@nexus-cortex/cli 4.124.27 → 4.124.28
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.cortex/orient +24 -0
- package/.env.defaults +9 -0
- package/package.json +2 -2
package/.cortex/orient
CHANGED
|
@@ -85,6 +85,30 @@ CHANGED
|
|
|
85
85
|
fi
|
|
86
86
|
BN=$(wc -l < "$W/.cortex/baseline.cksum" 2>/dev/null | tr -d ' ')
|
|
87
87
|
echo "-- baseline: ${BN:-0} files recorded at start; \`sh .cortex/changed\` lists files you modified/added/deleted. Before finishing, run it and revert every change the task did not require."
|
|
88
|
+
elif [ "${CORTEX_ORIENT_BASELINE:-0}" = "auto" ] && ! { command -v git >/dev/null 2>&1 && git -C "$W" rev-parse --is-inside-work-tree >/dev/null 2>&1; }; then
|
|
89
|
+
# auto (general work, 2026-09-27): a git repo already has `git status`, so skip it; otherwise keep the baseline OUTSIDE the project
|
|
90
|
+
# (the user's ~/.cortex, else /tmp), keyed by the workspace path, so nothing is written into the tree being worked on.
|
|
91
|
+
KEY=$(printf %s "$W" | cksum | cut -d' ' -f1)
|
|
92
|
+
BD="${HOME:-/tmp}/.cortex/baselines/$KEY"
|
|
93
|
+
mkdir -p "$BD" 2>/dev/null && [ -w "$BD" ] || { BD="${TMPDIR:-/tmp}/cortex-baselines/$KEY"; mkdir -p "$BD" 2>/dev/null; }
|
|
94
|
+
{ printf '#!/bin/sh\n# Lists files modified / added / deleted in %s since orient recorded its start state.\nW=%s\nB=%s\n' "$W" "'$W'" "'$BD/baseline.cksum'"
|
|
95
|
+
cat <<'CHANGED'
|
|
96
|
+
cd "$W" || exit 0
|
|
97
|
+
[ -f "$B" ] || { echo "no baseline recorded ($B missing)"; exit 0; }
|
|
98
|
+
T=$(mktemp 2>/dev/null || echo /tmp/cortex-now.$$)
|
|
99
|
+
find . \( -path ./.git -o -path ./.cortex -o -name node_modules -o -name __pycache__ -o -name .venv -o -name venv -o -name .cache -o -name .pytest_cache -o -name .mypy_cache \) -prune -o -type f -size -20M -print 2>/dev/null | head -20000 | tr '\n' '\0' | xargs -0 cksum 2>/dev/null > "$T"
|
|
100
|
+
awk 'function k( f) { f=$0; sub(/^[0-9]+[ \t]+[0-9]+[ \t]+/, "", f); return f }
|
|
101
|
+
NR==FNR { b[k()]=$1" "$2; next } { n[k()]=$1" "$2 } END {
|
|
102
|
+
for (f in n) { if (!(f in b)) print "added " f; else if (b[f] != n[f]) print "modified " f }
|
|
103
|
+
for (f in b) if (!(f in n)) print "deleted " f }' "$B" "$T" | sort -k2
|
|
104
|
+
rm -f "$T"
|
|
105
|
+
CHANGED
|
|
106
|
+
} > "$BD/changed" 2>/dev/null
|
|
107
|
+
if [ ! -f "$BD/baseline.cksum" ]; then
|
|
108
|
+
( cd "$W" && find . \( -path ./.git -o -path ./.cortex -o -name node_modules -o -name __pycache__ -o -name .venv -o -name venv -o -name .cache -o -name .pytest_cache -o -name .mypy_cache \) -prune -o -type f -size -20M -print 2>/dev/null | head -20000 | tr '\n' '\0' | xargs -0 cksum 2>/dev/null > "$BD/baseline.cksum" )
|
|
109
|
+
fi
|
|
110
|
+
BN=$(wc -l < "$BD/baseline.cksum" 2>/dev/null | tr -d ' ')
|
|
111
|
+
echo "-- baseline: ${BN:-0} files recorded at start (not a git repo); \`sh $BD/changed\` lists files you modified/added/deleted. Before finishing, run it and revert changes to existing files the task did not require."
|
|
88
112
|
fi
|
|
89
113
|
# Bare-box hint (2026-09-03, operator-simplified): a task container may ship WITHOUT the language, compiler,
|
|
90
114
|
# package or library the objective needs — on purpose (the agent phase is bare; internet is usually allowed).
|
package/.env.defaults
CHANGED
|
@@ -1766,6 +1766,8 @@ CORTEX_ORIENT_BOOTSTRAP=1
|
|
|
1766
1766
|
# the MiMo validation cell (4.124.26, orient-v2 arm vs spec-tests arm vs the 4.124.20 K=3 control 59.9%) decides it.
|
|
1767
1767
|
# [LEDGER 2026-09-27 @4.124.26] MiMo tb64 validation arm o2: 43/64 = 67.2% vs control 59.9% (K=3), +7.3 pts, P=0.019 vs per-task control
|
|
1768
1768
|
# rates (one run). Positive single run — replicate (K=2 more) before flipping the default (r-mimo64-val-2026-09-27.md).
|
|
1769
|
+
# [LEDGER 2026-09-27 @4.124.27] NOT REPLICATED: five sweep arms on o2 + edit-scope pair pooled 194/320 = 60.6% (control 59.9%); o2's 43/64 was
|
|
1770
|
+
# most likely a high draw (the pair could offset it — no o2-only replicate). Stays DARK.
|
|
1769
1771
|
#CORTEX_ORIENT_V2=
|
|
1770
1772
|
# CORTEX_ORIENT_BASELINE / CORTEX_LIFT_PLAN_SCOPE — edit-scope pair (2026-09-27, r-mimo64-val-2026-09-27.md §8). MiMo tb64: the grader's
|
|
1771
1773
|
# protected-file guard rejected 4-9 sessions per run (31 across 5 runs, ~10-14% of tasks) — a file outside the task's repair set was
|
|
@@ -1775,6 +1777,13 @@ CORTEX_ORIENT_BOOTSTRAP=1
|
|
|
1775
1777
|
# are modified; ends with the changed-files check.
|
|
1776
1778
|
# [LEDGER 2026-09-27 @4.124.27] DARK (both default off) — fixture-tested (dash; start state kept across orient re-runs; spaces in names);
|
|
1777
1779
|
# efficacy UNMEASURED: the MiMo reasoning sweep carries both ON with an o2+both base arm.
|
|
1780
|
+
# [LEDGER 2026-09-27 in tree] general-work variants (operator: "how will this work in interactive sessions or other benchmarks"):
|
|
1781
|
+
# CORTEX_LIFT_PLAN_SCOPE=adaptive (named files bind; tests/runners/fixtures/data in scope only when the task asks; revert EXISTING files only)
|
|
1782
|
+
# and CORTEX_ORIENT_BASELINE=auto (skip git repos; baseline stored outside the project in ~/.cortex/baselines, else /tmp). true / 1 are
|
|
1783
|
+
# unchanged = what the MiMo sweep measures. Scope only reaches headless sessions: the lift planner is skipped on a TTY unless
|
|
1784
|
+
# CORTEX_LIFT_PLAN_INTERACTIVE=true. DARK; validate adaptive/auto on TB2.1 or TB4.0 (build-style tasks) before any default.
|
|
1785
|
+
# [LEDGER 2026-09-27 @4.124.27] MiMo sweep (r-mimo64-sweep-2026-09-27.md): with SCOPE=true + BASELINE=1 on the o2 config, protected-file rejects
|
|
1786
|
+
# fell to 13/320 (4.1%) vs 31/320 (9.7%) in the five prior runs — mechanism works; pass rate unchanged (5 arms pooled 60.6% vs control 59.9%). DARK.
|
|
1778
1787
|
#CORTEX_ORIENT_BASELINE=
|
|
1779
1788
|
#CORTEX_LIFT_PLAN_SCOPE=
|
|
1780
1789
|
# [LEDGER 2026-09-02 @4.89.0] code default 30 (4.88.0) — 20 made the break rung (2N=24) unreachable; preserves 40%/80% density semantics.
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@nexus-cortex/cli",
|
|
3
|
-
"version": "4.124.
|
|
3
|
+
"version": "4.124.28",
|
|
4
4
|
"description": "Nexus Cortex CLI - Terminal interface for multi-provider LLM orchestration",
|
|
5
5
|
"main": "dist/index.js",
|
|
6
6
|
"type": "module",
|
|
@@ -19,7 +19,7 @@
|
|
|
19
19
|
"prepack": "node ../../scripts/copy-pkg-cortex-scaffold.mjs"
|
|
20
20
|
},
|
|
21
21
|
"dependencies": {
|
|
22
|
-
"@nexus-cortex/core": "4.124.
|
|
22
|
+
"@nexus-cortex/core": "4.124.28",
|
|
23
23
|
"chalk": "^4.1.2",
|
|
24
24
|
"cli-spinners": "^2.9.0",
|
|
25
25
|
"commander": "^11.0.0",
|