flowviant 0.125.6 → 0.125.9

package/bin/lib/workSessionTurns.mjs

@@ -1,3 +1,4 @@
+import { sessionWork, reportWorkOutcome } from './workActivity.mjs';
 /**
  * A WORKBENCH TAB'S TURN — offered, admitted, placed, adopted, pinned to its
  * CLI, credentialed, resumed, run, persisted and settled.
@@ -19,7 +20,7 @@
 import { existsSync, realpathSync, rmSync, statSync, writeFileSync } from 'node:fs';
 import { git } from './git.mjs';
 import { measureRepositoryBaseline } from './repositoryBaseline.mjs';
-import { REPOSITORY_BASELINE_PARAGRAPH } from './prompts.mjs';
+import { REPOSITORY_BASELINE_PARAGRAPH, tabFilesNote } from './prompts.mjs';
 import { processStartTime } from './procRegistry.mjs';
 import { c, note, ok, warn } from './ui.mjs';
 import { mcpFor } from './claude.mjs';
@@ -31,10 +32,11 @@ import { isTerminalSessionLive } from './claudeSessions.mjs';
 import { isAgyConversationLive } from './agySessions.mjs';
 import { adoptFence } from './localSessions.mjs';
 import { resumeRetriesFresh } from './resumeLost.mjs';
+import { CODEX_CONTRACT_MARKER, codexContractKnown, recordCodexContract } from './codexContract.mjs';
 import { turnLockedByLivePid } from './turnLock.mjs';
 import { createDeliveryReporter } from './workDelivery.mjs';
 import { createLiveMessages } from './workLiveMessages.mjs';
-import { killGroup } from './processGroup.mjs';
+import { interruptThenKill, killGroup } from './processGroup.mjs';
 import { printable } from './printable.mjs';
 import { brainFor } from './workBrain.mjs';
 import { AGY_CONV_RE, CLAUDE_SESSION_RE, CODEX_THREAD_RE, RESUME_MARKERS } from './conversationIds.mjs';
@@ -76,6 +78,9 @@ export function createWorkSessionTurns({
   agyRegistryLookup,
   carryDirtyState,
   fetchAttachments,
+  // Each file landed or named as missed (workAttachments.mjs); a caller that
+  // hands only `fetchAttachments` (an older test) has its paths as landed.
+  fetchTabFiles = async (wt, attachments) => (await fetchAttachments(wt, attachments)).map((path) => ({ path })),
   makeNarrator,
   sayTurnDeferred,
   lastDeferSaid,
@@ -145,6 +150,8 @@ export function createWorkSessionTurns({
     return settledHere.has(turnId);
   };
   const settleTurn = (turnId, payload) => {
+    const run = runs.get(turnId);
+    if (run?.job) reportWorkOutcome(sessionWork(run.job), { ...payload, outcome: payload.ok ? 'replied' : 'nothing' });
     const at = now();
     settledHere.delete(turnId); // re-inserted at the end: the map stays in time order
     settledHere.set(turnId, at);
@@ -219,7 +226,7 @@ export function createWorkSessionTurns({
       const releaseSlot = admit.reserve();
       lastDeferSaid.delete(job.sessionId);
       workAnswering.add(job.id);
-      const run = { interrupted: false, spawned: [], sessionId: job.sessionId, live: null };
+      const run = { interrupted: false, spawned: [], sessionId: job.sessionId, live: null, job };
       runs.set(job.id, run);
       /** The execution generation the server handed, echoed on every settle
        *  so the server can tell this execution's answer from an older one's. */
@@ -654,6 +661,17 @@ export function createWorkSessionTurns({
           // refused for want of an approval, by `server.tool`.
           let codexSpoke = false;
           const approvalRefused = [];
+          // WHAT THE CLI SAID OF ITS OWN RUN (2026-10-19): a failed result or
+          // error event, or an exit that was not a clean one. A usage limit, a
+          // signed-out CLI or a crash printed its words and settled as the
+          // tab's REPLY; they settle as the failure they are now, in the same
+          // words. Reset for the fresh retry, which is its own run.
+          let cliFailed = false;
+          let badExit = null;
+          // A Codex tab's contract and whether this turn sent it (codexContract.mjs).
+          const contractPath = rt.id === 'codex' ? sessionMetaPath(dir.wt, CODEX_CONTRACT_MARKER, job.sessionId) : null;
+          let turnSystem = '';
+          let contractSent = true;
           const spawned = []; // this turn's children, for the teardown registry
           // THE TURN'S TOOL LOG — see workToolLog.mjs for its shape rules.
           const turnLog = createToolLog(dir.wt);
@@ -675,12 +693,7 @@ export function createWorkSessionTurns({
             // Files first, then the message that references them: the agent
             // must be able to open what it is being told about. Only the ones
             // that actually landed are named.
-            const files = await fetchAttachments(dir.wt, job.attachments);
-            const filesNote = files.length
-              ? `[FILES THE HUMAN ATTACHED TO THIS MESSAGE — already on disk in this worktree]\n${files
-                  .map((f) => `- ${f}`)
-                  .join('\n')}`
-              : '';
+            const filesNote = tabFilesNote(await fetchTabFiles(dir.wt, job.attachments));
             const message = [job.body, filesNote, adoptNote, carryNote]
               .filter(Boolean)
               .join('\n\n');
@@ -689,8 +702,7 @@ export function createWorkSessionTurns({
                 scope: 'session', turnId: job.id, attempt: job.attempt, cwd: dir.wt, sessionMetaPath,
                 interrupted: () => run.interrupted,
                 format: async (followup) => {
-                  const paths = await fetchAttachments(dir.wt, followup.attachments);
-                  const note = paths.length ? `[FILES THE HUMAN ATTACHED TO THIS MESSAGE — already on disk in this worktree]\n${paths.map((f) => `- ${f}`).join('\n')}` : '';
+                  const note = tabFilesNote(await fetchTabFiles(dir.wt, followup.attachments));
                   const words = shape.capture ? (followup.libraryPrompt ?? '') + followup.body : followup.body + (followup.libraryPrompt ?? '');
                   return sessionTurnKickoff(shape, { sessionId: job.sessionId, sessionName: job.sessionName,
                     askedByName: followup.askedByName, message: [words, note].filter(Boolean).join('\n\n') });
@@ -728,11 +740,18 @@ export function createWorkSessionTurns({
                   // artifacts paragraph, the planning sentence appended). The
                   // artifacts paragraph also waits on the server — ARTIFACTS
                   // (0.94.0): only while it can show one, read here at spawn.
-                  system: sessionTurnSystem(shape, {
-                    knowledgeDir,
-                    artifactsAccepted: getArtifactsAccepted(),
-                    runtime: rt.id,
-                  }) + (toolLog.baseline ? '\n\n' + REPOSITORY_BASELINE_PARAGRAPH(toolLog.baseline) : '') + (artifactScan?.instructions ? `\n\n${artifactScan.instructions}` : ''),
+                  // …and a Codex thread that already holds it is sent none
+                  // (codexContract.mjs); `turnSystem` keeps the text for a
+                  // fresh retry and the marker.
+                  system: (() => {
+                    turnSystem = sessionTurnSystem(shape, {
+                      knowledgeDir,
+                      artifactsAccepted: getArtifactsAccepted(),
+                      runtime: rt.id,
+                    }) + (toolLog.baseline ? '\n\n' + REPOSITORY_BASELINE_PARAGRAPH(toolLog.baseline) : '') + (artifactScan?.instructions ? `\n\n${artifactScan.instructions}` : '');
+                    contractSent = !codexContractKnown(contractPath, codexResumeId, turnSystem);
+                    return contractSent ? turnSystem : '';
+                  })(),
                   // The directory is OUTSIDE a worktree's cwd; Claude Code is
                   // told it may read there (`--add-dir`) rather than left to
                   // refuse a read-only capture turn a path it was just handed.
@@ -774,6 +793,15 @@ export function createWorkSessionTurns({
               onModel: (m) => {
                 model = m;
               },
+              // The CLI's own context measure, live while the turn runs
+              // (claudeInput.mjs); the server clears it at settle.
+              onContext: (ctx) => narrator.context(ctx),
+              onCliFailure: () => {
+                cliFailed = true;
+              },
+              onExit: (code, signal) => {
+                badExit = signal || (code ? `exit ${code}` : null);
+              },
               // What this CLI says it can be asked for by name, and which of
               // its connectors need a sign-in, are harvested off this init by
               // runTurn itself (`initHarvestFor`, runtimeCapabilities.mjs) —
@@ -794,7 +822,7 @@ export function createWorkSessionTurns({
               // chat does not (workTurnShape.mjs).
               display: shape.display,
               runtime: rt.id,
-              label: c.cyan('[tab]'),
+              work: sessionWork(job),
               // Only codex announces one (thread.started); held here so the id
               // this turn actually SPOKE under is what gets persisted after it
               // ends. Last write wins on purpose: a failed resume that fell
@@ -871,9 +899,19 @@ export function createWorkSessionTurns({
              */
             if (
               !adopting &&
-              resumeRetriesFresh({ resume, out, runtime: rt.id, sawInit: Boolean(seenClaudeSession) })
+              resumeRetriesFresh({
+                resume,
+                out,
+                runtime: rt.id,
+                // The CLI's own word that it found the conversation: Claude's
+                // init, Codex's `thread.started` naming the pinned thread.
+                sawInit: rt.id === 'codex' ? Boolean(codexResumeId && seenThreadId === codexResumeId) : Boolean(seenClaudeSession),
+              })
             ) {
-              out = await runCli({ ...turnArgs, resume: false });
+              cliFailed = false;
+              badExit = null;
+              contractSent = true;
+              out = await runCli({ ...turnArgs, resume: false, system: turnSystem });
             }
           } finally {
             // The CLI has stopped printing, so stop relaying. The LINE itself
@@ -896,7 +934,10 @@ export function createWorkSessionTurns({
           // before it ever touches disk — it later rides in argv as
           // `resume <id>` — and best-effort, like the runtime pin: an
           // unwritable marker just means the tab runs fresh next turn.
-          if (rt.id === 'codex' && seenThreadId && CODEX_THREAD_RE.test(seenThreadId)) pinId(sessionMetaPath(dir.wt, RESUME_MARKERS.codex, job.sessionId), CODEX_THREAD_RE, seenThreadId);
+          if (rt.id === 'codex' && seenThreadId && CODEX_THREAD_RE.test(seenThreadId)) {
+            pinId(sessionMetaPath(dir.wt, RESUME_MARKERS.codex, job.sessionId), CODEX_THREAD_RE, seenThreadId);
+            recordCodexContract(contractPath, seenThreadId, turnSystem, contractSent);
+          }
           /**
            * A PERSON INTERRUPTED IT. Whatever the CLI printed was cut off, so
            * it is not an answer: the turn settles `interrupted`, with what it
@@ -974,7 +1015,7 @@ export function createWorkSessionTurns({
               artifactFailure = `Artifact delivery is incomplete: ${error.message}`;
             }
           }
-          const spoke = answer.length > 0 && !loginFailure;
+          const spoke = answer.length > 0 && !loginFailure && !cliFailed && !badExit;
           await settleTurn(job.id, {
               ...attempt,
             ok: spoke && !artifactFailure,
@@ -997,6 +1038,7 @@ export function createWorkSessionTurns({
           });
           if (spoke) ok(`${c.cyan('tab')} ${c.dim('— replied in the session')}`);
           else if (loginFailure) warn('codex answered 401 for this session turn — settled as failed, in words');
+          else if (answer && (cliFailed || badExit)) warn(`the CLI reported its turn failed${badExit ? ` (${badExit})` : ''} — settled as failed, in its words`);
           else warn('session turn produced no output — settled as failed');
         } catch (e) {
           if (toolLog) toolLog.state = 'partial';
@@ -1072,7 +1114,7 @@ export function createWorkSessionTurns({
       if (run) {
         if (!run.interrupted) note(`${c.cyan('tab')} ${c.dim('— interrupting the running turn')}`);
         run.interrupted = true;
-        for (const ch of run.spawned) killGroup(ch);
+        for (const ch of run.spawned) interruptThenKill(ch);
         continue;
       }
       if (pendingWorkReports.has(id) || settledRecently(id) || notRunningSaid.has(id)) continue;

package/bin/lib/worktreeDiff.mjs

@@ -191,14 +191,38 @@ function branchCommits(wt, base) {
         .split('\n')
         .filter(Boolean)
     );
+    // EVERY COMMIT ON THE BRANCH, NOT ONLY THE CARDED ONES (2026-10-19). A
+    // Terminal tab's prompt says a commit for no named card needs no trailer,
+    // so its own commits were never reported and the rail said "N commits
+    // ahead" with none of them listed or openable. One more exec names the
+    // rest — sha, author, date and the folded subject, NUL-separated (git
+    // allows a NUL in none of them) and one per line (`%s` folds its
+    // newlines), keyed by sha from the rev-list set with the first line
+    // winning, the same membership rule the trailered read keeps. No body and
+    // no numstat: the rail lists a commit by its subject, and its patch is the
+    // commit-diff job's.
+    const listed = new Map();
+    for (const line of git(['log', '--no-merges', '-n', String(MAX_COMMITS), '--format=%H%x00%an%x00%aI%x00%s', `${base}..HEAD`], wt).split('\n')) {
+      const [sha, author, at, subject] = line.split('\0');
+      if (real.includes(sha) && !listed.has(sha)) listed.set(sha, { author, at, subject });
+    }
     for (const sha of real) {
-      if (!named.has(sha)) continue;
       // A commit we cannot stat (a numstat past the pipe's buffer) still names
       // its cards — send it with zero counts rather than drop it, as before.
-      const rec = commitRecord(sha, wt, { numstat: true }) ?? commitRecord(sha, wt);
-      if (!rec) continue;
-      const taskIds = taskIdsFromMessage(stripDelims(rec.body));
-      if (taskIds.length === 0) continue;
+      const rec = named.has(sha) ? commitRecord(sha, wt, { numstat: true }) ?? commitRecord(sha, wt) : null;
+      const taskIds = rec ? taskIdsFromMessage(stripDelims(rec.body)) : [];
+      if (taskIds.length === 0) {
+        const row = listed.get(sha);
+        if (!row) continue;
+        out.push({
+          sha: sha.slice(0, 64),
+          subject: stripDelims(row.subject).slice(0, 200),
+          author: stripDelims(row.author).slice(0, 80),
+          at: stripDelims(row.at).slice(0, 40),
+          taskIds: [],
+        });
+        continue;
+      }
       out.push({
         // Clamped to the server's zod caps, same rule as everything else in
         // this file: one over-cap string 400s the whole batch.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "flowviant",
-  "version": "0.125.6",
+  "version": "0.125.9",
   "description": "Run your own coding CLIs as build agents for Flowviant \u2014 Claude Code, Codex or Antigravity, on your own credentials. Holds your sessions, keeps a worktree per tab, and ships branches on your word.",
   "type": "module",
   "bin": {