pwn 0.5.697 → 0.5.699

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4218feff0995079766c3b31b6beec85c9b45b1f971ac0bc78a9794c67ea8016e
-  data.tar.gz: 5a47207b366764a60b8c560bdcf77c5b12288d7fdbeb3e2d74c655a8e92e12e3
+  metadata.gz: 5d7edeea1940ad56143190285565ba9bd76f4d78a47294238eafaad1d002686f
+  data.tar.gz: c3a594906ca8d239b8f5dfc3fe3343ec718bf1a71e3c85334b3b6db7e852b5fe
 SHA512:
-  metadata.gz: 326c061669305b7069c4a66e68d958b6935ef3a9c3f90e02ee1242ad6bedfed9e5e31170fd4ae8b7b9570ab7c63fb1e7198c0cc97a5a8e3eeb96400bc329c611
-  data.tar.gz: ed8cdb221e58fdb00ae799d902121e0f2b6f629871ceb458a54ef35ee5e2ad326070c431d9c17346db442e0d62374f90f85cba074a00967d3193d41753a091a7
+  metadata.gz: fee1edba85d565609b592498feff5eaf38bc01a4771af2a2e22f3719d1eac7faaf54c8f8772837358cf5cbf18db59736aee90fe4a7114c1e6353be423d42e2e8
+  data.tar.gz: 4ad32cb15b7bd78d0fa150cf63379ecf61184b84a902426a3581a558b48b17b0a7bc23048a2d2576155b538403e9472a1ebe542a8154eece95a36e4106e352f9

data/etc/default_skills/pwn/ai/http_retry/SKILL.md

@@ -35,6 +35,7 @@ PWN::AI::HttpRetry.timeout_s(opts)
 
 - `timeout_s`
 - `max_attempts`
+- `retryable`
 - `retry_after_s`
 - `report_event`
 - `authors`

data/lib/pwn/ai/agent/loop.rb

@@ -851,6 +851,17 @@ module PWN
           opts[:session_id].to_s
         end
 
+        private_class_method def self.engine_transient?(opts = {})
+          err = opts[:error]
+          return false unless err
+
+          return PWN::AI::HttpRetry.retryable?(error: err) if defined?(PWN::AI::HttpRetry) && PWN::AI::HttpRetry.respond_to?(:retryable?)
+
+          err.message.to_s.match?(/HTTP 50[234]|Gateway Time-out|stream absolute timeout/i)
+        rescue StandardError
+          false
+        end
+
         # E1 — did the environment change under this tool? If Metrics CUSUM
         # tripped for it in the last hour AND Extrospection.drift shows a
         # toolchain/net/repo change, blame the WORLD not the AGENT.
@@ -1416,8 +1427,10 @@ module PWN
           messages = opts[:messages]
           return messages unless messages.is_a?(Array) && messages.length > 12
 
-          keep_pairs = (agent_flag(key: :history_keep_tool_pairs, default: 6) || 6).to_i
-          max_chars  = (agent_flag(key: :history_tool_max_chars, default: 2_000) || 2_000).to_i
+          keep_pairs = opts[:keep_pairs]
+          keep_pairs = (agent_flag(key: :history_keep_tool_pairs, default: 6) || 6).to_i if keep_pairs.nil?
+          max_chars = opts[:max_chars]
+          max_chars = (agent_flag(key: :history_tool_max_chars, default: 2_000) || 2_000).to_i if max_chars.nil?
 
           head = []
           rest = messages.dup
@@ -2473,20 +2486,43 @@ module PWN
 
           turn_fails = Hash.new(0)
           escalated  = false
+          engine_blips = 0
           maybe_park_budget_scars!
           maybe_extinguish_parked!
 
           i = 0
           loop do
             i += 1
-            # 3.1 — compact history on local so tool dumps don't fill num_ctx
-            compact_history!(messages: messages) if local
+            # 3.1 — compact fat tool dumps so remote ReadTimeout hops do not
+            # retry the same 70-message payload (R1 201215 Anthropic 180s×5).
+            compact_history!(messages: messages)
             # English-task-as-primary: when plan_idx advanced, tell the model
             # which plain-English task is active before the next tool batch.
             inject_task_focus!(messages: messages, state: ts_state, request: request) unless skip_compass
 
             t0 = Time.now
-            msg = call_engine(messages: messages, tools: tools, ts_state: ts_state)
+            begin
+              msg = call_engine(messages: messages, tools: tools, ts_state: ts_state)
+            rescue StandardError => e
+              if engine_transient?(error: e)
+                engine_blips += 1
+                debug_progress(msg: "engine hop failed #{e.class}: #{e.message.to_s[0, 240]} blip=#{engine_blips}")
+                if engine_blips >= 5
+                  txt = "[pwn-ai] engine hop failed after #{engine_blips} tries: #{e.message.to_s[0, 240]}"
+                  debug_final_text!(text: txt)
+                  final_chars = txt.length
+                  return txt
+                end
+                compact_history!(messages: messages, keep_pairs: 3, max_chars: 800)
+                messages << {
+                  role: 'user',
+                  content: '[pwn-ai] engine hop failed (transient). Keep calling CORE_TOOLS. ' \
+                           "Do not stop. (#{e.message.to_s[0, 180]})"
+                }
+                next
+              end
+              raise
+            end
             engine_s += (Time.now - t0)
             PWN::Plugins::TTYSpinner.halt_all! if defined?(PWN::Plugins::TTYSpinner)
             if msg.nil?

data/lib/pwn/ai/agent/task_summarizer.rb

@@ -573,8 +573,8 @@ module PWN
         # model cannot stall the user's real answer, and stay quiet on timeout.
         private_class_method def self.sidecar_timeout
           n = (PWN::Env.dig(:ai, :agent, :sidecar_llm_timeout) if defined?(PWN::Env)).to_i
-          n = 20 if n < 5
-          n = 60 if n > 60
+          n = 180 if n < 30
+          n = 180 if n > 180
           n
         rescue StandardError
           20
@@ -661,6 +661,17 @@ module PWN
             ]
           end
 
+          artifact = goal_text[%r{(?:/(?:tmp|var|home|opt|usr)/\S+\.(?:pdf|html|md|json|txt|csv)|~/\S+\.(?:pdf|html|md|json|txt|csv))}i]
+          if artifact && goal_lc.match?(/\b(analy[sz]e|test|scan|report|generat|store|write|export)\b/)
+            tasks << 'Carry out the requested analysis using the named skills and live evidence'
+            tasks << "Write the requested report to #{artifact}"
+            if goal_lc.match?(/\b(json|ya?ml|table|csv|tsv)\b/)
+              fmt = goal_lc[/\b(json|ya?ml|table|csv|tsv)\b/]
+              tasks << "Present the results in #{fmt} format"
+            end
+            return tasks
+          end
+
           tasks << "Understand the request: #{truncate_goal(goal: goal_text)}"
           tasks << "Carry out the core work for: #{truncate_goal(goal: goal_text)}"
 

data/lib/pwn/ai/agent/tool_guard.rb

@@ -24,6 +24,7 @@ module PWN
         # Conservative bash-only constructs. POSIX `$(())` is allowed.
         BASHISM_RX = /
           \bPIPESTATUS\b
+          |\$\{?RANDOM\}?\b
           |\[\[(?:\s|\z)
           |(?:^|[\s;|&])source\s+\S
           |<\([^)]

data/lib/pwn/ai/agent/tools/shell.rb

@@ -65,7 +65,7 @@ PWN::AI::Agent::Registry.register(
 
     if PWN::AI::Agent::ToolGuard.bashism?(text: cmd) && !PWN::AI::Agent::ToolGuard.shell_bash?
       return PWN::AI::Agent::ToolGuard.invalid_payload(
-        hint: 'Command uses bash-only syntax (PIPESTATUS, [[ ]], process substitution, ' \
+        hint: 'Command uses bash-only syntax (PIPESTATUS, $RANDOM, [[ ]], process substitution, ' \
               'source, &>). This handler runs /bin/sh (dash) unless ' \
               'PWN::Env[:ai][:agent][:shell_bash]=true. Rewrite as POSIX or opt in to bash.'
       )

data/lib/pwn/ai/http_retry.rb

@@ -27,6 +27,15 @@ module PWN
         DEFAULT_MAX_ATTEMPTS
       end
 
+      public_class_method def self.retryable?(opts = {})
+        err = opts[:error]
+        msg = err.respond_to?(:message) ? err.message.to_s : err.to_s
+        msg = opts[:message].to_s if msg.empty?
+        msg.match?(/HTTP 50[234]|Gateway Time-out|stream absolute timeout|ReadTimeout|Net::ReadTimeout|Timed out reading/i)
+      rescue StandardError
+        false
+      end
+
       public_class_method def self.retry_after_s(opts = {})
         retry_count = [opts[:retry_count].to_i, 1].max
         retry_after = 0

data/lib/pwn/ai/ollama.rb

@@ -198,6 +198,21 @@ module PWN
           retry if retry_count < max_attempts
 
           nil
+        rescue StandardError => e
+          if PWN::AI::HttpRetry.retryable?(error: e) && retry_count + 1 < max_attempts
+            retry_count += 1
+            unless opts[:quiet]
+              PWN::AI::HttpRetry.report_event(
+                label: 'ollama', which_self: self, quiet: opts[:quiet],
+                http_method: http_method, rest_call: rest_call,
+                extra: "retryable attempt=#{retry_count}/#{max_attempts}",
+                error: e
+              )
+            end
+            sleep(PWN::AI::HttpRetry.retry_after_s(retry_count: retry_count) + rand(0.3..2.0))
+            retry
+          end
+          raise e
         end
       rescue RestClient::ExceptionWithResponse => e
         body = begin

data/lib/pwn/plugins/repl.rb

@@ -1284,8 +1284,9 @@ module PWN
                 raise
               rescue StandardError => e
                 PWN::Plugins::Log.note_exception!(error: e, where: 'native agent loop', which_self: PWN::Plugins::REPL) if defined?(PWN::Plugins::Log) && PWN::Plugins::Log.respond_to?(:note_exception!)
-                warn "[pwn-ai] native agent loop failed (#{e.class}: #{e.message}\n#{e.backtrace}); " \
-                     'falling back to legacy regex-ReAct.'
+                warn "[pwn-ai] native agent loop failed (#{e.class}: #{e.message.to_s.split("\n").first})"
+                request.replace('nil')
+                next
               ensure
                 PWN::Plugins::REPL.ready_tty!
               end

data/lib/pwn/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module PWN
-  VERSION = '0.5.697'
+  VERSION = '0.5.699'
 end

data/spec/lib/pwn/ai/agent/loop_spec.rb

@@ -424,9 +424,43 @@ describe PWN::AI::Agent::Loop do # rubocop:disable Metrics/BlockLength
       expect(src).to match(/note_exception!/)
       expect(src).to match(/rescue StandardError/)
       expect(src).to include("debug_progress(msg: 'engine returned no message')")
+      expect(src).to match(/engine_transient|retryable\?/)
       expect(src).not_to match(/dispatch #\{name\} ok_len=/)
     end
 
+    it 'keeps Loop.run going after a transient engine 504' do
+      src = File.read(described_class.method(:run).source_location.first)
+      expect(src).to match(/engine_transient\?|HttpRetry\.retryable\?/)
+      expect(src).to match(/engine hop failed|engine_blip/)
+    end
+
+    it 'compacts tool history for every engine, not only local' do
+      src = File.read(described_class.method(:run).source_location.first)
+      expect(src).to match(/compact_history!\(messages: messages\)/)
+      expect(src).not_to match(/compact_history!\(messages: messages\) if local/)
+    end
+
+    it 'shrinks tool bodies when history is fat' do
+      msgs = [
+        { role: 'system', content: 'sys' },
+        { role: 'user', content: 'goal' }
+      ]
+      8.times do |n|
+        msgs << { role: 'assistant', content: '', tool_calls: [{ id: "c#{n}" }] }
+        msgs << { role: 'tool', tool_call_id: "c#{n}", name: 'shell', content: ('x' * 4_000) }
+      end
+      out = described_class.send(:compact_history!, messages: msgs)
+      tool_bodies = out.select { |m| m[:role].to_s == 'tool' }
+      expect(tool_bodies.length).to be <= 6
+      expect(tool_bodies.map { |m| m[:content].to_s.length }.max).to be <= 2_100
+    end
+
+    it 'compacts again after a transient engine hop before retrying' do
+      src = File.read(described_class.method(:run).source_location.first)
+      hop = src[/if engine_transient\?.*?raise/m]
+      expect(hop).to match(/compact_history!/)
+    end
+
     it 'records a timeout increment mistake instead of treating success:true as ok' do
       tmp = Dir.mktmpdir
       stub_const('PWN::AI::Agent::Mistakes::MISTAKES_FILE', File.join(tmp, 'mistakes.json'))

data/spec/lib/pwn/ai/agent/signal_hygiene_spec.rb

@@ -36,6 +36,14 @@ describe 'P0 signal hygiene (handler + inbox + policy-cold + calibrate)' do
     expect(r[:hint].to_s).to match(/bash-only/i)
   end
 
+  it 'rejects $RANDOM under /bin/sh so the model rewrites instead of getting USER=' do
+    expect(PWN::AI::Agent::ToolGuard.bashism?(text: 'U="p4tester$RANDOM"; echo $U')).to eq(true)
+    handler = PWN::AI::Agent::Registry.lookup(name: 'shell').handler
+    r = handler.call(command: 'U="p4tester$RANDOM"; echo $U')
+    expect(r[:error]).to eq('invalid_payload')
+    expect(r[:hint].to_s).to match(/RANDOM|bash-only|POSIX/i)
+  end
+
   it 'does not refuse pwn_eval host-discovery as unauthorized' do
     src = File.read('/opt/pwn/lib/pwn/ai/agent/tools/ruby_eval.rb')
     expect(src).not_to match(/recon_blocked/)

data/spec/lib/pwn/ai/agent/task_summarizer_spec.rb

@@ -425,6 +425,16 @@ describe PWN::AI::Agent::TaskSummarizer do
       )
     end
 
+    it 'splits analyze-plus-artifact goals instead of pasting the full request twice' do
+      goal = 'Exhaustively analyze the running application for vulnerabilities. ' \
+             'Once complete, generate a complete PDF report and store in /tmp/container_pentest-p4.pdf'
+      tasks = described_class.fallback_decompose(goal: goal)
+      expect(tasks.length).to be >= 2
+      expect(tasks.join("\n")).to include('/tmp/container_pentest-p4.pdf')
+      expect(tasks.grep(/Understand the request:/).length).to eq(0)
+      expect(tasks.grep(/Carry out the core work for:/).length).to eq(0)
+    end
+
     it 'accepts injected :tasks / :llm_tasks without network I/O' do
       custom = ['step alpha', 'step beta', 'verify completion']
       tasks = described_class.plan(request: 'anything at all', tasks: custom)

data/spec/lib/pwn/ai/http_retry_spec.rb

@@ -2,6 +2,7 @@
 
 require 'spec_helper'
 require 'stringio'
+require 'rest-client'
 
 describe PWN::AI::HttpRetry do
   it 'defaults timeout to 180 and max_attempts to 5' do
@@ -14,6 +15,12 @@ describe PWN::AI::HttpRetry do
     expect(described_class.max_attempts(max_attempts: 3)).to eq(3)
   end
 
+  it 'treats gateway 504 and stream absolute timeouts as retryable' do
+    expect(described_class.retryable?(error: 'ERROR: Ollama HTTP 504 for api/chat: Gateway Time-out')).to eq(true)
+    expect(described_class.retryable?(error: 'ERROR: Ollama stream absolute timeout after 20s for api/chat')).to eq(true)
+    expect(described_class.retryable?(error: '400 Bad Request')).to eq(false)
+  end
+
   it 'tees emit lines into the open debug request log' do
     path = PWN::Plugins::Log.start_debug(
       tee: StringIO.new,
@@ -50,6 +57,7 @@ describe PWN::AI::HttpRetry do
       expect(src).not_to match(/timeout \|\|= 900/)
       expect(src).to match(/Exceptions::Timeout/)
       expect(src).to match(/HttpRetry\.report_event/)
+      expect(src).to match(/retryable\?|HTTP 50/) if name == 'PWN::AI::Ollama'
     end
   end
 end

data/spec/lib/pwn/plugins/repl_pwn_vault_spec.rb

@@ -174,4 +174,14 @@ describe 'pwn-ai debug final answer placement' do
     expect(hook).to include('→ pwn-ai →')
     expect(hook).to include('→ result')
   end
+
+  it 'does not fall through to regex-ReAct after a native Loop.run error' do
+    native_start = hook.index('final = PWN::AI::Agent::Loop.run')
+    rescue_at = hook.index('rescue StandardError', native_start)
+    expect(rescue_at).not_to be_nil
+    chunk = hook[rescue_at, 900]
+    expect(chunk).to match(/request\.replace\('nil'\)/)
+    expect(chunk).to match(/\bnext\b/)
+    expect(chunk).not_to include('legacy regex-ReAct')
+  end
 end

data/third_party/pwn_rdoc.jsonl

@@ -339,6 +339,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.degrade_text_only Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.degrade_text_only`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.dispatch_fail_n Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.dispatch_fail_n`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.emit_task_summary Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.emit_task_summary`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.engine_transient? Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.engine_transient?`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.escalate Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.escalate`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.evidence_enough_to_finalize? Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.evidence_enough_to_finalize?`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.expose_current_session Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.expose_current_session`: "}]}
@@ -793,6 +794,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::HttpRetry.max_attempts Usage"},{"role":"assistant","content":"`PWN::AI::HttpRetry.max_attempts`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::HttpRetry.report_event Usage"},{"role":"assistant","content":"`PWN::AI::HttpRetry.report_event`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::HttpRetry.retry_after_s Usage"},{"role":"assistant","content":"`PWN::AI::HttpRetry.retry_after_s`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::HttpRetry.retryable? Usage"},{"role":"assistant","content":"`PWN::AI::HttpRetry.retryable?`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::HttpRetry.timeout_s Usage"},{"role":"assistant","content":"`PWN::AI::HttpRetry.timeout_s`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Ollama.assemble_native_chat_stream Usage"},{"role":"assistant","content":"`PWN::AI::Ollama.assemble_native_chat_stream`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Ollama.assemble_ollama_stream Usage"},{"role":"assistant","content":"`PWN::AI::Ollama.assemble_ollama_stream`: "}]}

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pwn
 version: !ruby/object:Gem::Version
-  version: 0.5.697
+  version: 0.5.699
 platform: ruby
 authors:
 - 0day Inc.