@nexus-cortex/cli 4.124.32 → 4.124.34

package/.cortex/inventory

@@ -41,8 +41,19 @@ W = os.path.abspath(sys.argv[1])
 SKIP_DIRS = {'.git', 'node_modules', '__pycache__', '.venv', 'venv', '.cortex', '.cache', '.pytest_cache', '.mypy_cache', '.tox', '.npm',
              '.gradle', '.m2', 'site-packages', 'dist-packages', '.idea', '.vscode', '.addon-tools'}
 MAX_WALK = 60000
-out = []
-p = out.append
+# Output is bucketed per section and emitted DECISIVE-FIRST (2026-09-29): the model reads orient through `| head -50/-100` in ~93% of sessions
+# (504/541, .bench/mimo/boot-prompt-audit.py) and cut the inventory in 8/17 cw sessions — so what decides a task (how it is checked, which
+# copy of the code runs, missing deps, services, data outside the workdir) comes before the long per-file listing. Computation order is unchanged.
+SECTIONS, CUR = {}, ['header']
+EMIT = ['header', 'checked', 'python', 'node', 'services', 'outside', 'structure', 'files', 'env']
+
+
+def sec(name):
+    CUR[0] = name
+
+
+def p(line):
+    SECTIONS.setdefault(CUR[0], []).append(line)
 
 
 def rel(x):
@@ -191,6 +202,7 @@ counts = ', '.join(f'{r} {len(v)}' for r, v in sorted(by_role.items(), key=lambd
 p('by role: ' + counts)
 
 # ---------------------------------------------------------------- 1. verification + contracts (PRIORITY — never capped)
+sec('checked')
 chk = by_role.get('checker/runner', []); con = by_role.get('contract/spec', [])
 if chk or con:
     p('-- HOW THIS TASK IS CHECKED (task-shipped checkers/runners and contracts — read these before building) --')
@@ -213,6 +225,7 @@ if chk or con:
         p('  test dirs: ' + ', '.join(shown) + (f' (+{len(tests)-12} more — find . -type d -name "test*")' if len(tests) > 12 else ''))
 
 # ---------------------------------------------------------------- 2. structure: every dir with counts (capped with marker)
+sec('structure')
 p('-- STRUCTURE (dir: files directly inside) --')
 dl = sorted(d for d in dirs if d != '.')
 top = [d for d in dl if d.count('/') <= 1]
@@ -226,6 +239,7 @@ if deep:
     p(f'  ({len(deep)} deeper directories — find . -mindepth 3 -type d)')
 
 # ---------------------------------------------------------------- 3. files by role with a peek (priority roles in full, bulk capped)
+sec('files')
 CAPS = {'source': 40, 'doc': 30, 'data': 40, 'manifest/config': 30, 'test': 20, 'other': 20}
 ORDER = ['doc', 'manifest/config', 'source', 'data', 'test', 'other']
 for r in ORDER:
@@ -244,6 +258,7 @@ for r in ORDER:
                f'list: find . -type f | grep -v -e .git/ -e node_modules  (role here = {r})'))
 
 # ---------------------------------------------------------------- 4. python: which interpreter, which copy of the code runs, missing deps
+sec('python')
 def run(cmd, timeout=20):
     try:
         r = subprocess.run(cmd, capture_output=True, text=True, timeout=timeout, cwd=W)
@@ -307,6 +322,7 @@ if declared:
         p(f'  all {len(declared)} declared python dependencies are installed')
 
 # ---------------------------------------------------------------- 5. node
+sec('node')
 pj = [f for f, _ in by_role.get('manifest/config', []) if os.path.basename(f) == 'package.json' and f.count('/') <= 2 and 'node_modules' not in f]
 for f in pj[:3]:
     try:
@@ -321,6 +337,7 @@ for f in pj[:3]:
       (f', NOT installed: {", ".join(miss[:20])}' + (' …' if len(miss) > 20 else '') if miss else (', node_modules present' if os.path.isdir(nm) else ', no node_modules')))
 
 # ---------------------------------------------------------------- 6. task environment (from the container's own init env; secrets redacted)
+sec('env')
 BOILER = {'PATH', 'HOME', 'HOSTNAME', 'TERM', 'SHLVL', 'PWD', 'OLDPWD', '_', 'LANG', 'LC_ALL', 'DEBIAN_FRONTEND', 'container', 'GPG_KEY',
           'PYTHON_VERSION', 'PYTHON_SHA256', 'PYTHON_PIP_VERSION', 'PYTHON_GET_PIP_URL', 'PYTHON_GET_PIP_SHA256', 'PYTHON_SETUPTOOLS_VERSION',
           'NODE_VERSION', 'YARN_VERSION', 'LS_COLORS', 'SSL_CERT_DIR', 'SSL_CERT_FILE', 'NVIDIA_VISIBLE_DEVICES', 'NVIDIA_DRIVER_CAPABILITIES'}
@@ -349,6 +366,7 @@ if src_note is not None:
         if len(keep) > 40: p(more(len(keep) - 40, 'variables', 'cat /proc/1/environ | tr "\\0" "\\n"'))
 
 # ---------------------------------------------------------------- 7. services + paths outside the workdir
+sec('services')
 hosts = []
 try:
     for l in open('/etc/hosts'):
@@ -372,6 +390,7 @@ if hosts or ports or urlvars:
     if hosts: p('  other hosts on this network: ' + ', '.join(sorted(set(hosts))[:20]))
     if ports: p('  listening locally: ' + ', '.join(str(x) for x in sorted(set(ports))[:20]))
     for u in urlvars[:12]: p('  ' + u)
+sec('outside')
 outside = []
 for d in ('/shared', '/data', '/mnt', '/srv', '/workspace', '/input', '/inputs', '/output', '/outputs', '/reference', '/resources', '/task', '/opt/task'):
     if os.path.isdir(d) and not W.startswith(d) and not d.startswith(W):
@@ -386,5 +405,5 @@ for d in ('/shared', '/data', '/mnt', '/srv', '/workspace', '/input', '/inputs',
         if n: outside.append(f'{d} ({n}{"+" if capped else ""} files)')
 if outside:
     p('-- OUTSIDE THE WORKDIR (task data may live here) -- ' + ', '.join(outside) + ' — list: find <dir> -type f | head -100')
-print('\n'.join(out))
+print('\n'.join(line for name in EMIT for line in SECTIONS.get(name, [])))
 PYINV

package/.cortex/orient

@@ -22,8 +22,11 @@ fi
 # files with a content peek, which copy of the code imports, missing deps, task env, services, data outside the workdir) replaces the
 # workspace map and the v2 verification section; the tooling / git / baseline / setup lines below still print. Unset = unchanged output.
 INV=""; if [ "${CORTEX_INVENTORY:-0}" = "1" ]; then for c in "$(dirname "$0")/inventory" "${CORTEX_ROOT:-/nonexistent}/.cortex/inventory"; do [ -f "$c" ] && { INV="$c"; break; }; done; fi
+INV_OUT=""
 if [ -n "$INV" ]; then
-  sh "$INV" "$W" 2>/dev/null
+  # Captured here, PRINTED after the setup line below (2026-09-29): the model reads orient through `| head -50/-100` in ~93% of sessions,
+  # and with the inventory first 8/17 cw sessions lost orient's own short lines (git, the `-- baseline:` revert instruction, setup).
+  INV_OUT=$(sh "$INV" "$W" 2>/dev/null)
 else
 echo "== WORKSPACE MAP: $W =="
 ls -1A "$W" 2>/dev/null | head -40
@@ -148,7 +151,22 @@ if [ "${CORTEX_ORIENT_BOOTSTRAP:-0}" = "1" ] && { ! command -v python3 >/dev/nul
   PM=""; for m in apt-get apk dnf yum pacman zypper brew; do command -v "$m" >/dev/null 2>&1 && { PM="$m"; break; }; done
   ROOT=$([ "$(id -u 2>/dev/null)" = "0" ] && echo yes || echo no)
   GUARD=""; [ "$V2" = "1" ] && GUARD=" Install into the system or a venv, never into the task's own source or vendored tree."
+  # With the inventory on and a python interpreter present, the inventory has already checked the workspace manifests (requirements/
+  # pyproject/package.json) against what is installed. Replace the speculative "may be missing" paragraph with those facts
+  # (r-orient-deficiencies-2026-09-29 §3.3: "print only the MISSING ones … keep the generic line otherwise"); keep the install discipline
+  # (once, cached, never into the task's source/vendored tree). No manifest, no python, or no inventory = the generic line below, unchanged.
+  INV_MF=""; INV_MISS=""
+  if [ -n "$INV_OUT" ] && { command -v python3 >/dev/null 2>&1 || command -v python >/dev/null 2>&1; }; then
+    printf '%s\n' "$INV_OUT" | grep -qE "declared but NOT installed|declared python dependencies are installed|^-- NODE " && INV_MF=1
+    printf '%s\n' "$INV_OUT" | grep -qE "declared but NOT installed|^-- NODE .*NOT installed:" && INV_MISS=1
+  fi
+  if [ -n "$INV_MF" ] && [ -n "$INV_MISS" ]; then
+    echo "-- setup: the workspace manifests declare dependencies that are NOT installed (listed in the inventory below; pkg-mgr=${PM:-none}, root=$ROOT). Install what the task needs ONCE and reuse it, with a fast cached installer for the language (uv for Python, bun for JS) — do not re-download a large dependency twice.$GUARD Pin versions, then verify with a real run before finishing."
+  elif [ -n "$INV_MF" ]; then
+    echo "-- setup: the declared dependencies are installed (see the inventory below). If you still need a package, install it once$( [ "$V2" = "1" ] && echo " into the system or a venv, never into the task's own source or vendored tree" )."
+  else
   echo "-- setup: this box may be missing a tool/language/library the task needs (pkg-mgr=${PM:-none}, root=$ROOT). If so, INSTALL what you need yourself following standard practice (the system package manager, or the language's own installer); install ONCE and reuse it, and prefer a fast cached installer for the language (uv for Python, bun for JS) — do not re-download a large dependency twice.$GUARD Pin versions, then verify with a real run before finishing. An empty box is part of the task, not an error."
+  fi
 fi
 # orient v2 — VERIFICATION ENTRY POINTS: how this task's correctness is checked, so work is built and verified against the real
 # criteria, not the agent's own guess (MiMo K=3: in 57/79 failures the agent's own test passed on a wrong solution). Same usage-line
@@ -169,6 +187,8 @@ for f in pytest.ini tox.ini setup.cfg pyproject.toml Makefile package.json; do
 done
 TD=$(find . -maxdepth 3 -type d \( -name tests -o -name test \) 2>/dev/null | grep -vE "node_modules|/\.|site-packages" | head -4 | tr '\n' ' ')
 fi
+# The inventory (captured above) prints AFTER orient's short facts — tooling, git, baseline, setup — so a `| head -N` read keeps them.
+[ -n "$INV" ] && printf '%s\n' "$INV_OUT"
 if [ -n "$VE$TC$TD" ] && [ -z "$INV" ]; then
   echo "== VERIFICATION ENTRY POINTS (build to and verify against these — not a test you invent) =="
   for f in $VE; do

package/.env.defaults

@@ -872,6 +872,37 @@ CORTEX_ENDTURN_GATE=true
 # [LEDGER 2026-09-15 @4.108.21] BUILT + unit-tested (timeBudget.test.ts bands/line/levers; TurnEndGuards.test.ts open-items shapes).
 #CORTEX_BUDGET_VISIBILITY=
 #CORTEX_BUDGET_CONTINUE_MIN_REMAINING=
+# CORTEX_TURN_STATUS — HB-TURN-STATUS (DARK, 2026-09-29). TB4.0 early self-finish root-cause (r-tb4-early-finish-rootcause-2026-09-29):
+# 141/162 fails finished with >50% of the budget left; the model's sense of elapsed time ran 6-10x fast and it gave up "out of time"
+# at <1 h of 8 h — the only clock it saw was the R151 WALL BUDGET line, once per 10% band. off (default; unset = byte-identical requests)
+# | auto (on when CORTEX_TURN_DEADLINE_MS > 0 — the bench always sets it) | on (always; elapsed only without a deadline). On EVERY
+# tool-result round (sendMessage + streamMessage, every handleToolCalls path) one line is appended to the TAIL of the round's newest
+# tool_result: <system-reminder>STATUS: elapsed 0h12m of 7h12m (3%), ~7h00m left; context ~84K of 1M tokens (8%); bg: <id> "<cmd>"
+# 4m12s running</system-reminder> (context = last API prompt_tokens + request growth, R132 anchor; bg = the 3 most recent entries of
+# executors' BackgroundProcessRegistry — the Bash run_in_background / BashOutput / KillShell registry). CACHE: persisted in the in-memory
+# history like the band line, so every earlier request is a byte-identical prefix; only the new tail varies. The band line stays.
+# Event: one turn_status row per turn {lines, mode, streaming}.
+# [LEDGER 2026-09-29 @4.124.33] DARK — BUILT + unit-tested (turnStatus.test.ts line/levers/bg/placement; turnStatus.integration.test.ts
+# both loops: STATUS tail on every continuation, prior request prefixes byte-identical, unset/auto-without-deadline = no STATUS). NO bench arm yet.
+#CORTEX_TURN_STATUS=
+# CORTEX_EFFORT_RAMP — HB-EFFORT-RAMP (DARK, 2026-09-30). TB4.0 t4cw traces: the action model hit the 65,536-token output wall right
+# after reading the task (photonic r2 call 3, protein r1 call 6), trying to solve the routing plan / hidden rule in its head before writing
+# code; it only started acting once forced to low. The first N MAIN action calls of a turn (0 = initial request; retry/synthesis calls
+# don't count) run at the ramp level, then the configured effort. off (default) | <level>[:<calls>] (low:8) | on (= low:8). DeepSeek
+# levels are low | high | max (medium = high). Precedence: R153 backoff > effort pulse > ramp > request param > card. Events: effort_ramp
+# {phase start|end}. CAUTION: GLOBAL low was harmful on TB4.0 (e6: 3/29 vs high 10/29) — this lever only covers the opening calls.
+# [LEDGER 2026-09-30 @4.124.34] DARK — BUILT + tested (effortLevers.test.ts; effortLevers.integration.test.ts both loops: calls 0..N-1 at
+# the ramp level, then the configured effort; unset = untouched). NO bench arm yet.
+#CORTEX_EFFORT_RAMP=
+# CORTEX_COMPUTE_NUDGE / _TOKENS / _COOLDOWN / _MAX — HB-COMPUTE-NUDGE (DARK, 2026-09-30). Same traces: steps that DID act often carried
+# 35K-52K reasoning tokens first; photonic r2's productive stretch began when it ran the task's checker. After a round whose response
+# reasoned past _TOKENS (default 16000), one line is appended to the tail of the newest tool_result (the HB-TURN-STATUS carrier — earlier
+# request prefixes stay byte-identical): "COMPUTE, DON'T DELIBERATE: your last step spent ~NK reasoning tokens ... write and run a short
+# script or the task's own checker ...". Cooldown _COOLDOWN rounds (3), cap _MAX per turn (6). off (default) | on. Event compute_nudge.
+# [LEDGER 2026-09-30 @4.124.34] DARK — BUILT + tested (unit + both-loop integration: fires on a heavy step, cooldown respected; unset =
+# never). NO bench arm yet.
+#CORTEX_COMPUTE_NUDGE=
+#CORTEX_COMPUTE_NUDGE_TOKENS=
 # (no lever) R152 HB-EMPTY-COMPLETION-RETRY (4.108.22, 2026-09-15): during a 27-min DeepSeek incident (1:30-1:57 AM PT) chat/completions
 # returned bodies that parsed to null; the SDK handed them back, the translation layer dereferenced `.choices`, and the tool loop ended the turn
 # on the TypeError — 30 sessions across tb21-a21 A/B + tb4-flash-v3 (largest-eigenval died at iteration 1). R150's ladder did not apply (a null
@@ -910,6 +941,12 @@ CORTEX_ENDTURN_GATE=true
 # from it) and the counter covers the retry call plus TURNS continuations.
 #CORTEX_REASONING_EXHAUST_BACKOFF=
 #CORTEX_REASONING_EXHAUST_BACKOFF_TURNS=
+# CORTEX_REASONING_EXHAUST_LADDER — R153c (2026-09-30, DEFAULT CHANGE). The R153 backoff stepped high/max → 'medium', which DeepSeek
+# runs as HIGH (reasoning_effort ladder low | high | max) — the first backoff was a NO-OP on DeepSeek (t4cw photonic r2: wall at high,
+# "medium" wall again, only then low). provider (default): step down the provider's real ladder (DeepSeek high → low, max → high; others
+# unchanged low | medium | high) | legacy: the pre-R153c steps. reasoning_exhaustion events now carry {ladder, provider}.
+# [LEDGER 2026-09-30 @4.124.34] ON (bug fix) — unit-tested (effortLevers.test.ts). NO bench arm.
+#CORTEX_REASONING_EXHAUST_LADDER=
 # (no lever) R154 HB-IMAGE-REJECTION-HEAL (4.109.0, 2026-09-16): retro-console-soc generated a 128x128 8-bit GRAYSCALE PNG, called
 # ReadImage, and DeepSeek answered the next continuation with `400 .messages[290].image[0]: You have uploaded an unsupported image`;
 # the exception escaped after the tool batch, the already-executed guard left the loop, and the R29a synthesis failed on the same block —
@@ -1806,6 +1843,14 @@ CORTEX_ORIENT_BOOTSTRAP=1
 # end with the exact command to list the rest. Python-first, POSIX-sh fallback; fixture 0.37 s. Pair with CORTEX_STEER_INPUTS=full (the
 # default env-report caps of 2000-3000 chars would cut it — with a marker only under full).
 # [LEDGER 2026-09-29 in tree] DARK — built + fixture-tested; not released; efficacy UNMEASURED (TB4.0 f66-shape cell vs f66 control).
+# [LEDGER 2026-09-29 @4.124.32] released dark; MiMo cw1/cw2 validation launched. Next release: with the inventory on (python present + a
+# manifest found) orient's bootstrap `-- setup:` paragraph states the inventory's facts instead of "this box may be missing…" — missing
+# declared deps → install-once line (+ vendored-tree guard under v2); all installed → one short line; no manifest / no python → the
+# generic line unchanged (r-orient-deficiencies §3.3). Unset inventory = byte-identical (tested).
+# Next release also REORDERS for the model's own `| head -50/-100` (504/541 sessions): orient prints its short lines (tooling, git,
+# `-- baseline:`, setup) BEFORE the inventory, and the inventory emits decisive-first (checked → python → node → services → outside →
+# structure → files → env). Evidence: first 17 cw2 sessions on 4.124.32 — orient cut before its end in 8/17, losing the baseline revert
+# line (the scope pair's mechanism). Fixture: 0/4 short lines survived head -50 before, 4/4 after; line set unchanged.
 #CORTEX_INVENTORY=
 # CORTEX_STEER_INPUTS — steering-prompt input sizing (2026-09-29, r-steering-input-truncation-2026-09-29): every steering prompt sliced its
 # inputs with NO marker (judge workspace delta cut in 56% of verdicts, lift plan cut for the judge/exit planner in >=62%, the planner saw 2000

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nexus-cortex/cli",
-  "version": "4.124.32",
+  "version": "4.124.34",
   "description": "Nexus Cortex CLI - Terminal interface for multi-provider LLM orchestration",
   "main": "dist/index.js",
   "type": "module",
@@ -19,7 +19,7 @@
     "prepack": "node ../../scripts/copy-pkg-cortex-scaffold.mjs"
   },
   "dependencies": {
-    "@nexus-cortex/core": "4.124.32",
+    "@nexus-cortex/core": "4.124.34",
     "chalk": "^4.1.2",
     "cli-spinners": "^2.9.0",
     "commander": "^11.0.0",