pwn 0.5.721 → 0.5.722

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c26adc03f68a0e0275c0eb1a85aa2bb06def95aeb014411b000f52c6a83e3e40
-  data.tar.gz: 6a082b81f8f1885de1b3ac42c604f1f21a08d48e7260609ce35e00d1461afa79
+  metadata.gz: 31f54d1cd6a7bfbc4cef3bcc989650930b1b31bbf2bbc47654b20d6f20158775
+  data.tar.gz: 50bfda31eecd4b99d17324a8c4c548f14b8df10561c183f27bc9e7157c8f88bf
 SHA512:
-  metadata.gz: 719d713266b6a3d4200895bb41d4ac5ec7ccfc6d142463cb70085cd6c8d02a514191553b5b12317de45a38abf3d4d1ccb47101f54dc5e588e596a13bc3e77af8
-  data.tar.gz: 271d9ff938ac37275ebe9da7d3031305bd09f587e5b6d14f0386fc006bd120b6e684b00aa651c935da7eefbd3a71c298b114ef07ec21b7a9099a75d9f71df647
+  metadata.gz: 3a43118f36da1f9712dbfb908b5f0c9d4650d3bfde63239240d7c9cefcc50d29d78468e14e9cd9bf44f9fdba5df971b48659b348619dc79a7e17667028930025
+  data.tar.gz: 867d72ec34f85caaf18fa5abeb38b466d5663c1361becf17cb4af87115dd0d45411527b0f5616dd54874c241544416fb0b9427f16476248ed38a864b3951daef

data/etc/default_skills/pwn/ai/agent/learning/SKILL.md

@@ -34,6 +34,8 @@ PWN::AI::Agent::Learning.note_outcome(opts)
 ## Public methods
 
 - `note_outcome`
+- `consistency_check`
+- `disputed_save`
 - `outcomes`
 - `stats`
 - `to_context`

data/etc/default_skills/pwn/ai/agent/metrics/SKILL.md

@@ -38,6 +38,7 @@ PWN::AI::Agent::Metrics.load(opts)
 - `record`
 - `append_jsonl`
 - `summary`
+- `snapshot`
 - `to_context`
 - `proxy_trust`
 - `ucb`

data/etc/default_skills/pwn/ai/agent/mistakes/SKILL.md

@@ -37,6 +37,7 @@ PWN::AI::Agent::Mistakes.load(opts)
 - `save`
 - `signature`
 - `error_class`
+- `family`
 - `find`
 - `for_tool`
 - `record`
@@ -48,6 +49,7 @@ PWN::AI::Agent::Mistakes.load(opts)
 - `operator_inbox`
 - `to_context`
 - `correction_hint`
+- `note_hint_outcome`
 - `correction`
 - `check_user_correction`
 - `lean`

data/etc/default_skills/pwn/plugins/exploit_dev/SKILL.md

@@ -53,6 +53,7 @@ PWN::Plugins::ExploitDev.required_bins(opts)
 - `fmtstr`
 - `io`
 - `scaffold`
+- `from_crash`
 - `authors`
 - `help`
 

data/etc/default_skills/pwn/plugins/gdb/SKILL.md

@@ -42,6 +42,7 @@ PWN::Plugins::GDB.required_bins(opts)
 - `breakpoints`
 - `crash_info`
 - `debug_session`
+- `ptrace_preflight`
 - `authors`
 - `help`
 

data/etc/default_skills/pwn/plugins/jobs/SKILL.md

@@ -35,6 +35,7 @@ PWN::Plugins::Jobs.required_bins(opts)
 
 - `required_bins`
 - `start`
+- `watch`
 - `status`
 - `tail`
 - `result`

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

@@ -72,6 +72,15 @@ module PWN
               error: 'refused: memory_remember/skills_update text copied from last tool output'
             )
           end
+          if blob.match?(/open_sockraw|sockraw/) && defined?(PWN::Plugins::PreflightChecker) &&
+             PWN::Plugins::PreflightChecker.respond_to?(:cap_net_raw?) && !PWN::Plugins::PreflightChecker.cap_net_raw?
+            return JSON.generate(
+              success: false,
+              error: 'capability missing CAP_NET_RAW',
+              substitute: 'PWN::Plugins::Packet.tcp_connect_scan',
+              code: 'CAP_DENY'
+            )
+          end
           result = entry.handler.call(args)
           result = ToolGuard.quarantine_output(text: result) if defined?(ToolGuard) && result.is_a?(String) && ToolGuard.respond_to?(:quarantine_output)
           note_taint(text: result)

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

@@ -26,6 +26,7 @@ module PWN
       # restarts and is shared by every future session.
       module Learning
         LEARNING_FILE      = File.join(Dir.home, '.pwn', 'learning.jsonl')
+        DISPUTED_FILE      = File.join(Dir.home, '.pwn', 'learning', 'disputed.jsonl')
         LESSONS_FILE       = File.join(Dir.home, '.pwn', 'lessons.json')
         FINETUNE_DIR       = File.join(Dir.home, '.pwn', 'finetune')
         # P0 — post-answer introspect must not train "stop early" while
@@ -138,6 +139,12 @@ module PWN
             entry[:success] = true
             success = true
           end
+          check = consistency_check(details: details, success: success, rationale: opts[:rationale])
+          if check == :disputed
+            entry[:status] = 'disputed'
+            disputed_save(entry: entry)
+            return entry
+          end
           FileUtils.mkdir_p(File.dirname(LEARNING_FILE))
           File.open(LEARNING_FILE, 'a') { |f| f.puts(JSON.generate(entry)) }
           maybe_prune_outcomes!
@@ -159,6 +166,22 @@ module PWN
           entry
         end
 
+        public_class_method def self.consistency_check(opts = {})
+          details = "#{opts[:details]} #{opts[:rationale]}"
+          success = opts[:success]
+          return :disputed if details.match?(/\bPASS\b/) && success == false
+          return :disputed if details.match?(/\bFAIL\b/) && success == true && !details.match?(/\bPASS\b/)
+
+          :ok
+        end
+
+        public_class_method def self.disputed_save(opts = {})
+          entry = opts[:entry] || {}
+          FileUtils.mkdir_p(File.dirname(DISPUTED_FILE))
+          File.open(DISPUTED_FILE, 'a') { |f| f.puts(JSON.generate(entry)) }
+          entry
+        end
+
         # Supported Method Parameters::
         # rows = PWN::AI::Agent::Learning.outcomes(
         #   limit: 'optional - max entries returned newest-first (default 50)',
@@ -2081,6 +2104,18 @@ module PWN
               include_demoted: 'optional - include demoted lessons (defaults to false)'
             )
 
+            # Refuse PASS+success:false rows; they go to the disputed queue.
+            #{self}.consistency_check(
+              details: 'optional - details string that may contain PASS/FAIL',
+              rationale: 'optional - judge rationale text',
+              success: 'required - boolean success flag'
+            )
+
+            # Append a disputed outcome to ~/.pwn/learning/disputed.jsonl.
+            #{self}.disputed_save(
+              entry: 'required - Hash of the disputed learning row'
+            )
+
             # Print the AUTHOR(S) string for this module.
             #{self}.authors
           "

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

@@ -1176,6 +1176,16 @@ module PWN
           false
         end
 
+        private_class_method def self.session_tool_budget(opts = {})
+          name = opts[:name].to_s
+          result = opts[:result].to_s
+          return 16_384 if name.match?(/decompile|binary_triage|artifact_read|exploitdev/)
+          return 16_384 if name == 'pwn_eval' && result.match?(/r2 |objdump|xxd |pdf\b|afl-/)
+          return 16_384 if name == 'shell' && result.match?(/objdump|xxd|radare|r2 |disassembl/)
+
+          1_024
+        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.
@@ -3126,7 +3136,7 @@ module PWN
               append_session(
                 session_id: session_id,
                 role: 'tool',
-                content: "#{name} → #{result[0, 1_024]}"
+                content: "#{name} → #{result[0, session_tool_budget(name: name, result: result)]}"
               )
             end
 

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

@@ -156,6 +156,19 @@ module PWN
               .sort_by { |r| [-r[:calls], -r[:success_rate]] }.first(limit)
         end
 
+        public_class_method def self.snapshot(opts = {})
+          _day = opts[:day]
+          learn = defined?(Learning) ? Learning.stats : {}
+          {
+            success_rate: learn[:success_rate],
+            success_rate_orm: learn[:success_rate_orm],
+            success_rate_heur: learn[:success_rate_heur],
+            brier: (learn.dig(:calibration, :brier) if learn.is_a?(Hash)),
+            proxy_distrust: learn[:proxy_distrust],
+            tools: summary(limit: 8)
+          }
+        end
+
         # Supported Method Parameters::
         # ctx = PWN::AI::Agent::Metrics.to_context(
         #   limit: 'optional - cap number of tools included (default 8)',
@@ -814,6 +827,11 @@ module PWN
             # Run reset and return its result
             #{self}.reset
 
+            # Snapshot success rates per judge source for the LEARNING block.
+            #{self}.snapshot(
+              day: 'optional - reserved day key for rotation'
+            )
+
             # Print the AUTHOR(S) string for this module.
             #{self}.authors
           "

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

@@ -127,6 +127,31 @@ module PWN
           'other'
         end
 
+        public_class_method def self.family(opts = {})
+          klass = error_class(opts)
+          case klass
+          when 'docker_registry_auth', 'auth_denied' then 'auth_denied'
+          when 'name_conflict' then 'name_conflict'
+          when 'parse_error' then 'parse_error'
+          when 'missing_path', 'enoent' then 'missing_path'
+          when 'perm_denied', 'socket_perm', 'eacces' then 'permission_capability'
+          when 'net_unreach' then 'network_unreachable'
+          when 'timeout' then 'timeout'
+          else 'other'
+          end
+        end
+
+        FAMILY_FIXES = {
+          'auth_denied' => 'check registry credentials / image name; do NOT ls paths',
+          'name_conflict' => 'rename the resource or remove the existing name; do NOT ls paths',
+          'parse_error' => 'fix template/JSON syntax; do NOT ls paths',
+          'missing_path' => 'Path missing. ls/test -e the parent first, then run. Do not retry the same missing path.',
+          'permission_capability' => 'Raw sockets need CAP_NET_RAW. Do not retry open_sockraw. Use connect-scan or Sock.',
+          'network_unreachable' => 'network unreachable; check routing/VPN, do not ls paths',
+          'timeout' => 'deadline too short; retry same payload with timeout += 180',
+          'other' => 'inspect the exact error class before retrying'
+        }.freeze
+
         # Supported Method Parameters::
         # entry = PWN::AI::Agent::Mistakes.find(
         #   signature: 'optional - exact signature to fetch',
@@ -630,10 +655,38 @@ module PWN
 
           parts = ["seen #{m[:count]}× across #{Array(m[:sessions]).length} session(s), sig=#{m[:signature]}"]
           parts << 'REGRESSED (previous fix did not hold)' if m[:regressed]
-          parts << "KNOWN FIX: #{m[:fix]}" if m[:fix].to_s.strip.length.positive?
+          fam = family(error: opts[:error] || m[:error])
+          fam_fix = FAMILY_FIXES[fam]
+          stored = m[:fix].to_s.strip
+          stored = '' if stored.match?(%r{ls/test -e the parent}) && fam != 'missing_path'
+          stored = '' if m[:hint_confidence].to_f.negative?
+          stored = '' if stored.include?('[UNVERIFIED]')
+          hint = stored.empty? ? fam_fix : stored
+          parts << "KNOWN FIX: #{hint}" unless hint.to_s.empty?
           "[pwn-ai/mistakes] #{parts.join(' | ')}"
         end
 
+        public_class_method def self.note_hint_outcome(opts = {})
+          sig = (opts[:signature] || signature(tool: opts[:tool], error: opts[:error])).to_s
+          return nil if sig.empty?
+
+          store = load
+          m = store[sig.to_sym]
+          return nil unless m
+
+          m[:hint_outcomes] = m[:hint_outcomes].to_i + 1
+          unless opts[:helped]
+            m[:hint_misses] = m[:hint_misses].to_i + 1
+            if m[:hint_misses].to_i >= 3
+              m[:fix] = "[UNVERIFIED] #{m[:fix]}"
+              m[:hint_confidence] = -1.0
+            end
+          end
+          store[sig.to_sym] = m
+          save(store: store)
+          m
+        end
+
         # Supported Method Parameters::
         # bool = PWN::AI::Agent::Mistakes.correction?(request: user_text)
 
@@ -988,6 +1041,19 @@ module PWN
               signature: 'optional - signature value consumed by #effective_count'
             )
 
+            # Map an error string onto a coarse family (auth vs path vs parse).
+            #{self}.family(
+              error: 'required - raw error text'
+            )
+
+            # Record whether an injected KNOWN FIX helped; demote after 3 misses.
+            #{self}.note_hint_outcome(
+              tool: 'optional - tool name',
+              error: 'optional - raw error text',
+              signature: 'optional - explicit signature',
+              helped: 'optional - true when the hint resolved the failure'
+            )
+
             # Print the AUTHOR(S) string for this module.
             #{self}.authors
           "

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

@@ -143,17 +143,21 @@ module PWN
           #       confidence handles that in the bandit blend.
           eng = (PWN::Env.dig(:ai, :active) if defined?(PWN::Env)).to_s.downcase
           local = eng == 'ollama' || eng.empty?
-          if v[:source].to_s == 'heuristic'
-            v[:confidence] = local ? 0.35 : 0.5
+          if v[:source].to_s == 'heuristic' || v[:source].to_s.start_with?('heuristic')
+            toolbacked = Array(trace).any?
+            v[:heuristic_class] = toolbacked ? :toolbacked : :textual
+            v[:source] = :heuristic
+            v[:confidence] = if toolbacked
+                               local ? 0.55 : 0.7
+                             else
+                               local ? 0.35 : 0.5
+                             end
             raw = v[:score].to_f
             v[:score_raw] = raw
-            # preserve empty / failure-language / polite floors exactly (raw <= 0.15)
-            # and pass-through non-capped heuristics via the default arm.
-            if local && raw > 0.15 && trace.empty? && raw >= 0.6 && final.length < 400
+            if !toolbacked && local && raw > 0.15 && trace.empty? && raw >= 0.6 && final.length < 400
               v[:score] = [raw, 0.45].min
               v[:rationale] = "#{v[:rationale]} | P1:local_no_trace_cap"
-            elsif local && raw > 0.15 && trace.length < 2 && raw >= 0.85
-              # thin-evidence local highs: mild shrink toward 0.5
+            elsif !toolbacked && local && raw > 0.15 && trace.length < 2 && raw >= 0.85
               v[:score] = (0.5 + ((raw - 0.5) * 0.7)).round(3).clamp(0.0, 1.0)
             else
               v[:score] = raw

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

@@ -45,17 +45,22 @@ PWN::AI::Agent::Registry.register(
         path: { type: 'string' },
         offset: { type: 'integer' },
         length: { type: 'integer' },
-        mode: { type: 'string', description: 'text|hex|base64' }
+        mode: { type: 'string', description: 'text|hex|base64' },
+        grep: { type: 'string' },
+        ref: { type: 'string' },
+        sha256: { type: 'string' }
       },
       required: %w[path]
     }
   },
   handler: lambda { |args|
     PWN::Plugins::ArtifactRegistry.read_page(
-      path: args[:path] || args['path'],
+      path: args[:path] || args['path'] || args[:ref] || args['ref'],
       offset: args[:offset] || args['offset'],
       length: args[:length] || args['length'],
-      mode: args[:mode] || args['mode']
+      mode: args[:mode] || args['mode'],
+      grep: args[:grep] || args['grep'],
+      sha256: args[:sha256] || args['sha256']
     )
   }
 )

data/lib/pwn/plugins/artifact_registry.rb

@@ -51,13 +51,26 @@ module PWN
       end
 
       public_class_method def self.read_page(opts = {})
-        path = opts[:path].to_s
+        path = (opts[:path] || opts[:ref]).to_s
         raise 'ERROR: path is required' if path.empty?
         raise "ERROR: file not found: #{path}" unless File.file?(path)
 
+        raise 'ERROR: sha256 mismatch' if opts[:sha256].to_s != '' && Digest::SHA256.file(path).hexdigest != opts[:sha256].to_s
+
+        grep = opts[:grep].to_s
+        unless grep.empty?
+          rx = Regexp.new(grep, Regexp::IGNORECASE)
+          hits = []
+          File.foreach(path).with_index(1) do |ln, i|
+            hits << { line: i, text: ln.chomp } if ln.match?(rx)
+            break if hits.length >= 50
+          end
+          return { path: path, grep: grep, matches: hits }
+        end
+
         offset = opts[:offset].to_i
-        length = (opts[:length] || 4_096).to_i
-        length = 4_096 if length <= 0
+        length = (opts[:length] || 16_384).to_i
+        length = 16_384 if length <= 0
         mode = (opts[:mode] || 'text').to_s
         data = File.binread(path, length, offset).to_s
         case mode
@@ -117,14 +130,6 @@ module PWN
             path: 'required - filesystem path of a registered artifact'
           )
 
-          # Page an artifact as text, hex, or base64 from an offset.
-          #{self}.read_page(
-            path: 'required - filesystem path of the artifact',
-            offset: 'optional - byte offset (defaults to 0)',
-            length: 'optional - byte count (defaults to 4096)',
-            mode: 'optional - text, hex, or base64 (defaults to text)'
-          )
-
           # Store bytes under artifacts/sha256/<h2>/<hash> and append manifest.jsonl.
           #{self}.put(
             bytes: 'optional - raw bytes to store',
@@ -134,6 +139,17 @@ module PWN
             kind: 'optional - artifact kind'
           )
 
+          # Page an artifact; grep: searches lines, ref: alias for path.
+          #{self}.read_page(
+            path: 'required - filesystem path of the artifact',
+            ref: 'optional - alias for path',
+            offset: 'optional - byte offset (defaults to 0)',
+            length: 'optional - byte count (defaults to 16384)',
+            mode: 'optional - text, hex, or base64 (defaults to text)',
+            grep: 'optional - regex to search instead of paging',
+            sha256: 'optional - expected sha256 of the file'
+          )
+
           # Print the AUTHOR(S) string for this module.
           #{self}.authors
         "

data/lib/pwn/plugins/binary_parser.rb

@@ -120,7 +120,10 @@ module PWN
         end
         body = summary.merge(path: path, sha256: sha)
         File.write(art, JSON.pretty_generate(body))
-        body.merge(artifact: art, cached: false)
+        facts_dir = File.join(Dir.home, '.pwn', 'engagements', 'default', 'binaries', sha)
+        FileUtils.mkdir_p(facts_dir)
+        File.write(File.join(facts_dir, 'facts.json'), JSON.pretty_generate(body))
+        body.merge(artifact: art, cached: false, facts: File.join(facts_dir, 'facts.json'))
       end
 
       public_class_method def self.diff(opts = {})

data/lib/pwn/plugins/exploit_dev.rb

@@ -2,6 +2,7 @@
 
 require 'open3'
 require 'fileutils'
+require 'json'
 
 module PWN
   module Plugins
@@ -197,6 +198,14 @@ module PWN
         { path: out, triage: triage, gadgets: gadgets.first(8) }
       end
 
+      public_class_method def self.from_crash(opts = {})
+        path = (opts[:path] || opts[:from_crash]).to_s
+        raise 'ERROR: path is required' if path.empty?
+        raise "ERROR: file not found: #{path}" unless File.file?(path)
+
+        JSON.parse(File.read(path), symbolize_names: true)
+      end
+
       public_class_method def self.authors
         "AUTHOR(S):\n  0day Inc. <support@0dayinc.com>\n"
       end
@@ -344,6 +353,12 @@ module PWN
             names: 'optional - Array of symbol names (defaults to system, execve, __libc_start_main)'
           )
 
+          # Load a crash.json produced by a fuzz campaign.
+          #{self}.from_crash(
+            path: 'required - filesystem path of crash.json',
+            from_crash: 'optional - alias for path'
+          )
+
           # Print the AUTHOR(S) string for this module.
           #{self}.authors
         "

data/lib/pwn/plugins/gdb.rb

@@ -90,6 +90,18 @@ module PWN
         crash_info(opts.merge(commands: script))
       end
 
+      public_class_method def self.ptrace_preflight(opts = {})
+        _target = opts[:target]
+        scope = 0
+        path = '/proc/sys/kernel/yama/ptrace_scope'
+        scope = File.read(path).to_i if File.file?(path)
+        {
+          ptrace_scope: scope,
+          attach_ok: scope.zero?,
+          hint: scope.positive? ? 'spawn instead of attach' : 'ok'
+        }
+      end
+
       public_class_method def self.authors
         "AUTHOR(S):\n  0day Inc. <support@0dayinc.com>\n"
       end
@@ -147,6 +159,11 @@ module PWN
             script: 'optional - Array of gdb/MI commands'
           )
 
+          # Probe yama ptrace_scope before gdb/frida attach.
+          #{self}.ptrace_preflight(
+            target: 'optional - pid or binary path being attached'
+          )
+
           # Print the AUTHOR(S) string for this module.
           #{self}.authors
         "

data/lib/pwn/plugins/jobs.rb

@@ -40,6 +40,19 @@ module PWN
         row
       end
 
+      public_class_method def self.watch(opts = {})
+        row = load_job(opts)
+        pattern = opts[:pattern].to_s
+        raise 'ERROR: pattern is required' if pattern.empty?
+
+        log = row[:log].to_s
+        return { id: row[:id], hits: [] } unless File.file?(log)
+
+        rx = Regexp.new(pattern)
+        hits = File.readlines(log).grep(rx).last(20)
+        { id: row[:id], hits: hits.map(&:chomp), pattern: pattern }
+      end
+
       public_class_method def self.status(opts = {})
         row = load_job(opts)
         alive = begin
@@ -144,6 +157,12 @@ module PWN
           # Run stop and return its result
           #{self}.stop
 
+          # Tail a job log for a regex (AFL new-crash lines).
+          #{self}.watch(
+            id: 'required - job id from #start',
+            pattern: 'required - regex to match in the job log'
+          )
+
           # Print the AUTHOR(S) string for this module.
           #{self}.authors
         "

data/lib/pwn/plugins/repl.rb

@@ -494,7 +494,12 @@ module PWN
             puts '[*] TAB menus: leading `/` = commands (/cron /skills /sessions …); `/` later = host paths; otherwise Ruby completion (same as the pwn REPL).'
             puts "[*] tmux + terminator users: Ensure ~/.tmux.conf has 'set -s extended-keys on' and 'set -g xterm-keys on', then restart tmux. Use TERM=xterm-256color."
             tag = pi.config.pwn_ai_session_id.to_s.empty? ? '<SESSION_ID>' : pi.config.pwn_ai_session_id
-            puts "\n\n\npwn-ai debug ON → ~/.pwn/logs/pwn-ai-DEBUG-#{tag}-R<REQUEST_NUMBER>.log"
+
+            dbg_lvl = ''
+            dbg_lvl = 'trace' if pi.config.pwn_ai_trace
+            dbg_lvl = 'debug' if pi.config.pwn_ai_debug && !pi.config.pwn_ai_trace
+
+            puts "\n\n\npwn-ai #{dbg_lvl} ON → ~/.pwn/logs/pwn-ai-DEBUG-#{tag}-R<REQUEST_NUMBER>.log" unless dbg_lvl.empty?
           end
         end
 

data/lib/pwn/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module PWN
-  VERSION = '0.5.721'
+  VERSION = '0.5.722'
 end

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

@@ -216,4 +216,14 @@ describe PWN::AI::Agent::Mistakes do
     expect(classes).to eq(rows.map { |r| r['class'] })
     expect(classes.uniq.length).to eq(3)
   end
+
+  it 'gives pull-access-denied an auth family hint, never ls-the-parent' do
+    stub_const('PWN::AI::Agent::Mistakes::MISTAKES_FILE', File.join(Dir.mktmpdir, 'mistakes.json'))
+    described_class.reset
+    described_class.record(tool: 'shell', error: 'pull access denied for library/foo')
+    hint = described_class.correction_hint(tool: 'shell', error: 'pull access denied for library/foo')
+    expect(hint).to match(/credential|registry/i)
+    expect(hint).not_to match(%r{ls/test -e the parent})
+    expect(described_class.family(error: 'pull access denied')).to eq('auth_denied')
+  end
 end

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

@@ -438,4 +438,16 @@ describe 'PWN::AI::Agent::Reward vs TUI plan' do
     expect(v[:verifier_verdict]).to eq(:pass)
     expect(v[:judge_score]).to be < 0.6
   end
+
+  it 'tags tool-backed heuristics so they are not haircut' do
+    allow(PWN::AI::Agent::Reward).to receive(:llm_judge).and_return(nil)
+    v = PWN::AI::Agent::Reward.judge(
+      request: 'write /tmp/x and verify',
+      final: 'PASS file exists',
+      trace: ['{"success":true,"result":{"exit":0}}'],
+      commit: false
+    )
+    expect(v[:heuristic_class].to_s).to eq('toolbacked')
+    expect(v[:score]).to be >= 0.6
+  end
 end

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

@@ -22,4 +22,13 @@ describe PWN::Plugins::ArtifactRegistry do
       expect(rows.first[:kind]).to eq('loot')
     end
   end
+
+  it 'greps an artifact by regex' do
+    Dir.mktmpdir do |dir|
+      src = File.join(dir, 'dump.txt')
+      File.write(src, "nop\ncall system\nret\n")
+      hits = described_class.read_page(path: src, grep: 'call.*system')
+      expect(hits[:matches].first[:text]).to include('call system')
+    end
+  end
 end

data/third_party/pwn_rdoc.jsonl

@@ -280,9 +280,11 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.compact! Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.compact!`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.compress_exemplar Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.compress_exemplar`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.compress_finetune_trace Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.compress_finetune_trace`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.consistency_check Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.consistency_check`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.consolidate Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.consolidate`: Supported Method Parameters\n\nremoved = PWN::AI::Agent::Learning.consolidate(\n\nmax_entries: 'optional - hard cap on PWN::Memory size (default MAX_MEMORY_ENTRIES)'\n\n)\n\nDeduplicates near-identical lesson values and prunes the oldest entries once the cap is exceeded so the injected MEMORY block stays high-signal.\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.discount_success_rate Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.discount_success_rate`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.display_task Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.display_task`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.disputed_save Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.disputed_save`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.distill_skill Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.distill_skill`: Supported Method Parameters\n\nskill = PWN::AI::Agent::Learning.distill_skill(\n\nname: 'required - snake_case name for the new skill',\nsession_id: 'optional - PWN::Sessions id to mine (uses its transcript)',\ncontent: 'optional - explicit markdown body; overrides transcript mining',\nreferences: 'optional - Array of reference URLs / CWE / CVE / ATT&CK ids'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.exemplars_for Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.exemplars_for`: Supported Method Parameters\n\nmsgs = PWN::AI::Agent::Learning.exemplars_for(\n\nrequest: 'required - current user request',\nlimit: 'optional - max exemplar traces to return (default 1)',\nmax_msgs: 'optional - cap on messages per exemplar (default 6)'\n\n)\n\nRetrieval-augmented BEHAVIOUR: keyword-matches request against prior successful outcomes in learning.jsonl, loads the matching session, and compresses its (user, tool, assistant) trace into a short few-shot exemplar Loop.run splices between system and user. Local models are dramatically better with 1 concrete example than with 25 abstract lessons.\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Learning.export_finetune Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Learning.export_finetune`: "}]}
@@ -419,6 +421,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.run Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.run`: Supported Method Parameters\n\nfinal = PWN::AI::Agent::Loop.run(\n\nrequest: 'required - what the human typed',\nsession_id: 'optional - PWN::Sessions id (transcript is appended to it)',\nenabled_toolsets: 'optional - subset of Registry.toolsets, or nil for all',\non_tool: 'optional - ->(name, args, result) callback for live UI',\nsystem_role_content: 'optional - override default system prompt (built from session_id if not provided)'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.session_chat_history Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.session_chat_history`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.session_files Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.session_files`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.session_tool_budget Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.session_tool_budget`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.should_auto_introspect? Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.should_auto_introspect?`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.spill_tool_body Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.spill_tool_body`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Loop.spill_tool_history! Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Loop.spill_tool_history!`: "}]}
@@ -467,6 +470,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.save Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.save`: Supported Method Parameters\n\nPWN::AI::Agent::Metrics.save(\n\nmetrics: 'required - Hash returned by .load / mutated in place'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.scale_prediction Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.scale_prediction`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.scoreboard Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.scoreboard`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.snapshot Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.snapshot`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.summary Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.summary`: Supported Method Parameters\n\nrows = PWN::AI::Agent::Metrics.summary(\n\nlimit: 'optional - cap number of tools returned (default 25)',\nengine: 'optional - only that engine\\'s sub-bucket (falls back to global when absent)'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.thompson Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.thompson`: Supported Method Parameters\n\np = PWN::AI::Agent::Metrics.thompson(name: ‘shell’)\n\nC1 — Thompson sample from Beta(ok+1, fail+1). Naturally balances exploit/explore; used by Registry.rank as the tie-breaker.\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Metrics.to_context Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Metrics.to_context`: Supported Method Parameters\n\nctx = PWN::AI::Agent::Metrics.to_context(\n\nlimit: 'optional - cap number of tools included (default 8)',\nengine: 'optional - restrict to one engine\\'s telemetry'\n\n)\n"}]}
@@ -481,6 +485,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.error_class Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.error_class`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.extinguish! Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.extinguish!`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.extinguish_parked! Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.extinguish_parked!`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.family Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.family`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.find Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.find`: Supported Method Parameters\n\nentry = PWN::AI::Agent::Mistakes.find(\n\nsignature: 'optional - exact signature to fetch',\ntool: 'optional - with error:, compute signature and fetch',\nerror: 'optional - raw error text (used with tool:)'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.for_tool Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.for_tool`: Supported Method Parameters\n\nrows = PWN::AI::Agent::Mistakes.for_tool(\n\ntool: 'required - tool name',\nunresolved_only: 'optional - default false'\n\n)\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.help Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.help`: "}]}
@@ -488,6 +493,7 @@
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.lean! Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.lean!`: Supported Method Parameters\n\nresult = PWN::AI::Agent::Mistakes.lean!(\n\ndry_run: 'optional - Boolean (default false)',\nmax_resolved_kept: 'optional - cap on resolved-with-fix records',\nresolved_min_age_days: 'optional - age before resolved count=1 may drop'\n\n)\n\nCompact text fields on every record. Never drops unresolved, regressed, or high-count repeaters. Aged resolved-once fixes may drop after Memory already holds mistake_fix_<sig>.\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.load Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.load`: Supported Method Parameters\n\nstore = PWN::AI::Agent::Mistakes.load\n"}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.normalize_error Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.normalize_error`: "}]}
+{"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.note_hint_outcome Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.note_hint_outcome`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.operator_inbox Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.operator_inbox`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.park Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.park`: "}]}
 {"messages":[{"role":"user","content":"PWN::AI::Agent::Mistakes.previous_assistant Usage"},{"role":"assistant","content":"`PWN::AI::Agent::Mistakes.previous_assistant`: "}]}
@@ -2068,6 +2074,7 @@
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.flat Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.flat`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.fmt_writes Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.fmt_writes`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.fmtstr Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.fmtstr`: "}]}
+{"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.from_crash Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.from_crash`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.gadgets Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.gadgets`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.help Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.help`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::ExploitDev.io Usage"},{"role":"assistant","content":"`PWN::Plugins::ExploitDev.io`: "}]}
@@ -2121,6 +2128,7 @@
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.debug_session Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.debug_session`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.help Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.help`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.mitigations Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.mitigations`: "}]}
+{"messages":[{"role":"user","content":"PWN::Plugins::GDB.ptrace_preflight Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.ptrace_preflight`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.registers Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.registers`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.required_bins Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.required_bins`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::GDB.run_to_crash Usage"},{"role":"assistant","content":"`PWN::Plugins::GDB.run_to_crash`: "}]}
@@ -2293,6 +2301,7 @@
 {"messages":[{"role":"user","content":"PWN::Plugins::Jobs.status Usage"},{"role":"assistant","content":"`PWN::Plugins::Jobs.status`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::Jobs.stop Usage"},{"role":"assistant","content":"`PWN::Plugins::Jobs.stop`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::Jobs.tail Usage"},{"role":"assistant","content":"`PWN::Plugins::Jobs.tail`: "}]}
+{"messages":[{"role":"user","content":"PWN::Plugins::Jobs.watch Usage"},{"role":"assistant","content":"`PWN::Plugins::Jobs.watch`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::K8s.authors Usage"},{"role":"assistant","content":"`PWN::Plugins::K8s.authors`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::K8s.docker_socket? Usage"},{"role":"assistant","content":"`PWN::Plugins::K8s.docker_socket?`: "}]}
 {"messages":[{"role":"user","content":"PWN::Plugins::K8s.help Usage"},{"role":"assistant","content":"`PWN::Plugins::K8s.help`: "}]}

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pwn
 version: !ruby/object:Gem::Version
-  version: 0.5.721
+  version: 0.5.722
 platform: ruby
 authors:
 - 0day Inc.