@pentoshi/clai 3.6.0 → 3.7.2

package/dist/agent/scratch-write.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"scratch-write.js","sourceRoot":"","sources":["../../src/agent/scratch-write.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAG9C,MAAM,sBAAsB,GAAG,IAAI,GAAG,CAAC;IACrC,UAAU;IACV,cAAc;IACd,SAAS;IACT,iBAAiB;IACjB,WAAW;IACX,WAAW;CACZ,CAAC,CAAC;AAEH,SAAS,eAAe,CAAC,IAAY;IACnC,IAAI,IAAI,KAAK,GAAG;QAAE,OAAO,OAAO,EAAE,CAAC;IACnC,IAAI,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QACpD,OAAO,OAAO,CAAC,OAAO,EAAE,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3C,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,IAAc;IACpD,IAAI,IAAI,CAAC,IAAI,KAAK,cAAc,EAAE,CAAC;QACjC,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;QAC9B,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QACrC,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;YAC1B,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,IAAI,KAAK,EAAE,CAAC;gBAC1D,MAAM,CAAC,GAAI,KAA4B,CAAC,IAAI,CAAC;gBAC7C,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC;oBAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YAC3D,CAAC;QACH,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;IAC/B,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACnE,OAAO,CAAC,OAAO,CAAC,CAAC;AACnB,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,kBAAkB,CAAC,IAAc,EAAE,UAAkB;IACnE,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,MAAM,KAAK,GAAG,uBAAuB,CAAC,IAAI,CAAC,CAAC;IAC5C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACrC,MAAM,eAAe,GAAG,OAAO,CAAC,UAAU,CAAC,CAAC;IAC5C,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QACzB,MAAM,QAAQ,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;QACtC,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;QACnC,MAAM,GAAG,GAAG,QAAQ,CAAC,eAAe,EAAE,QAAQ,CAAC,CAAC;QAChD,OAAO,GAAG,KAAK,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/agent/session-policy.d.ts

@@ -14,6 +14,13 @@ export interface SessionPolicy {
 }
 export declare function createSessionPolicy(sessionId?: string): SessionPolicy;
 export declare function isPreApprovalAllowedTool(name: string): boolean;
+/** True when the tool is allowed in plan mode (gather-only). */
+export declare function isPlanModeAllowedTool(name: string): boolean;
+/**
+ * Shell commands blocked in plan mode: project scaffold/mutate and active
+ * exploit/C2. Recon, enum, fuzz, long nmap, installs of scanners are allowed.
+ */
+export declare function isPlanModeAllowedShellCommand(command: string): boolean;
 /**
  * A plan's persisted status is the durable source of truth for "has this
  * plan been approved" — session.planApproved is in-memory only and resets to

package/dist/agent/session-policy.js

@@ -22,11 +22,72 @@ const PRE_APPROVAL_ALLOWED_TOOLS = new Set([
     "fs.search",
     "sysinfo",
     "tool.batch",
+    "tool.check",
     "net.context",
+    "web.search",
+    "web.fetch",
+    "dns.lookup",
+    "whois.lookup",
+    "http.fetch",
+    "net.scan",
+    "pentest.recon",
+    "wordlist.find",
+    "image.ocr",
+    "pdf.read",
+    "shell.jobs",
+    "shell.tail",
+]);
+/**
+ * Plan mode: gather freely (recon, scans, enum, research, tool installs).
+ * Block only project mutation and clear active exploitation — put those in
+ * the plan as post-accept tasks.
+ */
+const PLAN_MODE_BLOCKED_TOOLS = new Set([
+    "fs.write",
+    "fs.writeMany",
+    "fs.edit",
+    "fs.append",
+    "fs.replaceLines",
+    "fs.delete",
 ]);
 export function isPreApprovalAllowedTool(name) {
     return PRE_APPROVAL_ALLOWED_TOOLS.has(name);
 }
+/** True when the tool is allowed in plan mode (gather-only). */
+export function isPlanModeAllowedTool(name) {
+    if (PLAN_MODE_BLOCKED_TOOLS.has(name))
+        return false;
+    return true;
+}
+/**
+ * Shell commands blocked in plan mode: project scaffold/mutate and active
+ * exploit/C2. Recon, enum, fuzz, long nmap, installs of scanners are allowed.
+ */
+export function isPlanModeAllowedShellCommand(command) {
+    const c = command.trim();
+    if (!c)
+        return false;
+    if (/\b(npm\s+create|npx\s+(?:--yes\s+)?create-|yarn\s+create|pnpm\s+create|bun\s+create|cargo\s+new|rails\s+new|poetry\s+new|flutter\s+create|django-admin\s+startproject|composer\s+create-project|npm\s+init\s+vite|create-next-app|create-react-app)\b/i.test(c)) {
+        return false;
+    }
+    if (/\b(fs\.write|tee\s+>|>>\s*\/|cat\s*>\s*|printf\s+.*>\s*)/i.test(c) &&
+        !/\b(tee\s+\/tmp|tee\s+\$\{?TMP|>>\s*\/tmp|\/var\/folders)\b/i.test(c)) {
+        // Writing into user project trees is blocked; tmp artifact writes OK.
+        if (!/\b(\/tmp|TMPDIR|scratch|\.clai\/outputs)\b/i.test(c)) {
+            return false;
+        }
+    }
+    if (/\b(msfconsole|msfvenom|metasploit|exploit\/|use\s+exploit|revshell|reverse\s+shell|nc\s+-[el].*\d|ncat\s+-[el]|bash\s+-i\s+>&\s*\/dev\/tcp)\b/i.test(c)) {
+        return false;
+    }
+    if (/\b(sqlmap\b.*(?:--os-shell|--os-pwn|--sql-shell|--crawl)|hydra\s+.+\s+-l\s|medusa\s+.*-M\s)\b/i.test(c)) {
+        return false;
+    }
+    if (/\brm\s+(-[a-zA-Z]*f|-[a-zA-Z]*r).*(\/|~)/i.test(c)) {
+        return false;
+    }
+    return true;
+}
 /**
  * A plan's persisted status is the durable source of truth for "has this
  * plan been approved" — session.planApproved is in-memory only and resets to

package/dist/agent/session-policy.js.map

@@ -1 +1 @@
-{"version":3,"file":"session-policy.js","sourceRoot":"","sources":["../../src/agent/session-policy.ts"],"names":[],"mappings":"AAWA,MAAM,UAAU,mBAAmB,CAAC,SAAkB;IACpD,OAAO;QACL,KAAK,EAAE,IAAI,GAAG,EAAE;QAChB,iBAAiB,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE;QACnC,SAAS,EACP,SAAS;YACT,QAAQ,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE;QAC7E,YAAY,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE;KAC/B,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAS;IACjD,aAAa;IACb,aAAa;IACb,SAAS;IACT,SAAS;IACT,WAAW;IACX,SAAS;IACT,YAAY;IACZ,aAAa;CACd,CAAC,CAAC;AAEH,MAAM,UAAU,wBAAwB,CAAC,IAAY;IACnD,OAAO,0BAA0B,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;AAC9C,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CAAC,MAAsB;IAC3D,OAAO,MAAM,KAAK,OAAO,CAAC;AAC5B,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,MAAkC;IAChE,OAAO,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,WAAW,CAAC;AACxD,CAAC;AAKD,MAAM,UAAU,YAAY,CAAC,KAAc,EAAE,MAAoB;IAC/D,OAAO,CACL,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;QACxB,CAAC,KAAK,YAAY,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,CAAC,CACxD,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,oBAAoB,CAClC,MAAc,EACd,iBAA0B;IAE1B,IAAI,CAAC,iBAAiB;QAAE,OAAO,IAAI,CAAC;IACpC,OAAO,sDAAsD,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC7E,CAAC"}
+{"version":3,"file":"session-policy.js","sourceRoot":"","sources":["../../src/agent/session-policy.ts"],"names":[],"mappings":"AAWA,MAAM,UAAU,mBAAmB,CAAC,SAAkB;IACpD,OAAO;QACL,KAAK,EAAE,IAAI,GAAG,EAAE;QAChB,iBAAiB,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE;QACnC,SAAS,EACP,SAAS;YACT,QAAQ,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE;QAC7E,YAAY,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE;KAC/B,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAS;IACjD,aAAa;IACb,aAAa;IACb,SAAS;IACT,SAAS;IACT,WAAW;IACX,SAAS;IACT,YAAY;IACZ,YAAY;IACZ,aAAa;IACb,YAAY;IACZ,WAAW;IACX,YAAY;IACZ,cAAc;IACd,YAAY;IACZ,UAAU;IACV,eAAe;IACf,eAAe;IACf,WAAW;IACX,UAAU;IACV,YAAY;IACZ,YAAY;CACb,CAAC,CAAC;AAEH;;;;GAIG;AACH,MAAM,uBAAuB,GAAG,IAAI,GAAG,CAAS;IAC9C,UAAU;IACV,cAAc;IACd,SAAS;IACT,WAAW;IACX,iBAAiB;IACjB,WAAW;CACZ,CAAC,CAAC;AAEH,MAAM,UAAU,wBAAwB,CAAC,IAAY;IACnD,OAAO,0BAA0B,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;AAC9C,CAAC;AAED,gEAAgE;AAChE,MAAM,UAAU,qBAAqB,CAAC,IAAY;IAChD,IAAI,uBAAuB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACpD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,6BAA6B,CAAC,OAAe;IAC3D,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC;IACzB,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC;IACrB,IACE,wPAAwP,CAAC,IAAI,CAC3P,CAAC,CACF,EACD,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IACE,2DAA2D,CAAC,IAAI,CAAC,CAAC,CAAC;QACnE,CAAC,6DAA6D,CAAC,IAAI,CAAC,CAAC,CAAC,EACtE,CAAC;QACD,sEAAsE;QACtE,IAAI,CAAC,6CAA6C,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;YAC3D,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IACD,IACE,gJAAgJ,CAAC,IAAI,CACnJ,CAAC,CACF,EACD,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IACE,gGAAgG,CAAC,IAAI,CACnG,CAAC,CACF,EACD,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,2CAA2C,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QACxD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CAAC,MAAsB;IAC3D,OAAO,MAAM,KAAK,OAAO,CAAC;AAC5B,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,MAAkC;IAChE,OAAO,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,WAAW,CAAC;AACxD,CAAC;AAKD,MAAM,UAAU,YAAY,CAAC,KAAc,EAAE,MAAoB;IAC/D,OAAO,CACL,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;QACxB,CAAC,KAAK,YAAY,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,CAAC,CACxD,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,oBAAoB,CAClC,MAAc,EACd,iBAA0B;IAE1B,IAAI,CAAC,iBAAiB;QAAE,OAAO,IAAI,CAAC;IACpC,OAAO,sDAAsD,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC7E,CAAC"}

package/dist/agent/session-state.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Compact SESSION STATE / WORKING MEMORY block for the model.
+ * Survives as an updatable system message; re-injected after compaction.
+ */
+export declare const SESSION_STATE_PREFIX = "SESSION STATE / WORKING MEMORY";
+export interface SessionStateSnapshot {
+    goal?: string | undefined;
+    projectRoot?: string | undefined;
+    packageManager?: string | undefined;
+    planStatus?: string | undefined;
+    planKind?: string | undefined;
+    /** e.g. t3 in_progress "implement todo" */
+    openTask?: string | undefined;
+    /** short list of pending task ids/titles */
+    pendingTasks?: string[] | undefined;
+    /** short list of done task ids */
+    doneTasks?: string[] | undefined;
+    featureAppRequired?: boolean | undefined;
+    featureSeen?: boolean | undefined;
+    scaffoldOk?: boolean | undefined;
+    serverStarted?: boolean | undefined;
+    serverProbedOk?: boolean | undefined;
+    lastProbeFailed?: boolean | undefined;
+    lastOkTool?: string | undefined;
+    nextHint?: string | undefined;
+    /** pentest residual one-liner */
+    engagementNote?: string | undefined;
+}
+/** Build a short, high-signal state block (aim < 400 tokens). */
+export declare function buildSessionStateBlock(s: SessionStateSnapshot): string;
+/** Infer a short next-action hint from flags (no LLM). */
+export declare function inferNextHint(s: SessionStateSnapshot): string | undefined;
+/**
+ * Upsert SESSION STATE as a system message after the primary system prompt.
+ * Mutates the messages array in place.
+ */
+export declare function upsertSessionStateMessage(messages: Array<{
+    role: string;
+    content: string;
+}>, block: string): void;
+export declare function isSessionStateMessage(content: string): boolean;

package/dist/agent/session-state.js

@@ -0,0 +1,107 @@
+/**
+ * Compact SESSION STATE / WORKING MEMORY block for the model.
+ * Survives as an updatable system message; re-injected after compaction.
+ */
+export const SESSION_STATE_PREFIX = "SESSION STATE / WORKING MEMORY";
+/** Build a short, high-signal state block (aim < 400 tokens). */
+export function buildSessionStateBlock(s) {
+    const lines = [SESSION_STATE_PREFIX];
+    if (s.goal)
+        lines.push(`goal: ${oneLine(s.goal, 160)}`);
+    if (s.projectRoot)
+        lines.push(`project_root: ${s.projectRoot}`);
+    if (s.packageManager)
+        lines.push(`package_manager: ${s.packageManager}`);
+    if (s.planStatus || s.planKind) {
+        lines.push(`plan: ${s.planKind ?? "?"} status=${s.planStatus ?? "?"}`.trim());
+    }
+    if (s.openTask)
+        lines.push(`open_task: ${oneLine(s.openTask, 120)}`);
+    if (s.doneTasks?.length) {
+        lines.push(`done: ${s.doneTasks.slice(0, 8).join("; ")}`);
+    }
+    if (s.pendingTasks?.length) {
+        lines.push(`pending: ${s.pendingTasks.slice(0, 8).join("; ")}`);
+    }
+    const flags = [];
+    if (s.featureAppRequired) {
+        flags.push(`feature_needed=true feature_seen=${s.featureSeen ? "true" : "false"}`);
+    }
+    if (s.scaffoldOk)
+        flags.push("scaffold_ok=true");
+    if (s.serverStarted)
+        flags.push("server_started=true");
+    if (s.serverProbedOk)
+        flags.push("probe_ok=true");
+    if (s.lastProbeFailed)
+        flags.push("last_probe_failed=true");
+    if (flags.length)
+        lines.push(`flags: ${flags.join(" ")}`);
+    if (s.lastOkTool)
+        lines.push(`last_ok_tool: ${s.lastOkTool}`);
+    if (s.engagementNote)
+        lines.push(`note: ${oneLine(s.engagementNote, 160)}`);
+    if (s.nextHint)
+        lines.push(`next: ${oneLine(s.nextHint, 200)}`);
+    lines.push("Use this state. Prefer evidence over claims. Update work via tools; do not invent progress.");
+    return lines.join("\n");
+}
+function oneLine(s, max) {
+    const t = s.replace(/\s+/g, " ").trim();
+    return t.length <= max ? t : t.slice(0, max - 1) + "…";
+}
+/** Infer a short next-action hint from flags (no LLM). */
+export function inferNextHint(s) {
+    if (s.featureAppRequired && !s.featureSeen) {
+        return "Implement the product feature (replace starter boilerplate); do not start the server yet.";
+    }
+    if (s.lastProbeFailed) {
+        return "Fix the app error (fs.edit/write), then re-probe localhost — diagnosis alone is incomplete.";
+    }
+    // Leave-running / verify open with server already proven → close out, don't restart
+    if (s.serverStarted &&
+        s.openTask &&
+        /\bleave\s+.*running|for\s+user\s+to\s+test|keep\s+.*running/i.test(s.openTask)) {
+        return "Runtime already proven: confirm still alive if needed, task.update done, report URL + port + job id — do not restart.";
+    }
+    if (s.serverStarted && !s.serverProbedOk && !s.lastProbeFailed) {
+        return "Confirm ready (shell.tail or lsof LISTEN or localhost GET); leave server running; report URL + job id. Do not thrash ports.";
+    }
+    if (s.openTask && /\brecon|enumerat|scan|dns|nmap/i.test(s.openTask)) {
+        return "Run remote recon tools against the in-scope target (dns/http/net.scan), then mark done with evidence.";
+    }
+    if (s.openTask) {
+        return `Finish open task, verify with tools, mark done only with evidence.`;
+    }
+    if (s.pendingTasks?.length) {
+        return `Open next pending task with task.update in_progress, then execute it.`;
+    }
+    return undefined;
+}
+/**
+ * Upsert SESSION STATE as a system message after the primary system prompt.
+ * Mutates the messages array in place.
+ */
+export function upsertSessionStateMessage(messages, block) {
+    const content = block.startsWith(SESSION_STATE_PREFIX)
+        ? block
+        : `${SESSION_STATE_PREFIX}\n${block}`;
+    const idx = messages.findIndex((m) => m.role === "system" &&
+        typeof m.content === "string" &&
+        m.content.startsWith(SESSION_STATE_PREFIX));
+    if (idx >= 0) {
+        messages[idx].content = content;
+        return;
+    }
+    // Insert after first system message if present, else at front.
+    if (messages[0]?.role === "system") {
+        messages.splice(1, 0, { role: "system", content });
+    }
+    else {
+        messages.unshift({ role: "system", content });
+    }
+}
+export function isSessionStateMessage(content) {
+    return content.startsWith(SESSION_STATE_PREFIX);
+}
+//# sourceMappingURL=session-state.js.map

package/dist/agent/session-state.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"session-state.js","sourceRoot":"","sources":["../../src/agent/session-state.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,CAAC,MAAM,oBAAoB,GAAG,gCAAgC,CAAC;AA0BrE,iEAAiE;AACjE,MAAM,UAAU,sBAAsB,CAAC,CAAuB;IAC5D,MAAM,KAAK,GAAa,CAAC,oBAAoB,CAAC,CAAC;IAC/C,IAAI,CAAC,CAAC,IAAI;QAAE,KAAK,CAAC,IAAI,CAAC,SAAS,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IACxD,IAAI,CAAC,CAAC,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;IAChE,IAAI,CAAC,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC;IACzE,IAAI,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC;QAC/B,KAAK,CAAC,IAAI,CACR,SAAS,CAAC,CAAC,QAAQ,IAAI,GAAG,WAAW,CAAC,CAAC,UAAU,IAAI,GAAG,EAAE,CAAC,IAAI,EAAE,CAClE,CAAC;IACJ,CAAC;IACD,IAAI,CAAC,CAAC,QAAQ;QAAE,KAAK,CAAC,IAAI,CAAC,cAAc,OAAO,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IACrE,IAAI,CAAC,CAAC,SAAS,EAAE,MAAM,EAAE,CAAC;QACxB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,CAAC,CAAC,YAAY,EAAE,MAAM,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,CAAC,CAAC,kBAAkB,EAAE,CAAC;QACzB,KAAK,CAAC,IAAI,CAAC,oCAAoC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;IACrF,CAAC;IACD,IAAI,CAAC,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACjD,IAAI,CAAC,CAAC,aAAa;QAAE,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IACvD,IAAI,CAAC,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAClD,IAAI,CAAC,CAAC,eAAe;QAAE,KAAK,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC5D,IAAI,KAAK,CAAC,MAAM;QAAE,KAAK,CAAC,IAAI,CAAC,UAAU,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC1D,IAAI,CAAC,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC;IAC9D,IAAI,CAAC,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,SAAS,OAAO,CAAC,CAAC,CAAC,cAAc,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAC5E,IAAI,CAAC,CAAC,QAAQ;QAAE,KAAK,CAAC,IAAI,CAAC,SAAS,OAAO,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAChE,KAAK,CAAC,IAAI,CACR,6FAA6F,CAC9F,CAAC;IACF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,SAAS,OAAO,CAAC,CAAS,EAAE,GAAW;IACrC,MAAM,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IACxC,OAAO,CAAC,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC;AACzD,CAAC;AAED,0DAA0D;AAC1D,MAAM,UAAU,aAAa,CAAC,CAAuB;IACnD,IAAI,CAAC,CAAC,kBAAkB,IAAI,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;QAC3C,OAAO,2FAA2F,CAAC;IACrG,CAAC;IACD,IAAI,CAAC,CAAC,eAAe,EAAE,CAAC;QACtB,OAAO,6FAA6F,CAAC;IACvG,CAAC;IACD,oFAAoF;IACpF,IACE,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,QAAQ;QACV,8DAA8D,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,EAC/E,CAAC;QACD,OAAO,uHAAuH,CAAC;IACjI,CAAC;IACD,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,CAAC,cAAc,IAAI,CAAC,CAAC,CAAC,eAAe,EAAE,CAAC;QAC/D,OAAO,6HAA6H,CAAC;IACvI,CAAC;IACD,IAAI,CAAC,CAAC,QAAQ,IAAI,iCAAiC,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC;QACrE,OAAO,uGAAuG,CAAC;IACjH,CAAC;IACD,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC;QACf,OAAO,oEAAoE,CAAC;IAC9E,CAAC;IACD,IAAI,CAAC,CAAC,YAAY,EAAE,MAAM,EAAE,CAAC;QAC3B,OAAO,uEAAuE,CAAC;IACjF,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,yBAAyB,CACvC,QAAkD,EAClD,KAAa;IAEb,MAAM,OAAO,GAAG,KAAK,CAAC,UAAU,CAAC,oBAAoB,CAAC;QACpD,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,GAAG,oBAAoB,KAAK,KAAK,EAAE,CAAC;IACxC,MAAM,GAAG,GAAG,QAAQ,CAAC,SAAS,CAC5B,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,IAAI,KAAK,QAAQ;QACnB,OAAO,CAAC,CAAC,OAAO,KAAK,QAAQ;QAC7B,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,oBAAoB,CAAC,CAC7C,CAAC;IACF,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC;QACb,QAAQ,CAAC,GAAG,CAAE,CAAC,OAAO,GAAG,OAAO,CAAC;QACjC,OAAO;IACT,CAAC;IACD,+DAA+D;IAC/D,IAAI,QAAQ,CAAC,CAAC,CAAC,EAAE,IAAI,KAAK,QAAQ,EAAE,CAAC;QACnC,QAAQ,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;IACrD,CAAC;SAAM,CAAC;QACN,QAAQ,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;IAChD,CAAC;AACH,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAe;IACnD,OAAO,OAAO,CAAC,UAAU,CAAC,oBAAoB,CAAC,CAAC;AAClD,CAAC"}

package/dist/agent/step-budget.d.ts

@@ -0,0 +1,14 @@
+import type { TaskAnalysis } from "./task-analyzer.js";
+export interface StepBudgetInput {
+    analysis: TaskAnalysis;
+    maxSteps: number;
+    buildLike: boolean;
+    pentestLike: boolean;
+    hasHistory: boolean;
+}
+/**
+ * Productive tool-step budget for a turn. Multi-step builds/pentests get the
+ * full ceiling; short follow-ups with history stay at least standard-sized.
+ */
+export declare function computeStepBudget(input: StepBudgetInput): number;
+export declare function computeMaxIterations(stepBudget: number): number;

package/dist/agent/step-budget.js

@@ -0,0 +1,26 @@
+/**
+ * Productive tool-step budget for a turn. Multi-step builds/pentests get the
+ * full ceiling; short follow-ups with history stay at least standard-sized.
+ */
+export function computeStepBudget(input) {
+    const { analysis, maxSteps, buildLike, pentestLike, hasHistory } = input;
+    let budget = analysis.complexity === "simple"
+        ? 20
+        : analysis.complexity === "standard"
+            ? 40
+            : maxSteps;
+    if (buildLike || pentestLike || analysis.shouldPlan) {
+        budget = Math.max(budget, maxSteps);
+    }
+    else if (hasHistory) {
+        budget = Math.max(budget, 40);
+    }
+    if (analysis.complexity === "complex") {
+        budget = Math.max(budget, Math.min(maxSteps, 80));
+    }
+    return budget;
+}
+export function computeMaxIterations(stepBudget) {
+    return stepBudget * 3;
+}
+//# sourceMappingURL=step-budget.js.map

package/dist/agent/step-budget.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"step-budget.js","sourceRoot":"","sources":["../../src/agent/step-budget.ts"],"names":[],"mappings":"AAUA;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,KAAsB;IACtD,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,SAAS,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,KAAK,CAAC;IACzE,IAAI,MAAM,GACR,QAAQ,CAAC,UAAU,KAAK,QAAQ;QAC9B,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,QAAQ,CAAC,UAAU,KAAK,UAAU;YAClC,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC,QAAQ,CAAC;IAEjB,IAAI,SAAS,IAAI,WAAW,IAAI,QAAQ,CAAC,UAAU,EAAE,CAAC;QACpD,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;IACtC,CAAC;SAAM,IAAI,UAAU,EAAE,CAAC;QACtB,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAChC,CAAC;IAED,IAAI,QAAQ,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACtC,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC;IACpD,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,UAAkB;IACrD,OAAO,UAAU,GAAG,CAAC,CAAC;AACxB,CAAC"}

package/dist/agent/task-analyzer.d.ts

@@ -1,6 +1,7 @@
 import type { TaskComplexity, TaskKind, PlanStep } from "./task-plan.js";
 export interface TaskAnalysis {
     complexity: TaskComplexity;
+    /** True when multi-step durable planning is warranted (soft signal). */
     shouldPlan: boolean;
     category: TaskKind;
     goal: string;
@@ -8,11 +9,19 @@ export interface TaskAnalysis {
     needsToolPreflight: boolean;
     likelyTools: string[];
     stopWhen: string;
+    /** Soft high-level steps — never force-executed; empty for trivial asks. */
     suggestedSteps: PlanStep[];
 }
 /**
- * Complexity estimator. Uses keyword detection for pentest/security tasks
- * (always complex) and falls back to word-count heuristic otherwise.
- * No pre-baked steps — the AI decides everything dynamically.
+ * A bounded, explicit nmap operation is not a request for a full pentest.
+ * Keep it to one net.scan call unless the user also asks for assessment,
+ * reconnaissance, enumeration, exploitation, or multiple named operations.
+ */
+export declare function isNarrowExplicitNmapOperation(prompt: string): boolean;
+/**
+ * Lightweight task analysis for step budgets and soft guidance.
+ * Does not execute or force a plan — the agent still owns decisions.
  */
 export declare function analyzeTask(prompt: string): TaskAnalysis;
+/** Compact one-liner for session/system context (optional inject). */
+export declare function formatTaskAnalysisHint(analysis: TaskAnalysis): string;

package/dist/agent/task-analyzer.js

@@ -1,33 +1,273 @@
+import { createPlanStep } from "./task-plan.js";
+const PENTEST_RE = /\b(?:pentest|pen[\s-]?test|penetration|security\s*(?:test|audit|scan|assess)|vapt|csrf|xss|sqli|sql[\s-]?inject|rce|lfi|rfi|ssrf|idor|xxe|deserialization|brute[\s-]?force|enumerat|exploit|vulnerabilit|recon|bug[\s-]?bounty|ctf|capture[\s-]?the[\s-]?flag|red[\s-]?team|offensive|nmap|nikto|nuclei|ffuf|gobuster|sqlmap|hydra|metasploit)\b/i;
+const BUILD_RE = /\b(?:build|scaffold|create|implement|refactor|migrate|bootstrap|set\s*up|setup)\b/i;
+const APP_RE = /\b(?:app|application|project|website|api|service|cli|dashboard|todo|blog|frontend|backend)\b/i;
+const FIX_RE = /\b(?:fix|debug|broken|error|failing|crash|bug|regression|not\s+work|doesn'?t\s+work|typeerror|exception)\b/i;
+const RESEARCH_RE = /\b(?:what\s+is|how\s+(?:do|does|to)|explain|compare|difference|latest|current|who\s+is|when\s+did|why\s+(?:is|does))\b/i;
+const NETWORK_RE = /\b(?:scan|nmap|ping|subnet|cidr|local\s*network|hosts?|ports?|dns|whois|http|https|url|domain|ip\s*address)\b/i;
+const SHELL_RE = /\b(?:run|execute|install|upgrade|brew|apt|npm\s+i|docker|kubectl|chmod|chown|kill|restart|service)\b/i;
+const FS_RE = /\b(?:read|write|edit|file|directory|folder|path|rename|delete|list\s+files?|open\s+)\b/i;
 /**
- * Pentest / security keywords — these tasks are inherently multi-step and
- * always deserve the full step budget regardless of how terse the prompt is.
+ * A bounded, explicit nmap operation is not a request for a full pentest.
+ * Keep it to one net.scan call unless the user also asks for assessment,
+ * reconnaissance, enumeration, exploitation, or multiple named operations.
  */
-const PENTEST_RE = /\b(?:pentest|pen[\s-]?test|penetration|security\s*(?:test|audit|scan|assess)|csrf|xss|sqli|sql[\s-]?inject|rce|lfi|rfi|ssrf|idor|xxe|deserialization|brute[\s-]?force|enumerat|exploit|vulnerabilit|recon|bug[\s-]?bounty|ctf|capture[\s-]?the[\s-]?flag|red[\s-]?team|offensive|nmap|nikto|nuclei|ffuf|gobuster|sqlmap|hydra|metasploit)\b/i;
+export function isNarrowExplicitNmapOperation(prompt) {
+    const text = prompt.replace(/\s+/g, " ").trim();
+    if (!/\bnmap\b/i.test(text))
+        return false;
+    const explicitRun = /^(?:sudo\s+)?nmap(?:\s|$)/i.test(text) ||
+        /\b(?:run|execute|start|perform)\b[\s\S]{0,40}\bnmap\b/i.test(text) ||
+        /\bnmap\b[\s\S]{0,20}\bscan\b/i.test(text) ||
+        /\bscan\b[\s\S]{0,40}\bwith\s+nmap\b/i.test(text);
+    if (!explicitRun)
+        return false;
+    if (/\b(?:pentest|pen[\s-]?test|penetration|security\s+(?:audit|assessment|test)|vapt|recon(?:naissance)?|enumerat\w*|attack\s+surface|vulnerabilit\w*|exploit\w*|bug\s*bounty|red\s*team|threat\s+model|web\s+assessment)\b/i.test(text)) {
+        return false;
+    }
+    if (/\b(?:and|then|also)\b[\s\S]{0,40}\b(?:whois|dns|dig|http|crawl|fuzz|nikto|nuclei|sqlmap|gobuster|ffuf)\b/i.test(text)) {
+        return false;
+    }
+    return true;
+}
+function wordCount(prompt) {
+    return prompt.trim().split(/\s+/).filter(Boolean).length;
+}
+function classifyCategory(prompt) {
+    if (PENTEST_RE.test(prompt)) {
+        if (/\b(?:dns|whois|subdomain)\b/i.test(prompt))
+            return "dns";
+        if (/\b(?:whois|registrar|owner)\b/i.test(prompt))
+            return "whois";
+        if (/\b(?:dir|fuzz|gobuster|ffuf|content\s*discover)\b/i.test(prompt)) {
+            return "web-enum";
+        }
+        if (/\b(?:recon|enumerat|osint|fingerprint)\b/i.test(prompt)) {
+            return "pentest-recon";
+        }
+        return "pentest-recon";
+    }
+    if (/\b(?:whois)\b/i.test(prompt))
+        return "whois";
+    if (/\b(?:dns|dig|nslookup|resolve)\b/i.test(prompt))
+        return "dns";
+    if (NETWORK_RE.test(prompt) && /\b(?:discover|sweep|live\s*hosts?)\b/i.test(prompt)) {
+        return "network-discovery";
+    }
+    if (/\b(?:pkg|package|brew\s+install|apt\s+install|install\s+\w+)\b/i.test(prompt)) {
+        return "package";
+    }
+    if (BUILD_RE.test(prompt) || FIX_RE.test(prompt) || FS_RE.test(prompt)) {
+        return "filesystem";
+    }
+    if (SHELL_RE.test(prompt))
+        return "shell";
+    if (RESEARCH_RE.test(prompt) || prompt.trim().endsWith("?"))
+        return "answer";
+    return "other";
+}
+function estimateComplexity(prompt, category, words) {
+    if (PENTEST_RE.test(prompt))
+        return "complex";
+    if (BUILD_RE.test(prompt) && APP_RE.test(prompt))
+        return "complex";
+    if (/\b(?:multi-?file|across\s+files|refactor|migrate|end-?to-?end|full\s+stack)\b/i.test(prompt)) {
+        return "complex";
+    }
+    if (FIX_RE.test(prompt)) {
+        return words > 40 ? "complex" : "standard";
+    }
+    if (category === "answer" && words <= 20)
+        return "simple";
+    if (words <= 6 && !BUILD_RE.test(prompt) && !PENTEST_RE.test(prompt)) {
+        return "simple";
+    }
+    if (words <= 28 && !BUILD_RE.test(prompt))
+        return "standard";
+    return words > 40 || BUILD_RE.test(prompt) ? "complex" : "standard";
+}
+function shouldPlanFor(prompt, complexity, category) {
+    if (complexity === "simple")
+        return false;
+    if (category === "answer")
+        return false;
+    if (PENTEST_RE.test(prompt))
+        return true;
+    if (BUILD_RE.test(prompt) && APP_RE.test(prompt))
+        return true;
+    if (/\b(?:implement|refactor|migrate|multi-?step|several\s+files|end-?to-?end)\b/i.test(prompt)) {
+        return true;
+    }
+    return complexity === "complex" && wordCount(prompt) > 25;
+}
+function inferLikelyTools(prompt, category) {
+    const tools = new Set();
+    if (category === "answer" || RESEARCH_RE.test(prompt)) {
+        tools.add("web.search");
+        tools.add("web.fetch");
+    }
+    if (category === "filesystem" || BUILD_RE.test(prompt) || FIX_RE.test(prompt)) {
+        tools.add("fs.list");
+        tools.add("fs.read");
+        tools.add("fs.write");
+        tools.add("fs.edit");
+        tools.add("shell.exec");
+    }
+    if (BUILD_RE.test(prompt) && APP_RE.test(prompt)) {
+        tools.add("plan.create");
+        tools.add("task.update");
+        tools.add("shell.start");
+    }
+    if (PENTEST_RE.test(prompt) || category.startsWith("pentest") || category === "web-enum") {
+        tools.add("dns.lookup");
+        tools.add("whois.lookup");
+        tools.add("http.fetch");
+        tools.add("net.scan");
+        tools.add("tool.check");
+        tools.add("plan.create");
+    }
+    if (category === "network-discovery") {
+        tools.add("net.context");
+        tools.add("net.pingSweep");
+        tools.add("net.scan");
+    }
+    if (category === "dns")
+        tools.add("dns.lookup");
+    if (category === "whois")
+        tools.add("whois.lookup");
+    if (category === "package")
+        tools.add("pkg.install");
+    if (SHELL_RE.test(prompt))
+        tools.add("shell.exec");
+    if (/\b(?:pdf)\b/i.test(prompt))
+        tools.add("pdf.read");
+    if (/\b(?:ocr|image|screenshot)\b/i.test(prompt))
+        tools.add("image.ocr");
+    return [...tools];
+}
+function softSteps(prompt, category) {
+    if (PENTEST_RE.test(prompt)) {
+        return [
+            createPlanStep("Recon and fingerprint stack", "pentest-recon", {
+                toolHint: "dns.lookup / http.fetch / net.scan",
+                successCriteria: "real findings from tool output",
+            }),
+            createPlanStep("Plan from findings", "other", {
+                toolHint: "plan.create",
+                successCriteria: "durable plan with verifiable tasks",
+            }),
+            createPlanStep("Test high-value vectors", "web-enum", {
+                successCriteria: "validated evidence or clean negatives",
+            }),
+            createPlanStep("Report with evidence", "other", {
+                successCriteria: "findings with repro + severity",
+            }),
+        ];
+    }
+    if (BUILD_RE.test(prompt) && APP_RE.test(prompt)) {
+        return [
+            createPlanStep("Explore destination and stack", "filesystem", {
+                toolHint: "fs.list / fs.read",
+            }),
+            createPlanStep("Plan multi-step build", "other", {
+                toolHint: "plan.create",
+            }),
+            createPlanStep("Implement requested feature", "filesystem", {
+                toolHint: "fs.write / fs.edit",
+            }),
+            createPlanStep("Verify (build and/or live server)", "shell", {
+                toolHint: "shell.start / shell.exec",
+            }),
+        ];
+    }
+    if (FIX_RE.test(prompt)) {
+        return [
+            createPlanStep("Reproduce and capture error", "shell"),
+            createPlanStep("Localize root cause", "filesystem", {
+                toolHint: "fs.read / fs.search",
+            }),
+            createPlanStep("Apply minimal fix", "filesystem", {
+                toolHint: "fs.edit",
+            }),
+            createPlanStep("Re-run failing check", "shell"),
+        ];
+    }
+    if (category === "network-discovery") {
+        return [
+            createPlanStep("Local network context", "network-discovery", {
+                toolHint: "net.context",
+            }),
+            createPlanStep("Discover live hosts", "network-discovery", {
+                toolHint: "net.pingSweep",
+            }),
+            createPlanStep("Summarize findings", "other"),
+        ];
+    }
+    return [];
+}
+function stopWhenFor(prompt, category) {
+    if (PENTEST_RE.test(prompt)) {
+        return "Findings report with evidence; no local dev server; scope respected.";
+    }
+    if (BUILD_RE.test(prompt) && APP_RE.test(prompt)) {
+        return "Requested feature works; verified via build and/or live localhost probe; server left running when applicable.";
+    }
+    if (FIX_RE.test(prompt)) {
+        return "Original failing check now passes; root cause addressed.";
+    }
+    if (category === "answer") {
+        return "Accurate answer grounded in tools when facts are volatile.";
+    }
+    return "User request completed with tool-backed evidence.";
+}
 /**
- * Complexity estimator. Uses keyword detection for pentest/security tasks
- * (always complex) and falls back to word-count heuristic otherwise.
- * No pre-baked steps — the AI decides everything dynamically.
+ * Lightweight task analysis for step budgets and soft guidance.
+ * Does not execute or force a plan — the agent still owns decisions.
  */
 export function analyzeTask(prompt) {
-    const words = prompt.trim().split(/\s+/).length;
-    // Pentest tasks are inherently multi-step — always complex.
-    const complexity = PENTEST_RE.test(prompt)
-        ? "complex"
-        : words <= 5
-            ? "simple"
-            : words <= 30
-                ? "standard"
-                : "complex";
+    const text = prompt.replace(/\s+/g, " ").trim();
+    if (isNarrowExplicitNmapOperation(text)) {
+        return {
+            complexity: "standard",
+            shouldPlan: false,
+            category: "pentest-recon",
+            goal: text.slice(0, 100),
+            needsNetworkContext: false,
+            needsToolPreflight: true,
+            likelyTools: ["net.scan"],
+            stopWhen: "The requested nmap operation has started or completed and its canonical job ID, status, and output are reported; do not broaden scope.",
+            suggestedSteps: [],
+        };
+    }
+    const words = wordCount(text);
+    const category = classifyCategory(text);
+    const complexity = estimateComplexity(text, category, words);
+    const shouldPlan = shouldPlanFor(text, complexity, category);
+    const likelyTools = inferLikelyTools(text, category);
+    const suggestedSteps = complexity === "simple" || category === "answer" ? [] : softSteps(text, category);
     return {
         complexity,
-        shouldPlan: false,
-        category: "other",
-        goal: prompt.slice(0, 100),
-        needsNetworkContext: false,
-        needsToolPreflight: false,
-        likelyTools: [],
-        stopWhen: "",
-        suggestedSteps: [],
+        shouldPlan,
+        category,
+        goal: text.slice(0, 100),
+        needsNetworkContext: category === "network-discovery" ||
+            /\b(?:local\s*network|subnet|cidr|lan|my\s+network)\b/i.test(text),
+        needsToolPreflight: PENTEST_RE.test(text) ||
+            /\b(?:nmap|ffuf|gobuster|sqlmap|nuclei|hydra|nikto)\b/i.test(text),
+        likelyTools,
+        stopWhen: stopWhenFor(text, category),
+        suggestedSteps,
     };
 }
+/** Compact one-liner for session/system context (optional inject). */
+export function formatTaskAnalysisHint(analysis) {
+    const tools = analysis.likelyTools.length > 0
+        ? analysis.likelyTools.slice(0, 8).join(", ")
+        : "—";
+    const tracking = analysis.shouldPlan
+        ? "durable task tracking strongly recommended: create one concise plan before mutation, then keep task.update states aligned with real work"
+        : "direct execution ok";
+    return (`TASK ANALYSIS: category=${analysis.category} complexity=${analysis.complexity} (${tracking}). ` +
+        `Likely tools: ${tools}. Done when: ${analysis.stopWhen}`);
+}
 //# sourceMappingURL=task-analyzer.js.map