pentesting 0.55.0 → 0.55.2

package/dist/main.js

@@ -342,7 +342,7 @@ var ORPHAN_PROCESS_NAMES = [
 
 // src/shared/constants/agent.ts
 var APP_NAME = "Pentest AI";
-var APP_VERSION = "0.55.0";
+var APP_VERSION = "0.55.2";
 var APP_DESCRIPTION = "Autonomous Penetration Testing AI Agent";
 var LLM_ROLES = {
   SYSTEM: "system",
@@ -4940,19 +4940,145 @@ var CHALLENGE_TYPE_SIGNALS = {
   ]
 };
 
+// src/shared/constants/prompts.ts
+var PROMPT_PATHS = {
+  BASE: "base.md",
+  OFFENSIVE_PLAYBOOK: "offensive-playbook.md",
+  AGENT_FILES: {
+    ORCHESTRATOR: "orchestrator.md",
+    RECON: "recon.md",
+    VULN: "vuln.md",
+    WEB: "web.md",
+    EXPLOIT: "exploit.md",
+    POST: "post.md",
+    REPORT: "report.md",
+    INFRA: "infra.md",
+    // Strategic methodology (always loaded as reference)
+    STRATEGY: "strategy.md",
+    EVASION: "evasion.md",
+    ZERO_DAY: "zero-day.md",
+    PAYLOAD_CRAFT: "payload-craft.md"
+  },
+  /** Technique reference library directory */
+  TECHNIQUES_DIR: "techniques"
+};
+var TECHNIQUE_FILES = {
+  INJECTION: "injection",
+  PRIVESC: "privesc",
+  LATERAL: "lateral",
+  AD_ATTACK: "ad-attack",
+  AUTH_ACCESS: "auth-access",
+  CONTAINER_ESCAPE: "container-escape",
+  SANDBOX_ESCAPE: "sandbox-escape",
+  FILE_ATTACKS: "file-attacks",
+  NETWORK_SVC: "network-svc",
+  CRYPTO: "crypto",
+  REVERSING: "reversing",
+  FORENSICS: "forensics",
+  PWN: "pwn",
+  SHELLS: "shells"
+};
+var PROMPT_XML = {
+  PHASE: (phase, content) => `<phase-instructions phase="${phase}">
+${content}
+</phase-instructions>`,
+  SCOPE: (allowed, domains, exclude, flags) => `<scope type="ABSOLUTE_CONSTRAINT">
+Authorized CIDR: ${allowed}
+Authorized Domains: ${domains}
+Exclusions: ${exclude}
+Constraints: ${flags}
+</scope>`,
+  STATE: (content) => `<current-state>
+${content}
+</current-state>`,
+  TODO: (content) => `<todo>
+${content}
+</todo>`
+};
+var PROMPT_DEFAULTS = {
+  NO_SCOPE: "<scope>NO SCOPE DEFINED. STOP.</scope>",
+  EMPTY_TODO: "Create initial plan",
+  USER_CONTEXT: (context) => `
+<user-input>
+"${context}"
+</user-input>
+
+<intent-rules>
+ANALYZE the user's intent before acting. Classify into ONE:
+  ABORT \u2192 stop current work, confirm with \`ask_user\`
+  CORRECTION \u2192 adjust approach, continue
+  INFORMATION \u2192 store and USE immediately (credentials, paths, hints)
+  COMMAND \u2192 execute EXACTLY what was asked, nothing more
+  TARGET_CHANGE \u2192 \`add_target\`, then begin testing
+  GUIDANCE \u2192 acknowledge via \`ask_user\`, adjust strategy, continue
+  STATUS_QUERY \u2192 report via \`ask_user\`, then RESUME previous work
+  CONVERSATION \u2192 respond via \`ask_user\`, do NOT scan or attack
+
+RULES:
+- No target set and none provided \u2192 \`ask_user\` to request target.
+- Conversation or greeting \u2192 respond conversationally, do NOT attack.
+- Uncertain intent \u2192 ask for clarification with \`ask_user\`.
+- This is a collaborative tool. The user is your partner.
+</intent-rules>`
+};
+var PROMPT_CONFIG = {
+  ENCODING: "utf-8"
+};
+var INITIAL_TASKS = {
+  RECON: "Initial reconnaissance and target discovery"
+};
+var CONTEXT_EXTRACTOR_PROMPT = `You are extracting actionable intelligence from a penetration testing session.
+DO NOT simply summarize or shorten. EXTRACT critical facts:
+
+1. DISCOVERED: Services, versions, paths, parameters (exact IPs, ports, versions)
+2. CONFIRMED: Vulnerabilities or access confirmed
+3. CREDENTIALS: Usernames, passwords, tokens, keys
+4. DEAD ENDS: What failed \u2014 include EXACT command, tool, arguments, wordlist/file used.
+   Distinguish between:
+   - "This approach itself is impossible" (e.g., SSH key-only \u2192 no password brute force works)
+   - "This specific attempt failed" (e.g., sqlmap with default tamper \u2192 try different tamper)
+5. OPEN LEADS: Unexplored paths worth pursuing
+
+Every line must include exact commands/tools/files used.
+The reader must be able to judge whether a retry with different parameters is worthwhile.`;
+var REFLECTION_PROMPT = `You are a tactical reviewer for a penetration testing agent.
+Review ALL actions from this turn \u2014 successes AND failures.
+
+1. ASSESSMENT: What did this turn accomplish? Rate: HIGH / MED / LOW / NONE.
+2. SUCCESSES: What worked? Can this pattern be replicated elsewhere?
+3. FAILURES: What failed? Is this a repeated pattern? If so \u2192 STOP this approach.
+4. BLIND SPOTS: What was missed or overlooked?
+5. NEXT PRIORITY: Single most valuable next action.
+
+3-5 lines. Every word must be actionable.`;
+var SUMMARY_REGENERATOR_PROMPT = `Update this penetration testing session summary with the new turn data.
+
+Must include:
+- All discovered hosts, services, versions (exact IPs, ports, software versions)
+- All confirmed vulnerabilities
+- All obtained credentials
+- Failed attempts with EXACT commands/tools/arguments/files used.
+  For each failure, state:
+  - The root cause (auth method? WAF? patched? wrong params?)
+  - Whether retrying with different parameters could work
+- Top unexplored leads
+
+Remove outdated/superseded info. Keep concise but COMPLETE.
+The reader must be able to decide what to retry and what to never attempt again.`;
+
 // src/shared/utils/auto-prompter.ts
 var SECONDARY_TYPE_RATIO = 0.5;
 var MIN_CHALLENGE_CONFIDENCE = 0.2;
 var MIN_RECON_OUTPUT_LENGTH = 100;
 var TYPE_TECHNIQUE_MAP = {
-  web: ["injection", "auth-access", "file-attacks", "crypto"],
-  pwn: ["pwn", "shells", "reversing"],
-  crypto: ["crypto"],
-  forensics: ["forensics", "reversing", "crypto"],
-  reversing: ["reversing", "pwn"],
-  misc: ["sandbox-escape", "crypto", "forensics"],
-  network: ["network-svc", "shells", "lateral", "ad-attack"],
-  unknown: ["network-svc", "injection", "shells", "file-attacks"]
+  web: [TECHNIQUE_FILES.INJECTION, TECHNIQUE_FILES.AUTH_ACCESS, TECHNIQUE_FILES.FILE_ATTACKS, TECHNIQUE_FILES.CRYPTO],
+  pwn: [TECHNIQUE_FILES.PWN, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.REVERSING],
+  crypto: [TECHNIQUE_FILES.CRYPTO],
+  forensics: [TECHNIQUE_FILES.FORENSICS, TECHNIQUE_FILES.REVERSING, TECHNIQUE_FILES.CRYPTO],
+  reversing: [TECHNIQUE_FILES.REVERSING, TECHNIQUE_FILES.PWN],
+  misc: [TECHNIQUE_FILES.SANDBOX_ESCAPE, TECHNIQUE_FILES.CRYPTO, TECHNIQUE_FILES.FORENSICS],
+  network: [TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.LATERAL, TECHNIQUE_FILES.AD_ATTACK],
+  unknown: [TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.INJECTION, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.FILE_ATTACKS]
 };
 var TYPE_PHASE_PROMPT_MAP = {
   web: "web.md",
@@ -9635,7 +9761,9 @@ var LLM_LIMITS = {
    *  can produce full analysis, tool calls, and reasoning without cutoff. */
   streamMaxTokens: 128e3,
   /** WHY: ~3.5 chars/token is a reasonable average for mixed English/CJK content */
-  charsPerTokenEstimate: 3.5
+  charsPerTokenEstimate: 3.5,
+  /** WHY: 5 minutes max timeout for streaming and non-streaming responses */
+  fetchTimeoutMs: 3e5
 };
 var LLM_ERROR_TYPES = {
   RATE_LIMIT: "rate_limit",
@@ -9774,6 +9902,8 @@ var LLMClient = class {
   }
   async makeRequest(body, signal) {
     const url = `${this.baseUrl}${LLM_API.MESSAGES_PATH}`;
+    const timeoutSignal = AbortSignal.timeout(LLM_LIMITS.fetchTimeoutMs);
+    const combinedSignal = signal ? AbortSignal.any([signal, timeoutSignal]) : timeoutSignal;
     const response = await fetch(url, {
       method: LLM_HTTP_METHOD.POST,
       headers: {
@@ -9784,7 +9914,7 @@ var LLMClient = class {
         [LLM_HEADER.ANTHROPIC_VERSION]: LLM_API.VERSION
       },
       body: JSON.stringify(body),
-      signal
+      signal: combinedSignal
     });
     if (!response.ok) {
       let errorBody = `HTTP ${response.status}`;
@@ -10475,19 +10605,19 @@ var ANALYST_SYSTEM_PROMPT = `You are an independent pentesting output analyst. Y
 
 FORMAT YOUR RESPONSE EXACTLY LIKE THIS:
 
-## Key Findings
+## ${MEMO_SECTIONS.KEY_FINDINGS}
 - [finding 1 with exact values: ports, versions, paths]
 - [finding 2]
 
-## Credentials/Secrets
+## ${MEMO_SECTIONS.CREDENTIALS}
 - [any discovered credentials, hashes, tokens, keys, certificates]
 - (write "None found" if none)
 
-## Attack Vectors
+## ${MEMO_SECTIONS.ATTACK_VECTORS}
 - [exploitable services, vulnerabilities, misconfigurations, CVEs]
 - (write "None identified" if none)
 
-## Failures/Errors
+## ${MEMO_SECTIONS.FAILURES}
 Classify EVERY failure using one of these types. Format: [TYPE] exact_command \u2192 why_failed \u2192 recommended_pivot
 
 Failure types:
@@ -10504,7 +10634,7 @@ Examples:
 - "[TIMEOUT] nmap -sV -p- target --min-rate=5000 \u2192 timed out 5min \u2192 rustscan first, then targeted nmap on found ports"
 - (write "No failures" if everything succeeded)
 
-## Suspicious Signals
+## ${MEMO_SECTIONS.SUSPICIONS}
 - [anomalies that are NOT confirmed vulnerabilities but suggest exploitable surface]
 - [e.g.: "Response time 3x slower on /admin path \u2014 possible auth check or backend processing"]
 - [e.g.: "X-Debug-Token header present \u2014 debug mode may be enabled"]
@@ -10512,11 +10642,11 @@ Examples:
 - [e.g.: "Unexpected 302 redirect with session param leaked in URL"]
 - (write "No suspicious signals" if nothing anomalous)
 
-## Attack Value
+## ${MEMO_SECTIONS.ATTACK_VALUE}
 - [ONE word: HIGH / MED / LOW / NONE]
 - Reasoning: [1 sentence why \u2014 what makes this worth pursuing or abandoning]
 
-## Next Steps
+## ${MEMO_SECTIONS.NEXT_STEPS}
 - [recommended immediate actions based on findings]
 
 RULES:
@@ -10530,7 +10660,7 @@ RULES:
 - Do NOT miss subtle signals: unusual HTTP headers, non-standard responses, timing differences
 - Write as much detail as needed \u2014 do NOT artificially shorten. Every detail matters for strategy.
 
-## Reflection
+## ${MEMO_SECTIONS.REFLECTION}
 - What this output tells us: [1-line assessment]
 - Recommended next action: [1-2 specific follow-up actions]`;
 function parseAnalystMemo(response) {
@@ -11328,116 +11458,6 @@ import { readFileSync as readFileSync6, existsSync as existsSync10 } from "fs";
 import { join as join11, dirname as dirname4 } from "path";
 import { fileURLToPath as fileURLToPath2 } from "url";
 
-// src/shared/constants/prompts.ts
-var PROMPT_PATHS = {
-  BASE: "base.md",
-  OFFENSIVE_PLAYBOOK: "offensive-playbook.md",
-  AGENT_FILES: {
-    ORCHESTRATOR: "orchestrator.md",
-    RECON: "recon.md",
-    VULN: "vuln.md",
-    WEB: "web.md",
-    EXPLOIT: "exploit.md",
-    POST: "post.md",
-    REPORT: "report.md",
-    INFRA: "infra.md",
-    // Strategic methodology (always loaded as reference)
-    STRATEGY: "strategy.md",
-    EVASION: "evasion.md",
-    ZERO_DAY: "zero-day.md",
-    PAYLOAD_CRAFT: "payload-craft.md"
-  },
-  /** Technique reference library directory (auto-discovered) */
-  TECHNIQUES_DIR: "techniques"
-};
-var PROMPT_XML = {
-  PHASE: (phase, content) => `<phase-instructions phase="${phase}">
-${content}
-</phase-instructions>`,
-  SCOPE: (allowed, domains, exclude, flags) => `<scope type="ABSOLUTE_CONSTRAINT">
-Authorized CIDR: ${allowed}
-Authorized Domains: ${domains}
-Exclusions: ${exclude}
-Constraints: ${flags}
-</scope>`,
-  STATE: (content) => `<current-state>
-${content}
-</current-state>`,
-  TODO: (content) => `<todo>
-${content}
-</todo>`
-};
-var PROMPT_DEFAULTS = {
-  NO_SCOPE: "<scope>NO SCOPE DEFINED. STOP.</scope>",
-  EMPTY_TODO: "Create initial plan",
-  USER_CONTEXT: (context) => `
-<user-input>
-"${context}"
-</user-input>
-
-<intent-rules>
-ANALYZE the user's intent before acting. Classify into ONE:
-  ABORT \u2192 stop current work, confirm with \`ask_user\`
-  CORRECTION \u2192 adjust approach, continue
-  INFORMATION \u2192 store and USE immediately (credentials, paths, hints)
-  COMMAND \u2192 execute EXACTLY what was asked, nothing more
-  TARGET_CHANGE \u2192 \`add_target\`, then begin testing
-  GUIDANCE \u2192 acknowledge via \`ask_user\`, adjust strategy, continue
-  STATUS_QUERY \u2192 report via \`ask_user\`, then RESUME previous work
-  CONVERSATION \u2192 respond via \`ask_user\`, do NOT scan or attack
-
-RULES:
-- No target set and none provided \u2192 \`ask_user\` to request target.
-- Conversation or greeting \u2192 respond conversationally, do NOT attack.
-- Uncertain intent \u2192 ask for clarification with \`ask_user\`.
-- This is a collaborative tool. The user is your partner.
-</intent-rules>`
-};
-var PROMPT_CONFIG = {
-  ENCODING: "utf-8"
-};
-var INITIAL_TASKS = {
-  RECON: "Initial reconnaissance and target discovery"
-};
-var CONTEXT_EXTRACTOR_PROMPT = `You are extracting actionable intelligence from a penetration testing session.
-DO NOT simply summarize or shorten. EXTRACT critical facts:
-
-1. DISCOVERED: Services, versions, paths, parameters (exact IPs, ports, versions)
-2. CONFIRMED: Vulnerabilities or access confirmed
-3. CREDENTIALS: Usernames, passwords, tokens, keys
-4. DEAD ENDS: What failed \u2014 include EXACT command, tool, arguments, wordlist/file used.
-   Distinguish between:
-   - "This approach itself is impossible" (e.g., SSH key-only \u2192 no password brute force works)
-   - "This specific attempt failed" (e.g., sqlmap with default tamper \u2192 try different tamper)
-5. OPEN LEADS: Unexplored paths worth pursuing
-
-Every line must include exact commands/tools/files used.
-The reader must be able to judge whether a retry with different parameters is worthwhile.`;
-var REFLECTION_PROMPT = `You are a tactical reviewer for a penetration testing agent.
-Review ALL actions from this turn \u2014 successes AND failures.
-
-1. ASSESSMENT: What did this turn accomplish? Rate: HIGH / MED / LOW / NONE.
-2. SUCCESSES: What worked? Can this pattern be replicated elsewhere?
-3. FAILURES: What failed? Is this a repeated pattern? If so \u2192 STOP this approach.
-4. BLIND SPOTS: What was missed or overlooked?
-5. NEXT PRIORITY: Single most valuable next action.
-
-3-5 lines. Every word must be actionable.`;
-var SUMMARY_REGENERATOR_PROMPT = `Update this penetration testing session summary with the new turn data.
-
-Must include:
-- All discovered hosts, services, versions (exact IPs, ports, software versions)
-- All confirmed vulnerabilities
-- All obtained credentials
-- Failed attempts with EXACT commands/tools/arguments/files used.
-  For each failure, state:
-  - The root cause (auth method? WAF? patched? wrong params?)
-  - Whether retrying with different parameters could work
-- Top unexplored leads
-
-Remove outdated/superseded info. Keep concise but COMPLETE.
-The reader must be able to decide what to retry and what to never attempt again.`;
-
 // src/shared/constants/scoring.ts
 var ATTACK_SCORING = {
   /** Base score for all attack prioritization */
@@ -11822,15 +11842,15 @@ var CORE_KNOWLEDGE_FILES = [
   // Active Directory / infrastructure attack methodology
 ];
 var PHASE_TECHNIQUE_MAP = {
-  [PHASES.RECON]: ["network-svc", "shells", "crypto"],
-  [PHASES.VULN_ANALYSIS]: ["injection", "network-svc", "file-attacks", "crypto", "reversing"],
-  [PHASES.EXPLOIT]: ["injection", "shells", "file-attacks", "network-svc", "pwn", "container-escape", "reversing"],
-  [PHASES.POST_EXPLOIT]: ["privesc", "lateral", "auth-access", "shells", "container-escape", "forensics"],
-  [PHASES.PRIV_ESC]: ["privesc", "auth-access", "shells", "pwn", "container-escape"],
-  [PHASES.LATERAL]: ["lateral", "ad-attack", "auth-access", "container-escape", "network-svc"],
-  [PHASES.PERSISTENCE]: ["shells", "privesc", "lateral"],
-  [PHASES.EXFIL]: ["lateral", "network-svc", "forensics"],
-  [PHASES.WEB]: ["injection", "file-attacks", "auth-access", "crypto", "shells"],
+  [PHASES.RECON]: [TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.CRYPTO],
+  [PHASES.VULN_ANALYSIS]: [TECHNIQUE_FILES.INJECTION, TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.FILE_ATTACKS, TECHNIQUE_FILES.CRYPTO, TECHNIQUE_FILES.REVERSING],
+  [PHASES.EXPLOIT]: [TECHNIQUE_FILES.INJECTION, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.FILE_ATTACKS, TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.PWN, TECHNIQUE_FILES.CONTAINER_ESCAPE, TECHNIQUE_FILES.SANDBOX_ESCAPE, TECHNIQUE_FILES.REVERSING],
+  [PHASES.POST_EXPLOIT]: [TECHNIQUE_FILES.PRIVESC, TECHNIQUE_FILES.LATERAL, TECHNIQUE_FILES.AUTH_ACCESS, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.CONTAINER_ESCAPE, TECHNIQUE_FILES.SANDBOX_ESCAPE, TECHNIQUE_FILES.FORENSICS],
+  [PHASES.PRIV_ESC]: [TECHNIQUE_FILES.PRIVESC, TECHNIQUE_FILES.AUTH_ACCESS, TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.PWN, TECHNIQUE_FILES.CONTAINER_ESCAPE, TECHNIQUE_FILES.SANDBOX_ESCAPE],
+  [PHASES.LATERAL]: [TECHNIQUE_FILES.LATERAL, TECHNIQUE_FILES.AD_ATTACK, TECHNIQUE_FILES.AUTH_ACCESS, TECHNIQUE_FILES.CONTAINER_ESCAPE, TECHNIQUE_FILES.NETWORK_SVC],
+  [PHASES.PERSISTENCE]: [TECHNIQUE_FILES.SHELLS, TECHNIQUE_FILES.PRIVESC, TECHNIQUE_FILES.LATERAL],
+  [PHASES.EXFIL]: [TECHNIQUE_FILES.LATERAL, TECHNIQUE_FILES.NETWORK_SVC, TECHNIQUE_FILES.FORENSICS],
+  [PHASES.WEB]: [TECHNIQUE_FILES.INJECTION, TECHNIQUE_FILES.FILE_ATTACKS, TECHNIQUE_FILES.AUTH_ACCESS, TECHNIQUE_FILES.CRYPTO, TECHNIQUE_FILES.SHELLS],
   [PHASES.REPORT]: []
   // Report phase: no attack techniques needed
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pentesting",
-  "version": "0.55.0",
+  "version": "0.55.2",
   "description": "Autonomous Penetration Testing AI Agent",
   "type": "module",
   "main": "dist/main.js",