@triflux/remote 10.1.0 → 10.26.0

package/hub/team/cli/services/task-model.mjs

@@ -1,30 +1,40 @@
-export function buildTasks(subtasks, workers) {
-  return subtasks.map((subtask, index) => ({
-    id: `T${index + 1}`,
-    title: subtask,
-    owner: workers[index]?.name || null,
-    status: "pending",
-    depends_on: index === 0 ? [] : [`T${index}`],
-  }));
-}
-
-export function normalizeTaskStatus(action) {
-  const value = String(action || "").toLowerCase();
-  if (value === "done" || value === "complete" || value === "completed") return "completed";
-  if (value === "progress" || value === "in-progress" || value === "in_progress") return "in_progress";
-  if (value === "pending") return "pending";
-  return null;
-}
-
-export function updateTaskStatus(tasks = [], taskId, nextStatus) {
-  const normalizedId = String(taskId || "").toUpperCase();
-  const target = tasks.find((task) => String(task.id).toUpperCase() === normalizedId);
-  if (!target) return { tasks, target: null };
-
-  return {
-    target: { ...target, status: nextStatus },
-    tasks: tasks.map((task) => (
-      String(task.id).toUpperCase() === normalizedId ? { ...task, status: nextStatus } : task
-    )),
-  };
-}
+export function buildTasks(subtasks, workers) {
+  return subtasks.map((subtask, index) => ({
+    id: `T${index + 1}`,
+    title: subtask,
+    owner: workers[index]?.name || null,
+    status: "pending",
+    depends_on: index === 0 ? [] : [`T${index}`],
+  }));
+}
+
+export function normalizeTaskStatus(action) {
+  const value = String(action || "").toLowerCase();
+  if (value === "done" || value === "complete" || value === "completed")
+    return "completed";
+  if (
+    value === "progress" ||
+    value === "in-progress" ||
+    value === "in_progress"
+  )
+    return "in_progress";
+  if (value === "pending") return "pending";
+  return null;
+}
+
+export function updateTaskStatus(tasks = [], taskId, nextStatus) {
+  const normalizedId = String(taskId || "").toUpperCase();
+  const target = tasks.find(
+    (task) => String(task.id).toUpperCase() === normalizedId,
+  );
+  if (!target) return { tasks, target: null };
+
+  return {
+    target: { ...target, status: nextStatus },
+    tasks: tasks.map((task) =>
+      String(task.id).toUpperCase() === normalizedId
+        ? { ...task, status: nextStatus }
+        : task,
+    ),
+  };
+}

package/hub/team/codex-review.mjs

@@ -0,0 +1,521 @@
+// Codex headless review helper. Runs a cross-model review of a git diff
+// without going through Bash-level headless-guard (the guard only inspects
+// `codex exec` invoked from the Bash tool — node `spawn` bypasses it
+// intentionally because `tfx` is the sanctioned entry point).
+//
+// Contract:
+//   runCodexReview({ ref = "HEAD", base?: string, timeoutMs = 180_000 })
+//     → { ok, code?, verdict, stdout, stderr, diffBytes, error? }
+//   runCodexReviewSharded({ ref, base, timeoutMs })
+//     → { ok, verdict, range, files, perFile: [{ file, verdict, ... }] }
+//   Verdict parsed from Codex output: APPROVED / REQUEST_CHANGES / COMMENT
+//   / UNKNOWN when none is stated.
+//
+// Motivation: session 12 post-mortem (BUG-J fix merged without cross-review).
+// Without an ergonomic `tfx review` path, every swarm bugfix skips the
+// Codex independent opinion — the exact check that caught BUG-I's path-only
+// filter bypass in round 1 (session 11 PR #134). Session 13 adds --shard
+// per-file for diffs that exceed the 32KB single-spawn ceiling.
+
+import { execFileSync, spawn } from "node:child_process";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+// OS arg-list ceiling. The whole chain (bash → codex.cmd → node) fails
+// with ENAMETOOLONG / "Argument list too long" well below the documented
+// 128KB on Windows. Observed breakage at 62KB diff (session 12 self-dogfood).
+// 32KB is a safe ceiling that covers most single-commit swarm fixes.
+// Larger ranges should be reviewed per-file with --shard per-file or
+// narrowed with --base.
+export const MAX_PROMPT_BYTES = 32_000;
+
+/**
+ * Expand a {ref, base} pair into a git range string.
+ * - `base` set → `base..ref`
+ * - `ref` already contains `..` → pass through
+ * - otherwise → `ref~1..ref` (single-commit review)
+ *
+ * Shared by single-mode `runCodexReview` and shard-mode
+ * `runCodexReviewSharded` so both interpret identical inputs identically.
+ *
+ * @param {{ ref?: string, base?: string }} opts
+ * @returns {string}
+ */
+export function expandRange({ ref = "HEAD", base } = {}) {
+  if (base) return `${base}..${ref}`;
+  if (ref.includes("..")) return ref;
+  return `${ref}~1..${ref}`;
+}
+
+/**
+ * Resolve the diff payload for a given ref.
+ * - `a..b` ranges pass through unchanged
+ * - `<commit>` expands to `<commit>~1..<commit>` (single-commit review)
+ * - `HEAD` with no prior commit returns an empty string
+ * - optional `file` scopes the diff to a single path via `-- <file>`
+ *
+ * @param {{ ref?: string, base?: string, file?: string, runner?: (cmd: string, args: string[]) => string }} opts
+ * @returns {{ diff: string, range: string, file?: string }}
+ */
+export function resolveReviewDiff({
+  ref = "HEAD",
+  base,
+  file,
+  runner = _defaultGitRunner,
+} = {}) {
+  const range = expandRange({ ref, base });
+
+  const args = ["log", "-p", "--stat", "--no-color", range];
+  if (file) {
+    args.push("--", file);
+  }
+  const diff = runner("git", args);
+  return { diff, range, file };
+}
+
+/**
+ * Parse the raw `git log --name-only --pretty=format:` output into a
+ * de-duplicated, non-blank file path array. Pure function — extracted
+ * so shard enumeration can be tested against canned log output without
+ * requiring a temp git repo fixture.
+ *
+ * @param {string|null|undefined} rawLogOutput
+ * @returns {string[]}
+ */
+export function parseChangedFilesFromLog(rawLogOutput) {
+  const seen = new Set();
+  for (const line of (rawLogOutput || "").split(/\r?\n/)) {
+    const trimmed = line.trim();
+    if (trimmed) seen.add(trimmed);
+  }
+  return [...seen];
+}
+
+/**
+ * List files changed in a range via `git log --name-only`, matching
+ * the commit-log semantics used by `resolveReviewDiff`. Using
+ * `git diff --name-only` would report only the net tree diff and
+ * silently drop files changed then reverted within a multi-commit
+ * range — those files would then be invisible to shard-mode review
+ * even though `git log -p` still contains their edits.
+ *
+ * Merge commits are included so enumeration stays aligned with
+ * `resolveReviewDiff`'s plain `git log -p`. Merges rarely introduce
+ * novel content but also rarely add noise since the dedupe Set
+ * collapses paths that already appear in parent commits.
+ *
+ * @param {string} range — e.g. "HEAD~1..HEAD" or "main..feature"
+ * @param {(cmd: string, args: string[]) => string} [runner] — injectable
+ *   process runner. Default shells out to git; tests can pass a stub to
+ *   verify argv contract (e.g. absence of `--no-merges`) without spawning.
+ * @returns {string[]}
+ */
+export function listChangedFiles(range, runner = _defaultGitRunner) {
+  const out = runner("git", ["log", "--name-only", "--pretty=format:", range]);
+  return parseChangedFilesFromLog(out);
+}
+
+function _defaultGitRunner(cmd, args) {
+  return execFileSync(cmd, args, {
+    encoding: "utf8",
+    windowsHide: true,
+    maxBuffer: 50 * 1024 * 1024,
+  });
+}
+
+/**
+ * Build the Codex review prompt.
+ * @param {string} diff
+ * @param {{ range: string }} ctx
+ */
+export function buildReviewPrompt(diff, { range }) {
+  return [
+    `You are acting as an independent cross-model reviewer.`,
+    `Review the following git log range: ${range}.`,
+    ``,
+    `Scope: correctness, regression risk, security, test coverage, code style.`,
+    ``,
+    `Classify each finding as:`,
+    `- HIGH: blocking — logic bug, security hole, regression`,
+    `- MEDIUM: should-fix — missing test, readability, maintainability`,
+    `- LOW: nit — style, micro-optimization`,
+    ``,
+    `Format each finding as: [SEVERITY] file:line — description. Suggested fix: ...`,
+    `If a category has no findings, write 'none'.`,
+    ``,
+    `End with exactly one line stating the verdict:`,
+    `VERDICT: APPROVED  (no HIGH findings)`,
+    `VERDICT: REQUEST_CHANGES  (any HIGH finding)`,
+    `VERDICT: COMMENT  (observations only, no blockers)`,
+    ``,
+    "```diff",
+    diff,
+    "```",
+  ].join("\n");
+}
+
+/**
+ * Parse the verdict marker from Codex stdout.
+ * @param {string} stdout
+ */
+export function parseVerdict(stdout) {
+  const match = stdout.match(
+    /VERDICT:\s*(APPROVED|REQUEST_CHANGES|COMMENT)\b/i,
+  );
+  if (match) return match[1].toUpperCase();
+  if (/REQUEST_CHANGES/i.test(stdout)) return "REQUEST_CHANGES";
+  if (/\bAPPROVED\b/i.test(stdout)) return "APPROVED";
+  return "UNKNOWN";
+}
+
+/**
+ * Aggregate per-file verdicts into an overall verdict.
+ * - Any REQUEST_CHANGES → REQUEST_CHANGES
+ * - Any COMMENT (no REQUEST_CHANGES) → COMMENT
+ * - All APPROVED (skipped excluded) → APPROVED
+ * - Mixed with UNKNOWN → UNKNOWN (conservative)
+ *
+ * @param {Array<{verdict: string|null, skipped?: boolean}>} fileResults
+ * @returns {string}
+ */
+export function aggregateVerdicts(fileResults) {
+  const active = fileResults.filter((r) => !r.skipped);
+  if (active.length === 0) return "UNKNOWN";
+  const verdicts = active.map((r) => r.verdict);
+  if (verdicts.some((v) => v === "REQUEST_CHANGES")) return "REQUEST_CHANGES";
+  if (verdicts.some((v) => v === "COMMENT")) return "COMMENT";
+  if (verdicts.every((v) => v === "APPROVED")) return "APPROVED";
+  return "UNKNOWN";
+}
+
+/**
+ * Internal: run Codex on a pre-built prompt and parse the response.
+ * Shared by single-mode runCodexReview and shard-mode runCodexReviewSharded.
+ */
+async function _runCodexOnPrompt({
+  prompt,
+  range,
+  diffBytes,
+  timeoutMs,
+  sandbox,
+  env,
+}) {
+  // Prompt is a full git diff — often >10KB, regularly >60KB on swarm
+  // bugfixes. Windows cmd.exe caps command lines at ~8KB and node's
+  // `spawn(..., [promptArg])` route cannot exceed the OS arg limit
+  // (ENAMETOOLONG on Windows). Persist the prompt to a tmp file and let
+  // bash read it back — same pattern session 11 used manually
+  // (`codex exec "$(cat prompt.md)" ...`).
+  const tmpDir = mkdtempSync(join(tmpdir(), "tfx-review-"));
+  const promptFile = join(tmpDir, "prompt.md").replace(/\\/g, "/");
+  writeFileSync(promptFile, prompt);
+
+  const shellCmd = [
+    "codex",
+    "exec",
+    "-s",
+    sandbox,
+    "--dangerously-bypass-approvals-and-sandbox",
+    `"$(cat '${promptFile.replace(/'/g, "'\\''")}')"`,
+  ].join(" ");
+
+  return new Promise((resolve) => {
+    const child = spawn("bash", ["-c", shellCmd], {
+      stdio: ["ignore", "pipe", "pipe"],
+      windowsHide: true,
+      env,
+    });
+
+    let stdout = "";
+    let stderr = "";
+    let timedOut = false;
+
+    const timer = setTimeout(() => {
+      timedOut = true;
+      child.kill("SIGTERM");
+    }, timeoutMs);
+
+    child.stdout.on("data", (d) => {
+      stdout += d.toString();
+    });
+    child.stderr.on("data", (d) => {
+      stderr += d.toString();
+    });
+    const cleanupTmp = () => {
+      try {
+        rmSync(tmpDir, { recursive: true, force: true });
+      } catch {
+        /* best-effort */
+      }
+    };
+
+    child.on("error", (err) => {
+      clearTimeout(timer);
+      cleanupTmp();
+      resolve({
+        ok: false,
+        error: err.message,
+        verdict: null,
+        stdout,
+        stderr,
+        diffBytes,
+        range,
+      });
+    });
+    child.on("close", (code) => {
+      clearTimeout(timer);
+      cleanupTmp();
+      if (timedOut) {
+        resolve({
+          ok: false,
+          code,
+          error: `codex review timed out after ${timeoutMs}ms`,
+          verdict: null,
+          stdout,
+          stderr,
+          diffBytes,
+          range,
+        });
+        return;
+      }
+      resolve({
+        ok: code === 0,
+        code,
+        verdict: parseVerdict(stdout),
+        stdout,
+        stderr,
+        diffBytes,
+        range,
+      });
+    });
+  });
+}
+
+/**
+ * Run Codex review of a git range (single-spawn mode).
+ * @param {object} opts
+ * @param {string} [opts.ref="HEAD"]
+ * @param {string} [opts.base]
+ * @param {number} [opts.timeoutMs=180000]
+ * @param {string} [opts.sandbox="read-only"]
+ * @param {object} [opts.env]
+ */
+export async function runCodexReview({
+  ref = "HEAD",
+  base,
+  timeoutMs = 180_000,
+  sandbox = "read-only",
+  env = process.env,
+  _deps = {},
+} = {}) {
+  let range;
+  let diff;
+  try {
+    ({ diff, range } = (_deps.resolveReviewDiff || resolveReviewDiff)({
+      ref,
+      base,
+    }));
+  } catch (err) {
+    return {
+      ok: false,
+      error: `git log failed: ${err.message}`,
+      verdict: null,
+      stdout: "",
+      stderr: "",
+      diffBytes: 0,
+    };
+  }
+
+  const diffBytes = Buffer.byteLength(diff, "utf8");
+  if (diffBytes === 0) {
+    return {
+      ok: false,
+      error: `empty diff for range ${range}`,
+      verdict: null,
+      stdout: "",
+      stderr: "",
+      diffBytes: 0,
+      range,
+    };
+  }
+
+  const prompt = buildReviewPrompt(diff, { range });
+  const promptBytes = Buffer.byteLength(prompt, "utf8");
+
+  if (promptBytes > MAX_PROMPT_BYTES) {
+    return {
+      ok: false,
+      error:
+        `prompt too large (${promptBytes} bytes > ${MAX_PROMPT_BYTES}). ` +
+        `Retry with --shard per-file or narrow with --base <sha>.`,
+      verdict: null,
+      stdout: "",
+      stderr: "",
+      diffBytes,
+      promptBytes,
+      range,
+    };
+  }
+
+  return (_deps.runCodexOnPrompt || _runCodexOnPrompt)({
+    prompt,
+    range,
+    diffBytes,
+    timeoutMs,
+    sandbox,
+    env,
+  });
+}
+
+/**
+ * Run Codex review sharded per-file. Each changed file in the range gets
+ * its own Codex spawn with its own 32KB budget. Results are aggregated.
+ *
+ * Sequential, not parallel — Codex headless startup is ~30-90s per call
+ * and parallel spawns risk auth contention. Wall-clock scales linearly
+ * with file count, budget accordingly (default timeout applies per-file).
+ *
+ * @param {object} opts — same shape as runCodexReview
+ * @returns {Promise<{ok: boolean, verdict: string, range: string, files: string[], perFile: object[], error?: string}>}
+ */
+export async function runCodexReviewSharded({
+  ref = "HEAD",
+  base,
+  timeoutMs = 180_000,
+  sandbox = "read-only",
+  env = process.env,
+  onFileStart,
+  onFileDone,
+} = {}) {
+  const range = expandRange({ ref, base });
+
+  let files;
+  try {
+    files = listChangedFiles(range);
+  } catch (err) {
+    return {
+      ok: false,
+      error: `git log --name-only failed: ${err.message}`,
+      verdict: "UNKNOWN",
+      range,
+      files: [],
+      perFile: [],
+    };
+  }
+
+  if (files.length === 0) {
+    return {
+      ok: false,
+      error: `no changed files in range ${range}`,
+      verdict: "UNKNOWN",
+      range,
+      files: [],
+      perFile: [],
+    };
+  }
+
+  const perFile = [];
+  for (const file of files) {
+    if (typeof onFileStart === "function") {
+      try {
+        onFileStart({ file, index: perFile.length, total: files.length });
+      } catch {
+        /* caller callback errors should not block review */
+      }
+    }
+
+    let fileDiff;
+    try {
+      ({ diff: fileDiff } = resolveReviewDiff({ ref, base, file }));
+    } catch (err) {
+      const entry = {
+        file,
+        ok: false,
+        error: `git log scope failed: ${err.message}`,
+        verdict: "UNKNOWN",
+        diffBytes: 0,
+        range: `${range} -- ${file}`,
+      };
+      perFile.push(entry);
+      if (typeof onFileDone === "function") {
+        try {
+          onFileDone(entry);
+        } catch {
+          /* ignore */
+        }
+      }
+      continue;
+    }
+
+    const diffBytes = Buffer.byteLength(fileDiff, "utf8");
+    if (diffBytes === 0) {
+      const entry = {
+        file,
+        ok: true,
+        skipped: true,
+        error: `empty diff for ${file} (likely rename/mode-only)`,
+        verdict: "APPROVED",
+        diffBytes: 0,
+        range: `${range} -- ${file}`,
+      };
+      perFile.push(entry);
+      if (typeof onFileDone === "function") {
+        try {
+          onFileDone(entry);
+        } catch {
+          /* ignore */
+        }
+      }
+      continue;
+    }
+
+    const scopedRange = `${range} -- ${file}`;
+    const prompt = buildReviewPrompt(fileDiff, { range: scopedRange });
+    const promptBytes = Buffer.byteLength(prompt, "utf8");
+    if (promptBytes > MAX_PROMPT_BYTES) {
+      const entry = {
+        file,
+        ok: false,
+        error:
+          `prompt too large (${promptBytes} bytes > ${MAX_PROMPT_BYTES}) ` +
+          `for single file. Review this file manually or split the commit.`,
+        verdict: "UNKNOWN",
+        diffBytes,
+        promptBytes,
+        range: scopedRange,
+      };
+      perFile.push(entry);
+      if (typeof onFileDone === "function") {
+        try {
+          onFileDone(entry);
+        } catch {
+          /* ignore */
+        }
+      }
+      continue;
+    }
+
+    const res = await _runCodexOnPrompt({
+      prompt,
+      range: scopedRange,
+      diffBytes,
+      timeoutMs,
+      sandbox,
+      env,
+    });
+    const entry = { file, ...res };
+    perFile.push(entry);
+    if (typeof onFileDone === "function") {
+      try {
+        onFileDone(entry);
+      } catch {
+        /* ignore */
+      }
+    }
+  }
+
+  const verdict = aggregateVerdicts(perFile);
+  const ok = perFile.every((r) => r.ok);
+  return { ok, verdict, range, files, perFile };
+}

package/hub/team/conductor-mesh-bridge.mjs

@@ -1,4 +1,4 @@
-import { createMessage, MSG_TYPES } from "../../mesh/mesh-protocol.mjs";
+import { createMessage, MSG_TYPES } from "@triflux/core/mesh/mesh-protocol.mjs";
 
 const BRIDGE_AGENT_ID = "conductor";
 
@@ -14,10 +14,7 @@ const BRIDGE_AGENT_ID = "conductor";
  * @returns {object} Bridge API
  */
 export function createConductorMeshBridge(conductor, registry, opts = {}) {
-  const {
-    bridgeAgentId = BRIDGE_AGENT_ID,
-    onMessage,
-  } = opts;
+  const { bridgeAgentId = BRIDGE_AGENT_ID, onMessage } = opts;
 
   const sessionAgentMap = new Map();
   let attached = false;
@@ -49,12 +46,13 @@ export function createConductorMeshBridge(conductor, registry, opts = {}) {
     }
 
     // Emit mesh event
-    const msg = createMessage(
-      MSG_TYPES.EVENT,
-      bridgeAgentId,
-      "*",
-      { event: "stateChange", sessionId, from, to, reason },
-    );
+    const msg = createMessage(MSG_TYPES.EVENT, bridgeAgentId, "*", {
+      event: "stateChange",
+      sessionId,
+      from,
+      to,
+      reason,
+    });
     emit(msg);
 
     // Unregister on terminal states
@@ -68,12 +66,10 @@ export function createConductorMeshBridge(conductor, registry, opts = {}) {
    * Handle conductor completed events.
    */
   function handleCompleted(event) {
-    const msg = createMessage(
-      MSG_TYPES.EVENT,
-      bridgeAgentId,
-      "*",
-      { event: "completed", sessionId: event.sessionId },
-    );
+    const msg = createMessage(MSG_TYPES.EVENT, bridgeAgentId, "*", {
+      event: "completed",
+      sessionId: event.sessionId,
+    });
     emit(msg);
   }
 
@@ -81,12 +77,11 @@ export function createConductorMeshBridge(conductor, registry, opts = {}) {
    * Handle conductor dead events.
    */
   function handleDead(event) {
-    const msg = createMessage(
-      MSG_TYPES.EVENT,
-      bridgeAgentId,
-      "*",
-      { event: "dead", sessionId: event.sessionId, reason: event.reason },
-    );
+    const msg = createMessage(MSG_TYPES.EVENT, bridgeAgentId, "*", {
+      event: "dead",
+      sessionId: event.sessionId,
+      reason: event.reason,
+    });
     emit(msg);
   }
 
@@ -117,5 +112,11 @@ export function createConductorMeshBridge(conductor, registry, opts = {}) {
     attached = false;
   }
 
-  return { attach, detach, get isAttached() { return attached; } };
+  return {
+    attach,
+    detach,
+    get isAttached() {
+      return attached;
+    },
+  };
 }