pwn 0.5.656 → 0.5.657
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.
- checksums.yaml +4 -4
- data/lib/pwn/ai/agent/loop.rb +11 -7
- data/lib/pwn/version.rb +1 -1
- metadata +1 -1
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: af75906b3c582dea27fb9ec31f19c7d3774aabcc928549cee27df74ad75220f3
|
|
4
|
+
data.tar.gz: c3424966d7b298c78b3edac59a777ee8e927f130ecefbf84fd0a46371256900e
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: 396a1f16a96e071396689cdfde74364ee9bae88544275f037137007b5ddd94dc1fc37f181a99ef239db5339b5fb24ecf750605ba67dd28bd62969ec9360ed5f6
|
|
7
|
+
data.tar.gz: 976c310a1e93bac33b383bcb7702a6350f2268062e71a0bb73ee28454d6ea23798fc654744b88dad048472d721d2417713d5664e81f39e819ddd2ccefe51c555
|
data/lib/pwn/ai/agent/loop.rb
CHANGED
|
@@ -146,12 +146,13 @@ module PWN
|
|
|
146
146
|
# is "iteration budget exhausted" ×N; more headroom only produces more
|
|
147
147
|
# empty terminal failures for ORM/PRM/DPO.
|
|
148
148
|
if budget_exhaustion_hot?
|
|
149
|
-
# P17
|
|
150
|
-
#
|
|
151
|
-
#
|
|
152
|
-
#
|
|
153
|
-
|
|
154
|
-
|
|
149
|
+
# P17 — always 8 for ALL engines while budget fingerprints dominate.
|
|
150
|
+
# P28 remote runway (40) applies only to W3 overconf above; stacking
|
|
151
|
+
# a 40 hot-cap on top re-opens multi-dozen-iter thrash and is the
|
|
152
|
+
# 2026-08-04 regression (22 exhaust hits / day). Finish-under-8 is
|
|
153
|
+
# the sustained drop; incomplete_final? keeps multi-step autonomy
|
|
154
|
+
# inside that budget without polite handoffs.
|
|
155
|
+
n = [n, 8].min
|
|
155
156
|
end
|
|
156
157
|
n
|
|
157
158
|
rescue StandardError
|
|
@@ -634,7 +635,9 @@ module PWN
|
|
|
634
635
|
# nothing the user (or ORM) can use.
|
|
635
636
|
# P17 deepen — when budget_hot, force text-only on the LAST TWO
|
|
636
637
|
# iters so a final tool_calls batch cannot burn the terminal slot.
|
|
637
|
-
|
|
638
|
+
# P17 deepen³ — under hot, force text-only on last THREE of the
|
|
639
|
+
# 8-iter cap so a late tool binge cannot burn every salvage slot.
|
|
640
|
+
text_only_iters = budget_exhaustion_hot? ? 3 : 1
|
|
638
641
|
last_iter = (i >= max_iters - text_only_iters)
|
|
639
642
|
if last_iter
|
|
640
643
|
tag = i >= max_iters - 1 ? 'FINAL ITERATION' : 'PENULTIMATE — wrap up'
|
|
@@ -830,6 +833,7 @@ module PWN
|
|
|
830
833
|
|
|
831
834
|
P28 autonomy: incomplete-final detector refuses mid-goal handoffs;
|
|
832
835
|
W3 overconf max_iters_cap is 40 on remote engines (8 on ollama).
|
|
836
|
+
P17 budget-hot always caps max_iters to 8 for ALL engines (not only ollama).
|
|
833
837
|
|
|
834
838
|
#{self}.authors
|
|
835
839
|
USAGE
|
data/lib/pwn/version.rb
CHANGED