
           () =>
             ({
               sendMessageStream: mockSendMessageStream,
+              setLastPromptTokenCount: vi.fn(),
             }) as unknown as GeminiChat,
         );