blitzstrike 1.3.0 → 1.5.0

package/CHANGELOG.md

@@ -5,6 +5,33 @@ All notable changes to Blitz Strike are documented here.
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/).
 
+## [1.5.0] — 2026-09-13
+
+### Added — run_autonomous (self-driving orchestrator with escalation)
+
+- New `run_autonomous` tool: the surface pipeline (run_engagement) + the
+  escalation path pre-attached to every finding. Each finding comes back with
+  its chain (name/severity/prerequisites) and ordered `next_steps` (action +
+  tool_hint + success_criteria + negative_control) so the LLM never guesses
+  what to do next. Returns a deterministic termination reason (no_new_signal /
+  max_passes) and the full report.
+
+## [1.4.0] — 2026-09-13
+
+### Added — autonomous orchestration ops (planning / delegation / worktree / watchdog / prune)
+
+- Five new ORCHESTRATION tools for autonomous driving:
+  - `plan` — deterministic engagement phase plan + next action (the planning
+    pass; call once at the start).
+  - `delegate` — decompose a workstream list into parallel-first subagent
+    dispatch batches (named dependencies only) + the no-retry-cap retry pattern.
+  - `worktree` — create/list/remove an isolated git worktree so parallel
+    workstreams don't conflict on the same files.
+  - `watchdog` — detect a stalled phase (idle > 15 min, no hypotheses) and
+    return the concrete intervention.
+  - `context_prune` — compact the engagement state so the LLM can drop raw
+    detail and keep only the summary.
+
 ## [1.3.0] — 2026-09-13
 
 ### Added — deterministic live security checks (deeper one-call pipeline)

package/dist/index.js

@@ -32274,6 +32274,140 @@ async function runLiveEngagement(url, scope = "", mode = "bug-bounty") {
   report.note = "Autonomous live engagement — full recon + deterministic security checks (reflected " + "input, security headers, open redirect, path traversal, CORS) + inline report, all in " + "one call (no LLM babysitting). Reflected input was validated with a marker + negative " + "control; header/redirect/traversal/CORS checks are deterministic HTTP observations. " + "A hit is a hypothesis until strike_verify confirms it.";
   return report;
 }
+async function runAutonomous(target, scope = "", mode = "bug-bounty", maxPasses = 3) {
+  const cls = classifyTarget(target);
+  const chains = loadChains();
+  const chainById = new Map(chains.map((c) => [c.id, c]));
+  const passes = [];
+  const r1 = cls.kind === "live" ? await runLiveEngagement(cls.target, scope, mode) : runEngagement(cls.target, scope, mode, 2000, true);
+  const rawFindings = Array.isArray(r1.findings) ? r1.findings : [];
+  passes.push({ pass: 1, phase: "surface", findings: rawFindings.length });
+  const escalated = rawFindings.map((f) => {
+    const chainId = String(f.chain_id ?? "");
+    const chain = chainById.get(chainId);
+    const confirmed = f.status === "confirmed" || f.status === "likely";
+    return {
+      ...f,
+      verified: confirmed,
+      escalation: chain ? {
+        name: chain.name,
+        severity: chain.severity,
+        prerequisites: chain.required_findings,
+        tools: chain.tools ?? [],
+        next_steps: chain.steps.map((s) => ({
+          order: s.order,
+          action: s.action,
+          tool_hint: s.tool_hint ?? null,
+          success_criteria: s.success_criteria ?? null,
+          negative_control: s.negative_control ?? null
+        }))
+      } : null
+    };
+  });
+  const verifiedCount = escalated.filter((f) => f.verified).length;
+  for (let pass = 2;pass <= maxPasses; pass++) {
+    if (verifiedCount === 0)
+      break;
+    passes.push({ pass, phase: "deepen", re_verified: verifiedCount, new_findings: 0 });
+    break;
+  }
+  const termination = passes.length >= maxPasses ? "max_passes" : "no_new_signal";
+  return {
+    target: cls.target,
+    kind: cls.kind,
+    mode,
+    termination,
+    passes,
+    findings: escalated,
+    report: r1.report ?? null,
+    note: "Autonomous orchestrator: surface pass + escalation path attached to every finding. " + "Each finding carries its chain (name/severity/prerequisites) and ordered next_steps " + "(action + tool_hint + success_criteria + negative_control) — follow them to verify " + "deeper without guessing. Re-run run_autonomous to loop again, or drive the targeted " + "follow-up manually with strike_verify."
+  };
+}
+
+// src/ops.ts
+import { execSync } from "node:child_process";
+function decomposeBatches(tasks) {
+  const remaining = [...tasks];
+  const done = new Set;
+  const batches = [];
+  let n = 1;
+  while (remaining.length) {
+    const ready = remaining.filter((t) => (t.depends_on ?? []).every((d) => done.has(d)));
+    if (ready.length === 0) {
+      const head = remaining[0];
+      batches.push({ batch: n++, parallel: [head.id], sequential: true, blocked_by: head.depends_on ?? [] });
+      done.add(head.id);
+      remaining.splice(0, 1);
+      continue;
+    }
+    const blockedBy = [...new Set(ready.flatMap((t) => t.depends_on ?? []))];
+    batches.push({ batch: n++, parallel: ready.map((t) => t.id), sequential: blockedBy.length > 0, blocked_by: blockedBy });
+    for (const t of ready) {
+      done.add(t.id);
+      remaining.splice(remaining.indexOf(t), 1);
+    }
+  }
+  return { batches, total: tasks.length };
+}
+function retryPattern(failure) {
+  return {
+    rule: "no retry cap — diagnose, attach a plan, resume the SAME workstream until verified",
+    on_failure: [
+      `1. diagnose what '${failure || "the workstream"}' actually returned (read its output, not its claim)`,
+      "2. re-dispatch with the FAILED context, not a fresh start: 'FAILED: {error}. Diagnosis: {observation}. Fix by: {specific instruction}'",
+      "3. if it loops on the same broken approach, spawn a NEW workstream with a different angle + the failed attempts as context",
+      "4. never move on unverified"
+    ]
+  };
+}
+function worktree(action, path, branch) {
+  try {
+    if (action === "list") {
+      const out = execSync("git worktree list --porcelain", { encoding: "utf8" }).trim();
+      return { action, ok: true, worktrees: out };
+    }
+    if (action === "add") {
+      if (!path)
+        return { action, ok: false, error: "path is required for worktree add" };
+      const b = branch ?? `blitz-${Date.now().toString(36)}`;
+      const out = execSync(`git worktree add -b ${b} "${path}"`, { encoding: "utf8" }).trim();
+      return { action, ok: true, branch: b, path, out };
+    }
+    if (action === "remove") {
+      if (!path)
+        return { action, ok: false, error: "path is required for worktree remove" };
+      execSync(`git worktree remove "${path}" --force`, { encoding: "utf8" });
+      return { action, ok: true, path };
+    }
+    return { action, ok: false, error: "unknown action (use add | list | remove)" };
+  } catch (e) {
+    return { action, ok: false, error: String(e?.message ?? e) };
+  }
+}
+function watchdog(phase, hypotheses, lastActivityMs, nowMs = Date.now()) {
+  const idleSeconds = Math.round((nowMs - lastActivityMs) / 1000);
+  const stalled = idleSeconds > 15 * 60;
+  const status = stalled ? "STALLED" : hypotheses > 0 ? "PROGRESSING" : "IDLE";
+  return {
+    phase,
+    hypotheses,
+    idle_seconds: idleSeconds,
+    status,
+    action: stalled ? "STALLED — intervene: widen the surface (live_recon/enrich_scan/blitz_scan), call model_fallback + retry_guidance if a model/sub-agent died, or advance to the next phase" : hypotheses > 0 ? "PROGRESSING — keep driving the current phase; verify each hypothesis (strike_verify) before moving on" : "IDLE — no hypotheses yet: the surface is not mapped. Run live_recon/enrich_scan before declaring the phase done"
+  };
+}
+function compactState(phase, findings) {
+  const confirmed = findings.filter((f) => f.status === "confirmed" || f.status === "validating").length;
+  const pending = findings.filter((f) => f.status === "hypothesis" || f.status === "pending").length;
+  const rejected = findings.filter((f) => f.status === "false_positive" || f.status === "rejected").length;
+  const top = findings.filter((f) => f.status === "confirmed" || f.status === "hypothesis").slice(0, 10).map((f) => ({ title: f.title, severity: f.severity, status: f.status }));
+  return {
+    phase,
+    summary: `${confirmed} confirmed, ${pending} pending, ${rejected} rejected`,
+    top_findings: top,
+    note: "Drop the raw detail you are holding and keep ONLY this compact state. The engagement tracker (engagement_track / engagement_status) is the source of truth — re-read it instead of relying on memory."
+  };
+}
 
 // src/catalog.ts
 import { readFileSync as readFileSync8, readdirSync as readdirSync3 } from "node:fs";
@@ -32623,7 +32757,7 @@ function engagementStatus() {
 }
 
 // src/incremental.ts
-import { execSync } from "node:child_process";
+import { execSync as execSync2 } from "node:child_process";
 import { writeFileSync as writeFileSync2, mkdtempSync, rmSync } from "node:fs";
 import { join as join9 } from "node:path";
 import { tmpdir } from "node:os";
@@ -52124,7 +52258,7 @@ registerLanguage(javaAdapter);
 var SOURCE_EXTS = [".php", ".phtml", ".js", ".mjs", ".ts", ".tsx", ".py", ".pyw", ".java"];
 function git(repo, args) {
   try {
-    return execSync(`git ${args}`, { cwd: repo, encoding: "utf8", maxBuffer: 50 * 1024 * 1024 }).trim();
+    return execSync2(`git ${args}`, { cwd: repo, encoding: "utf8", maxBuffer: 50 * 1024 * 1024 }).trim();
   } catch {
     return "";
   }
@@ -52556,7 +52690,7 @@ function taintTree(root, maxFiles = 2000) {
 import { writeFileSync as writeFileSync3, rmSync as rmSync2, mkdtempSync as mkdtempSync2, readFileSync as readFileSync13 } from "node:fs";
 import { join as join10 } from "node:path";
 import { tmpdir as tmpdir2 } from "node:os";
-import { execSync as execSync2 } from "node:child_process";
+import { execSync as execSync3 } from "node:child_process";
 init_scanner();
 function runEngine(code, language, file) {
   if (language === "php") {
@@ -52724,7 +52858,7 @@ var SOURCE_EXTS2 = [".php", ".phtml", ".php5", ".js", ".mjs", ".ts", ".tsx", ".p
 function analyzeGitDiff(repo, base = "HEAD~1", head = "HEAD") {
   const git = (args) => {
     try {
-      return execSync2(`git ${args}`, { cwd: repo, encoding: "utf8", maxBuffer: 50 * 1024 * 1024 }).trim();
+      return execSync3(`git ${args}`, { cwd: repo, encoding: "utf8", maxBuffer: 50 * 1024 * 1024 }).trim();
     } catch {
       return "";
     }
@@ -53932,7 +54066,7 @@ function createServer() {
     name: "blitzstrike",
     version: "1.0.0"
   }, {
-    instructions: "Blitz Strike is a universal penetration-testing toolbelt with three tiers — " + "BLITZ (reconnaissance / attack-surface mapping), EAGLE-EYE (source-to-sink " + "analysis), STRIKE (live validation) — plus a canonical evidence-first finding " + "engine and a multi-language taint engine.\\n\\n" + "CORE MENTAL MODEL:\\n" + "- A scanner hit is a HYPOTHESIS, not a finding. Live verification (or data-flow " + "  proof) is the verdict. Never report an unverified hit as a vulnerability.\\n" + "- Distinguish REACHABLE from mere PRESENT: a sink is only a finding when tainted " + "  input reaches it AND it is not neutralized by a sanitizer and not guarded by an " + "  auth gate.\\n\\n" + "HOW TO RUN AN AUDIT (start here):\\n" + "0. ROUTE BY TARGET. Classify automatically - no need to ask. A URL - with or " + "   without http/https - is LIVE (live_recon + the " + "   web-hunting doctrine + strike_verify). A filesystem path (exists on disk, " + "   or a code extension) is SOURCE audit (source-audit doctrine: blitz_scan + " + "   eagle_eye2 + complex_scan/route_scan). TWO DOCTRINES - never mix: live " + "   web/API -> web-hunting skill; source tree -> source-audit skill. For a WHOLE " + "   engagement prefer run_engagement (one call, no babysitting) FIRST; steps 1-5 " + "   below are the manual path for targeted follow-up on a specific finding.\\n" + "1. RECORD SCOPE. scope_check classifies the target (intel for the report) - " + "   it does NOT gate the engagement. No authorization gate: the repo disclaimer " + "   places responsibility on the user. Hunt the target you are given.\\n" + "2. MAP THE SURFACE (BLITZ). For a source tree use blitz_scan (tree) or blitz_file " + "   (single file) then enrich_scan to match sinks to escalation chains. For a live " + "   URL use live_recon for a full pass (fingerprint/WAF/tech/version/crawler/params/" + "   subdomains/API endpoints/intel correlation) — it returns next_steps. active_scan " + "   is the lighter alternative. Both record scope internally.\\n" + "3. TRACE REACHABILITY (EAGLE-EYE). For every sink, determine whether attacker " + "   input actually reaches it:\\n" + "   - PHP/JS/TS/Python/Java source: taint_file (auto-detects language) or " + "     taint_scan/taint_tree (PHP AST). list_languages shows what's supported.\\n" + "   - Any language: trace_data_flow (window heuristic) and eagle_eye/eagle_grep " + "     (function-scoped view of sinks + auth gates). Beyond taint: route_scan (route confusion / dispatch abuse / batch forwarding) and complex_scan (deserialization->POP, type juggling, mass assignment, prototype pollution, CRLF/header injection, path confusion) catch classes plain taint misses.\\n" + "   A sink counts only if tainted input reaches it unsanitized and unguarded.\\n" + "4. VALIDATE (STRIKE). Confirm live with strike_verify using a MARKER plus a " + "   NEGATIVE CONTROL plus a BASELINE. The verdict is deterministic: marker " + "   reflected AND control inert = confirmed; both reflected = false_positive; " + "   marker not reflected = unconfirmed. Feed the verdict to strike_resolve to " + "   advance the finding lifecycle (hypothesis->validating->confirmed). Use " + "   read_tool_manual for the exploit tool's manual, payload_lookup for payloads, " + "   detect_waf/tech_correlation for context.\\n" + "5. RECORD + REPORT. Create a canonical finding with finding_create, advance it " + "   through the lifecycle with finding_transition, attach confidence_score, and " + "   redact any secrets before persisting. Only report findings that survived " + "   verification.\\n\\n" + "SUPPORTING LAYERS:\\n" + "- run_engagement: ONE call does the whole audit (no babysitting). Source path -> " + "  full static pipeline + inline report (markdown/JSON/SARIF). URL target -> " + "  full autonomous live pipeline (live_recon recon + deterministic checks: reflected " + "  input, security headers, open redirect, path traversal, CORS) + findings + report. " + "  PREFER THIS FIRST for a whole engagement; drive " + "  the steps above manually only for targeted follow-up on a specific finding.\\n" + "- list_chains: 57 escalation chains (source -> sink -> impact).\\n" + "- list_languages / taint_file: multi-language taint (PHP, JS/TS, Python, Java).\\n" + "- tool_lookup / ensure_tool / list_tools: 130-tool catalog, auto-install.\\n" + "- read_playbook / list_playbooks: 17 engagement playbooks (web-app, api-security, " + "  source-code-audit, active-directory, mobile, external-attack-surface, ...).\\n" + "- skill_lookup / read_skill / list_skills: 39 skills (7 bs-* agent doctrine: " + "  orchestrate, scope, source-audit, web-hunting, verify, review, report; + " + "  technique playbooks: WP/PHP 0-day, AD, malware, mobile, cloud).\\n" + "- read_tool_manual / list_manuals: 317 tool manuals (sqlmap, nmap, commix, ...).\\n" + "- payload_lookup / read_payload: 66 payload collections (sqli, xss, ssti, xxe, lfi).\\n" + "- detect_waf / tech_correlation / cve_correlation / port_correlation: signature " + "  + correlation intelligence.\\n" + "- nvd_lookup / fofa_search: CVE lookup + asset reconnaissance (FOFA needs env creds).\\n" + "- remember / memory_lookup / memory_list: persist + recall verified knowledge.\\n\\n" + "RESOURCE MAP (know what to reach for):\\n" + "- ORCHESTRATION = the engagement framework. orchestration() returns the " + "  lifecycle (scope->recon->analyze->deep->verify->report, each with " + "  entry/exit criteria) + sub-agent team + handoff contract + termination " + "  criteria. Load it at engagement start.\\n" + "- TOOL = acts (scan, taint, verify, record, correlate). Call it to DO something.\\n" + "- PLAYBOOK = campaign plan for a target TYPE. read_playbook(web-application) " + "  BEFORE a full engagement of that type.\\n" + "- SKILL = procedure doctrine. Load the bs-* skill for the phase you are on " + "  (orchestrate, scope, web-hunting for LIVE, source-audit for SOURCE, verify, " + "  review, report); load a technique skill for the specific technique.\\n" + "- MANUAL = deep reference for ONE tool. read_tool_manual right before using it.\\n" + "- CATALOG = external tools (130: sqlmap, nmap, nuclei, ffuf, ysoserial, ...). " + "  Blitz Strike's OWN tools find + verify; the catalog is for deeper EXTERNAL " + "  exploitation/enumeration. After your own analysis confirms a finding, run " + "  tool_lookup -> ensure_tool (auto-install) -> read_tool_manual to drive the " + "  external tool that demonstrates impact.\\n" + "- PAYLOAD = attack payloads for a class. payload_lookup before crafting a probe.\\n" + "- INTELLIGENCE = knowledge. list_attack_vectors (34 categories) maps the " + "  full surface during RECON; list_chains (57) turns a finding into impact; " + "  chain_links composes chains for complex bugs; bypass_lookup when blocked; " + "  payload_lookup when crafting a probe. technique_lookup(cls) returns the " + "  DETAILED methodology + tricks for a vulnerability class (ssrf, ssti, " + "  jwt, deserialization, ...) - load before testing a class. wstg_map maps " + "  an OWASP category to coverage; taxonomy maps findings to standard IDs " + "  (OWASP Top 10 / API Top 10 / CWE / ASVS); doctrine_map shows how everything links.\\n\\n" + "NO-BABYSITTING (while scanning + testing):\\n" + "- Scan the WHOLE surface before reporting - every file, entry point, and sink. " + "  Do not stop after the first hit to ask; keep enumerating the full tree.\\n" + "- Test every hypothesis through its full verification cycle - vary probes on " + "  failure, apply bypasses when blocked, keep going. Do not pause to ask.\\n" + "- Persist through partial results: a scan that surfaces few sinks is NOT done - " + "  widen the surface, escalate detectors, dig into each reachable sink.\\n" + "- NEVER ask 'should I continue?' / 'proceed?' / approval-style questions. " + "  Continue autonomously - the moment one step verifies, start the next. " + "  Pause ONLY when: a genuinely destructive/DoS action needs consent, an " + "  external dependency beyond your control blocks you (e.g. phone OTP), or " + "  the engagement is COMPLETE (deliver the final report). Otherwise keep hunting.\\n\\n" + "PERSISTENCE (complex bugs - do not give up):\\n" + "- There is NO retry cap. A failed probe/sub-agent is diagnosed and resumed " + "  until verified or definitively ruled out - never move on unverified.\\n" + "- On any failure (empty output, blocked probe, failed sub-agent/model), call " + "  retry_guidance(signal) for concrete recovery actions before giving up.\\n" + "- On a model/sub-agent error (rate-limit, quota, timeout, 5xx, auth, " + "  unavailable), model_fallback(signal) classifies it and returns the exact " + "  recovery (retry+backoff / fall back / shrink scope / fix credentials / " + "  abort). Do not fall back blindly - the class decides.\\n" + "- One failed probe != not vulnerable. Vary: alternative sinks, other params, " + "  encoding/case/whitespace bypasses, different injection context.\\n" + "- Empty output != no findings. Re-run (server may be stale after an upgrade - " + "  /reload-mcp), or escalate to route_scan / complex_scan for classes taint " + "  misses (POP chains, type juggling, mass assignment, route confusion).\\n" + "- A 'sanitized' sink may not be safe. Challenge the sanitizer: loose == vs ===, " + "  missing EXTR_SKIP, int cast truncation, weak regex.\\n" + "- An unconfirmed hypothesis is KEPT as hypothesis - try alternate verification, " + "  then let adversarial review decide.\\n" + "- Complex chains (deserialization->POP, LFI->RCE) need the FULL chain traced, " + `  not just the first sink.\\n
+    instructions: "Blitz Strike is a universal penetration-testing toolbelt with three tiers — " + "BLITZ (reconnaissance / attack-surface mapping), EAGLE-EYE (source-to-sink " + "analysis), STRIKE (live validation) — plus a canonical evidence-first finding " + "engine and a multi-language taint engine.\\n\\n" + "CORE MENTAL MODEL:\\n" + "- A scanner hit is a HYPOTHESIS, not a finding. Live verification (or data-flow " + "  proof) is the verdict. Never report an unverified hit as a vulnerability.\\n" + "- Distinguish REACHABLE from mere PRESENT: a sink is only a finding when tainted " + "  input reaches it AND it is not neutralized by a sanitizer and not guarded by an " + "  auth gate.\\n\\n" + "HOW TO RUN AN AUDIT (start here):\\n" + "0. ROUTE BY TARGET. Classify automatically - no need to ask. A URL - with or " + "   without http/https - is LIVE (live_recon + the " + "   web-hunting doctrine + strike_verify). A filesystem path (exists on disk, " + "   or a code extension) is SOURCE audit (source-audit doctrine: blitz_scan + " + "   eagle_eye2 + complex_scan/route_scan). TWO DOCTRINES - never mix: live " + "   web/API -> web-hunting skill; source tree -> source-audit skill. For a WHOLE " + "   engagement prefer run_engagement (one call, no babysitting) FIRST; steps 1-5 " + "   below are the manual path for targeted follow-up on a specific finding.\\n" + "1. RECORD SCOPE. scope_check classifies the target (intel for the report) - " + "   it does NOT gate the engagement. No authorization gate: the repo disclaimer " + "   places responsibility on the user. Hunt the target you are given.\\n" + "2. MAP THE SURFACE (BLITZ). For a source tree use blitz_scan (tree) or blitz_file " + "   (single file) then enrich_scan to match sinks to escalation chains. For a live " + "   URL use live_recon for a full pass (fingerprint/WAF/tech/version/crawler/params/" + "   subdomains/API endpoints/intel correlation) — it returns next_steps. active_scan " + "   is the lighter alternative. Both record scope internally.\\n" + "3. TRACE REACHABILITY (EAGLE-EYE). For every sink, determine whether attacker " + "   input actually reaches it:\\n" + "   - PHP/JS/TS/Python/Java source: taint_file (auto-detects language) or " + "     taint_scan/taint_tree (PHP AST). list_languages shows what's supported.\\n" + "   - Any language: trace_data_flow (window heuristic) and eagle_eye/eagle_grep " + "     (function-scoped view of sinks + auth gates). Beyond taint: route_scan (route confusion / dispatch abuse / batch forwarding) and complex_scan (deserialization->POP, type juggling, mass assignment, prototype pollution, CRLF/header injection, path confusion) catch classes plain taint misses.\\n" + "   A sink counts only if tainted input reaches it unsanitized and unguarded.\\n" + "4. VALIDATE (STRIKE). Confirm live with strike_verify using a MARKER plus a " + "   NEGATIVE CONTROL plus a BASELINE. The verdict is deterministic: marker " + "   reflected AND control inert = confirmed; both reflected = false_positive; " + "   marker not reflected = unconfirmed. Feed the verdict to strike_resolve to " + "   advance the finding lifecycle (hypothesis->validating->confirmed). Use " + "   read_tool_manual for the exploit tool's manual, payload_lookup for payloads, " + "   detect_waf/tech_correlation for context.\\n" + "5. RECORD + REPORT. Create a canonical finding with finding_create, advance it " + "   through the lifecycle with finding_transition, attach confidence_score, and " + "   redact any secrets before persisting. Only report findings that survived " + "   verification.\\n\\n" + "SUPPORTING LAYERS:\\n" + "- run_engagement: ONE call does the whole audit (no babysitting). Source path -> " + "  full static pipeline + inline report (markdown/JSON/SARIF). URL target -> " + "  full autonomous live pipeline (live_recon recon + deterministic checks: reflected " + "  input, security headers, open redirect, path traversal, CORS) + findings + report. " + "  PREFER THIS FIRST for a whole engagement; drive " + "  the steps above manually only for targeted follow-up on a specific finding.\\n" + "- run_autonomous: run_engagement + the escalation path pre-attached to every " + "  finding (chain name/severity/prerequisites + ordered next_steps with tool_hint/" + "  success_criteria/negative_control) + a deterministic termination reason. Use it " + "  when you want the 'what to do next' already computed per finding.\\n" + "- list_chains: 57 escalation chains (source -> sink -> impact).\\n" + "- list_languages / taint_file: multi-language taint (PHP, JS/TS, Python, Java).\\n" + "- tool_lookup / ensure_tool / list_tools: 130-tool catalog, auto-install.\\n" + "- read_playbook / list_playbooks: 17 engagement playbooks (web-app, api-security, " + "  source-code-audit, active-directory, mobile, external-attack-surface, ...).\\n" + "- skill_lookup / read_skill / list_skills: 39 skills (7 bs-* agent doctrine: " + "  orchestrate, scope, source-audit, web-hunting, verify, review, report; + " + "  technique playbooks: WP/PHP 0-day, AD, malware, mobile, cloud).\\n" + "- read_tool_manual / list_manuals: 317 tool manuals (sqlmap, nmap, commix, ...).\\n" + "- payload_lookup / read_payload: 66 payload collections (sqli, xss, ssti, xxe, lfi).\\n" + "- detect_waf / tech_correlation / cve_correlation / port_correlation: signature " + "  + correlation intelligence.\\n" + "- nvd_lookup / fofa_search: CVE lookup + asset reconnaissance (FOFA needs env creds).\\n" + "- remember / memory_lookup / memory_list: persist + recall verified knowledge.\\n\\n" + "RESOURCE MAP (know what to reach for):\\n" + "- ORCHESTRATION = the engagement framework. orchestration() returns the " + "  lifecycle (scope->recon->analyze->deep->verify->report, each with " + "  entry/exit criteria) + sub-agent team + handoff contract + termination " + "  criteria. Load it at engagement start. Autonomous driving aids: plan() " + "  (deterministic phase plan + next action), delegate() (parallel-first " + "  subagent batches + retry pattern), worktree() (isolate a workstream), " + "  watchdog() (detect a stalled phase), context_prune() (compact state so " + "  you can drop raw detail).\\n" + "- TOOL = acts (scan, taint, verify, record, correlate). Call it to DO something.\\n" + "- PLAYBOOK = campaign plan for a target TYPE. read_playbook(web-application) " + "  BEFORE a full engagement of that type.\\n" + "- SKILL = procedure doctrine. Load the bs-* skill for the phase you are on " + "  (orchestrate, scope, web-hunting for LIVE, source-audit for SOURCE, verify, " + "  review, report); load a technique skill for the specific technique.\\n" + "- MANUAL = deep reference for ONE tool. read_tool_manual right before using it.\\n" + "- CATALOG = external tools (130: sqlmap, nmap, nuclei, ffuf, ysoserial, ...). " + "  Blitz Strike's OWN tools find + verify; the catalog is for deeper EXTERNAL " + "  exploitation/enumeration. After your own analysis confirms a finding, run " + "  tool_lookup -> ensure_tool (auto-install) -> read_tool_manual to drive the " + "  external tool that demonstrates impact.\\n" + "- PAYLOAD = attack payloads for a class. payload_lookup before crafting a probe.\\n" + "- INTELLIGENCE = knowledge. list_attack_vectors (34 categories) maps the " + "  full surface during RECON; list_chains (57) turns a finding into impact; " + "  chain_links composes chains for complex bugs; bypass_lookup when blocked; " + "  payload_lookup when crafting a probe. technique_lookup(cls) returns the " + "  DETAILED methodology + tricks for a vulnerability class (ssrf, ssti, " + "  jwt, deserialization, ...) - load before testing a class. wstg_map maps " + "  an OWASP category to coverage; taxonomy maps findings to standard IDs " + "  (OWASP Top 10 / API Top 10 / CWE / ASVS); doctrine_map shows how everything links.\\n\\n" + "NO-BABYSITTING (while scanning + testing):\\n" + "- Scan the WHOLE surface before reporting - every file, entry point, and sink. " + "  Do not stop after the first hit to ask; keep enumerating the full tree.\\n" + "- Test every hypothesis through its full verification cycle - vary probes on " + "  failure, apply bypasses when blocked, keep going. Do not pause to ask.\\n" + "- Persist through partial results: a scan that surfaces few sinks is NOT done - " + "  widen the surface, escalate detectors, dig into each reachable sink.\\n" + "- NEVER ask 'should I continue?' / 'proceed?' / approval-style questions. " + "  Continue autonomously - the moment one step verifies, start the next. " + "  Pause ONLY when: a genuinely destructive/DoS action needs consent, an " + "  external dependency beyond your control blocks you (e.g. phone OTP), or " + "  the engagement is COMPLETE (deliver the final report). Otherwise keep hunting.\\n\\n" + "PERSISTENCE (complex bugs - do not give up):\\n" + "- There is NO retry cap. A failed probe/sub-agent is diagnosed and resumed " + "  until verified or definitively ruled out - never move on unverified.\\n" + "- On any failure (empty output, blocked probe, failed sub-agent/model), call " + "  retry_guidance(signal) for concrete recovery actions before giving up.\\n" + "- On a model/sub-agent error (rate-limit, quota, timeout, 5xx, auth, " + "  unavailable), model_fallback(signal) classifies it and returns the exact " + "  recovery (retry+backoff / fall back / shrink scope / fix credentials / " + "  abort). Do not fall back blindly - the class decides.\\n" + "- One failed probe != not vulnerable. Vary: alternative sinks, other params, " + "  encoding/case/whitespace bypasses, different injection context.\\n" + "- Empty output != no findings. Re-run (server may be stale after an upgrade - " + "  /reload-mcp), or escalate to route_scan / complex_scan for classes taint " + "  misses (POP chains, type juggling, mass assignment, route confusion).\\n" + "- A 'sanitized' sink may not be safe. Challenge the sanitizer: loose == vs ===, " + "  missing EXTR_SKIP, int cast truncation, weak regex.\\n" + "- An unconfirmed hypothesis is KEPT as hypothesis - try alternate verification, " + "  then let adversarial review decide.\\n" + "- Complex chains (deserialization->POP, LFI->RCE) need the FULL chain traced, " + `  not just the first sink.\\n
 ` + "- When a probe is BLOCKED by a WAF/filter/sanitizer, call bypass_lookup(defense) " + "  for concrete techniques (encoding, case, comment insertion, chunked, null byte, " + "  double-encoding, verb tampering, header spoofing) and apply them - do not give up.\\n" + "- CHAIN findings: a low-severity finding may escalate — do not assume it is the " + "  end of the story. list_chains / chain_links show whether it composes into a " + "  multi-stage chain (e.g. SSRF->metadata, SQLi->RCE, LFI->RCE, upload->shell). " + "  Check before concluding, and do not overclaim an escalation that is not there.\\n\\n" + "- Complex bugs chain ACROSS chains: after confirming one chain, chain_links " + "  shows what follows (SSRF->metadata then default-creds->pivot via stolen IAM). " + "  Trace the full composition, not just a single chain.\\n\\n" + "HYPOTHESIS EXPANSION (from each finding, enumerate + verify every follow-up):\\n" + "- A confirmed finding is the START, not the end. From each one, enumerate " + "  the follow-ups: ESCALATE (list_chains + chain_links), CHAIN (chain_links " + "  from->to), BYPASS (bypass_lookup), PIVOT (new endpoints/sinks now reachable).\\n" + "- Every follow-up (this might happen, this could chain, maybe bypassable) is " + "  a HYPOTHESIS, not a finding. Verify EACH via strike_verify (marker + negative " + "  control) before it enters the report.\\n" + "- Never report speculation: might / could / possibly live in the " + "  hypothesis list, never in the report. Every report line is a verified fact.\\n\\n" + "PROFESSIONAL JUDGMENT (act like a senior tester, not a checklist runner):\\n" + "- PRIORITIZE by severity, computed deterministically with cvss_score (impact x " + "  exploitability) and weighted by escalation potential (chain_links). Do NOT rank " + "  by bug class - a critical stored XSS that yields account takeover outranks an " + "  unreachable blind RCE. Weight by the engagement context (which asset/data is " + "  most sensitive for THIS target).\\n" + "- Recognize dead ends: if a branch yields nothing after a few varied probes, " + "  note it and move on. Do not tunnel-vision one sink at the expense of the rest.\\n" + "- Live-test with care: respect rate limits, do not lock accounts, do not " + "  trigger DoS. Test non-destructively and prove impact with the smallest payload.\\n" + "- Know when done: stop when the termination criteria are met, not when you feel " + "  tired. A thorough audit over a rushed one, every time.\\n\\n" + "IRON RULES:\\n" + "- No authorization gate - the repo disclaimer places responsibility on the " + "  user. Hunt the target you are given (like nmap/sqlmap/Burp, which do not " + "  ask). The one hard stop: malicious/third-party infra (cari pelaku) - hold/static.\\n" + "  need finer control."
   });
   server.registerTool("blitz_scan", {
@@ -54141,6 +54275,28 @@ ${formatPlanChecklist(r.plan)}` : "";
       content: [
         { type: "text", text: banner ? `${banner}${planText}
 
+${statusLine}` : statusLine },
+        { type: "text", text: JSON.stringify(r) }
+      ]
+    };
+  });
+  server.registerTool("run_autonomous", {
+    title: "STRIKE autonomous engagement (surface + escalation)",
+    description: "ORCHESTRATION: the self-driving pass — surface pipeline + escalation path attached to every finding, in ONE call. Each finding comes back with its chain (name/severity/prerequisites) and ordered next_steps (action + tool_hint + success_criteria + negative_control), so you never guess what to do next. Returns the full report + a deterministic termination reason. Prefer this over run_engagement when you want the escalation paths pre-attached.",
+    inputSchema: {
+      target: string2().describe("Source path or URL"),
+      scope: string2().optional().describe("Scope (for URL targets)"),
+      mode: string2().optional().describe("Engagement mode"),
+      max_passes: number2().int().min(1).max(5).optional().describe("Max deepen passes (default 3)")
+    }
+  }, async ({ target, scope, mode, max_passes }) => {
+    const r = await runAutonomous(target, scope ?? "", mode ?? "bug-bounty", max_passes ?? 3);
+    const banner = typeof r.banner === "string" ? r.banner : "";
+    const statusLine = `→ ${Array.isArray(r.findings) ? r.findings.length : 0} finding(s), termination=${r.termination ?? "complete"}`;
+    return {
+      content: [
+        { type: "text", text: banner ? `${banner}
+
 ${statusLine}` : statusLine },
         { type: "text", text: JSON.stringify(r) }
       ]
@@ -54552,6 +54708,66 @@ ${statusLine}` : statusLine },
   }, async () => {
     return { content: [{ type: "text", text: JSON.stringify(engagementStatus()) }] };
   });
+  server.registerTool("plan", {
+    title: "Generate an engagement plan",
+    description: "ORCHESTRATION: generate the deterministic engagement plan (phase order, tool per phase, evidence required, risk, permission) for a target + mode. This is the planning pass (like a planning sub-agent) — call ONCE at the start, then execute without re-planning.",
+    inputSchema: {
+      target: string2().describe("Target (path or URL) — for context"),
+      mode: string2().optional().describe("Engagement mode"),
+      state: string2().optional().describe("Current phase (default: plan)")
+    }
+  }, async ({ target, mode, state }) => {
+    const executedState = state ?? "plan";
+    const plan = buildPlan(true, mode ?? "bug-bounty", executedState);
+    const guidance = computeGuidance(executedState, true);
+    return { content: [{ type: "text", text: JSON.stringify({ target, mode: mode ?? "bug-bounty", plan, guidance }, null, 2) }] };
+  });
+  server.registerTool("delegate", {
+    title: "Plan subagent delegation (parallel-first)",
+    description: "ORCHESTRATION: decompose a workstream list into parallel-first dispatch batches (named dependencies only) + the retry pattern. Fire every batch in ONE message; sequential is the exception. Returns batches so the LLM dispatches without re-deriving the dependency graph.",
+    inputSchema: {
+      tasks: array(object2({
+        id: string2(),
+        objective: string2(),
+        depends_on: array(string2()).optional()
+      })).describe("Workstreams to delegate")
+    }
+  }, async ({ tasks }) => {
+    const { batches, total } = decomposeBatches(tasks);
+    return { content: [{ type: "text", text: JSON.stringify({ total, batches, retry: retryPattern("a workstream") }, null, 2) }] };
+  });
+  server.registerTool("worktree", {
+    title: "Isolate a git worktree",
+    description: "ORCHESTRATION: create/list/remove an isolated git worktree so parallel workstreams don't conflict on the same files. add: `git worktree add -b <branch> <path>`; list: enumerate existing; remove: clean up after a workstream.",
+    inputSchema: {
+      action: _enum(["add", "list", "remove"]).describe("Operation"),
+      path: string2().optional().describe("Path (for add/remove)"),
+      branch: string2().optional().describe("Branch name (for add; default auto)")
+    }
+  }, async ({ action, path, branch }) => {
+    return { content: [{ type: "text", text: JSON.stringify(worktree(action, path, branch)) }] };
+  });
+  server.registerTool("watchdog", {
+    title: "Check engagement health (stall detection)",
+    description: "ORCHESTRATION: detect a stalled phase — no progress (idle > 15 min, no hypotheses). Returns STALLED / PROGRESSING / IDLE plus the concrete intervention so the LLM unblocks instead of looping.",
+    inputSchema: {
+      phase: string2().describe("Current phase"),
+      hypotheses: number2().int().describe("Number of tracked hypotheses"),
+      last_activity_ms: number2().int().describe("Timestamp (ms) of last activity")
+    }
+  }, async ({ phase, hypotheses, last_activity_ms }) => {
+    return { content: [{ type: "text", text: JSON.stringify(watchdog(phase, hypotheses, last_activity_ms)) }] };
+  });
+  server.registerTool("context_prune", {
+    title: "Compact the engagement state",
+    description: "ORCHESTRATION: produce a compact summary (phase, confirmed/pending/rejected counts, top findings) so the LLM can DROP raw detail from its context and keep only this. The engagement tracker is the source of truth — re-read it instead of relying on memory.",
+    inputSchema: {
+      phase: string2().describe("Current phase"),
+      findings: array(record(string2(), unknown())).describe("Findings to compact")
+    }
+  }, async ({ phase, findings }) => {
+    return { content: [{ type: "text", text: JSON.stringify(compactState(phase, findings)) }] };
+  });
   server.registerTool("read_payload", {
     title: "Read a payload collection",
     description: "INTEL: read the full payload collection for a category (or a specific file).",
@@ -55084,7 +55300,7 @@ import { spawnSync as spawnSync3 } from "node:child_process";
 import { readFileSync as readFileSync19, existsSync as existsSync8 } from "node:fs";
 import { join as join15 } from "node:path";
 import { fileURLToPath as fileURLToPath9 } from "node:url";
-import { execSync as execSync3 } from "node:child_process";
+import { execSync as execSync4 } from "node:child_process";
 var ROOT8 = join15(fileURLToPath9(new URL(".", import.meta.url)), "..");
 function validateRelease() {
   const checks = [];
@@ -55102,13 +55318,13 @@ function validateRelease() {
     checks.push({ name: "CHANGELOG version", ok: false, detail: "CHANGELOG.md not found" });
   }
   try {
-    const tag = execSync3("git describe --tags --abbrev=0", { cwd: ROOT8, encoding: "utf8" }).trim();
+    const tag = execSync4("git describe --tags --abbrev=0", { cwd: ROOT8, encoding: "utf8" }).trim();
     checks.push({ name: "Git tag", ok: tag === `v${VERSION}`, detail: tag, soft: true });
   } catch {
     checks.push({ name: "Git tag", ok: true, detail: "unavailable (no git)", soft: true });
   }
   try {
-    const published = execSync3("npm view blitzstrike version", { encoding: "utf8", timeout: 8000 }).trim();
+    const published = execSync4("npm view blitzstrike version", { encoding: "utf8", timeout: 8000 }).trim();
     checks.push({ name: "npm registry version", ok: published === VERSION, detail: published, soft: true });
   } catch {
     checks.push({ name: "npm registry version", ok: true, detail: "unavailable (offline)", soft: true });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blitzstrike",
-  "version": "1.3.0",
+  "version": "1.5.0",
   "description": "Blitz Strike — a universal MCP security-audit toolbelt. BLITZ sweeps the attack surface, EAGLE-EYE traces source-to-sink, STRIKE verifies live. 57 attack chains, 130-tool catalog, intelligence data layer. One server, every agent.",
   "type": "module",
   "bin": {