@nac3/forge-cli 0.2.0-alpha.9 → 1.0.0

package/dist/test_agent/gemini_user.d.ts

@@ -0,0 +1,9 @@
+import type { SyntheticUserClient } from './runner.js';
+export interface GeminiSyntheticUserOpts {
+    apiKey: string;
+    /** Default 'gemini-2.5-flash-lite' -- cheapest valid model for
+     *  multi-turn persona simulation. */
+    model?: string;
+}
+export declare function makeGeminiSyntheticUser(opts: GeminiSyntheticUserOpts): SyntheticUserClient;
+//# sourceMappingURL=gemini_user.d.ts.map

package/dist/test_agent/gemini_user.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"gemini_user.d.ts","sourceRoot":"","sources":["../../src/test_agent/gemini_user.ts"],"names":[],"mappings":"AAiBA,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAEvD,MAAM,WAAW,uBAAuB;IACtC,MAAM,EAAE,MAAM,CAAC;IACf;yCACqC;IACrC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AA4BD,wBAAgB,uBAAuB,CAAC,IAAI,EAAE,uBAAuB,GAAG,mBAAmB,CA8B1F"}

package/dist/test_agent/gemini_user.js

@@ -0,0 +1,58 @@
+import { GeminiClient } from '../llm/gemini_client.js';
+function buildSystemPrompt(profile, turnIndex) {
+    const lines = [
+        'You are a SYNTHETIC USER in a test harness. You are NOT the assistant.',
+        'You drive a multi-turn conversation against a target Forge chat server.',
+        'On each turn you produce the SINGLE next user message and nothing else.',
+        '',
+        '--- Persona ---',
+        profile.persona,
+        '',
+        '--- Goal ---',
+        profile.goal,
+        '',
+        '--- Constraints ---',
+        ...profile.constraints.map((c) => '- ' + c),
+        '',
+        '--- Output rules ---',
+        'Output ONLY the next user message in plain text. No JSON, no markdown',
+        'fences, no role tags, no commentary. Reply with what a real user would',
+        'type. Keep it conversational and natural. Length 1-4 sentences typical;',
+        'longer only when the persona would write a longer message.',
+        '',
+        'This is turn ' + turnIndex + '.',
+    ];
+    return lines.join('\n');
+}
+export function makeGeminiSyntheticUser(opts) {
+    const gemini = new GeminiClient({
+        apiKey: opts.apiKey,
+        defaultModel: opts.model ?? 'gemini-2.5-flash-lite',
+    });
+    return {
+        async nextTurn({ profile, history, turn_index }) {
+            const messages = history.map((h) => ({
+                /* Flip the perspective: the conversation history is from the
+                 * REAL chat; for the synthetic user, what the real chat said
+                 * (assistant) is what the synthetic user is RESPONDING to,
+                 * and what the synthetic user previously said (user) is its
+                 * own prior turn. */
+                role: (h.role === 'user' ? 'assistant' : 'user'),
+                content: h.content,
+            }));
+            /* If this is the first turn, kick the conversation off with a
+             * single system-only call. Gemini complains when there are no
+             * user messages, so emit a synthetic seed. */
+            if (messages.length === 0) {
+                messages.push({ role: 'user', content: '(begin the conversation in line with your persona + goal)' });
+            }
+            const r = await gemini.chat({
+                messages,
+                system: buildSystemPrompt(profile, turn_index),
+                maxTokens: 400,
+            });
+            return { text: (r.text ?? '').trim() };
+        },
+    };
+}
+//# sourceMappingURL=gemini_user.js.map

package/dist/test_agent/gemini_user.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gemini_user.js","sourceRoot":"","sources":["../../src/test_agent/gemini_user.ts"],"names":[],"mappings":"AAeA,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAWvD,SAAS,iBAAiB,CAAC,OAAgB,EAAE,SAAiB;IAC5D,MAAM,KAAK,GAAG;QACZ,wEAAwE;QACxE,yEAAyE;QACzE,yEAAyE;QACzE,EAAE;QACF,iBAAiB;QACjB,OAAO,CAAC,OAAO;QACf,EAAE;QACF,cAAc;QACd,OAAO,CAAC,IAAI;QACZ,EAAE;QACF,qBAAqB;QACrB,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,GAAG,CAAC,CAAC;QAC3C,EAAE;QACF,sBAAsB;QACtB,uEAAuE;QACvE,wEAAwE;QACxE,yEAAyE;QACzE,4DAA4D;QAC5D,EAAE;QACF,eAAe,GAAG,SAAS,GAAG,GAAG;KAClC,CAAC;IACF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,IAA6B;IACnE,MAAM,MAAM,GAAG,IAAI,YAAY,CAAC;QAC9B,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,YAAY,EAAE,IAAI,CAAC,KAAK,IAAI,uBAAuB;KACpD,CAAC,CAAC;IACH,OAAO;QACL,KAAK,CAAC,QAAQ,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE;YAC7C,MAAM,QAAQ,GAAkB,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAClD;;;;qCAIqB;gBACrB,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAyB;gBACxE,OAAO,EAAE,CAAC,CAAC,OAAO;aACnB,CAAC,CAAC,CAAC;YACJ;;0DAE8C;YAC9C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC1B,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,2DAA2D,EAAE,CAAC,CAAC;YACxG,CAAC;YACD,MAAM,CAAC,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC;gBAC1B,QAAQ;gBACR,MAAM,EAAE,iBAAiB,CAAC,OAAO,EAAE,UAAU,CAAC;gBAC9C,SAAS,EAAE,GAAG;aACf,CAAC,CAAC;YACH,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QACzC,CAAC;KACF,CAAC;AACJ,CAAC"}

package/dist/test_agent/profiles.d.ts

@@ -0,0 +1,23 @@
+/**
+ * Agente de Test profile library (PND-024 Etapa 1).
+ *
+ * Three baseline profiles cover the common Forge entry points:
+ *
+ *  - greenfield-spec-writer: user opens Forge against an empty
+ *    directory with a partial spec idea. Expects Forge to ingest
+ *    the spec and propose next steps.
+ *  - brownfield-bug-fixer: user opens Forge against an existing
+ *    repo to fix a small bug. Expects Forge to read manifest, find
+ *    relevant files, propose changes.
+ *  - casual-out-of-scope: user opens Forge and asks for a song /
+ *    poem / chat. Expects Forge to skip the warmup (PND-022) and
+ *    respond conversationally instead of asking target/mode.
+ *
+ * Each profile is an immutable export -- the runner reads them via
+ * getProfile(slug) and never mutates. Adding profiles is a single
+ * declaration here; the CLI surface picks them up automatically.
+ */
+import type { Profile } from './types.js';
+export declare function listProfiles(): readonly Profile[];
+export declare function getProfile(slug: string): Profile | null;
+//# sourceMappingURL=profiles.d.ts.map

package/dist/test_agent/profiles.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"profiles.d.ts","sourceRoot":"","sources":["../../src/test_agent/profiles.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AACH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AA2H1C,wBAAgB,YAAY,IAAI,SAAS,OAAO,EAAE,CAEjD;AAED,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI,CAEvD"}

package/dist/test_agent/profiles.js

@@ -0,0 +1,117 @@
+const GREENFIELD_SPEC_WRITER = {
+    slug: 'greenfield-spec-writer',
+    name: 'Greenfield spec writer',
+    persona: 'You are a startup founder with a half-baked product idea. You '
+        + 'just installed Yujin Forge to scaffold a new project. You have '
+        + 'a rough sense of what you want but you write in plain Spanish, '
+        + 'not technical jargon.',
+    goal: 'Get Forge to ingest your spec idea and propose a concrete '
+        + 'next step (e.g. specific files to create, a target stack to use, '
+        + 'or a clarification question that nails the ambiguity).',
+    constraints: [
+        'Reply in plain Spanish.',
+        'Do not paste code -- you are not a developer.',
+        'When Forge asks the target/mode warmup, answer briefly and move on.',
+        'If Forge proposes a concrete next step (file scaffold, stack choice, '
+            + 'clarifying question), the goal is met -- thank them.',
+    ],
+    max_turns: 6,
+    assertions: [
+        {
+            kind: 'reply_not_contains',
+            label: 'Forge does not paste raw code into the first reply',
+            payload: { text: 'function ', case_insensitive: true },
+        },
+        {
+            kind: 'turn_count_at_most',
+            label: 'Forge proposes a next step within 4 turns',
+            payload: { max: 4 },
+        },
+    ],
+    success_when: {
+        kind: 'reply_matches_regex',
+        label: 'Forge proposes a scaffold / stack / clarifying question',
+        payload: {
+            regex: '(scaffold|estructura inicial|stack|template|que tipo de|precisas|necesitas|cu[aá]l|escribir(e|ia) el|propongo)',
+            flags: 'i',
+        },
+    },
+};
+const BROWNFIELD_BUG_FIXER = {
+    slug: 'brownfield-bug-fixer',
+    name: 'Brownfield bug fixer',
+    persona: 'You are an experienced developer who just cloned an existing '
+        + 'project and opened Forge in its directory. You have a bug '
+        + 'report and want Forge to help you locate the relevant files.',
+    goal: 'Have Forge invoke a manifest-read / file-read / fs-find verb '
+        + 'so you can confirm it is grounding itself in the actual code '
+        + 'instead of inventing context.',
+    constraints: [
+        'Write in technical English.',
+        'When Forge asks clarification, give a concrete file path + symptom.',
+        'Do not accept hand-waved answers -- ask for the actual file.',
+    ],
+    max_turns: 6,
+    assertions: [
+        {
+            kind: 'turn_count_at_most',
+            label: 'Forge grounds itself in actual code within 3 turns',
+            payload: { max: 3 },
+            blocker: false,
+        },
+    ],
+    success_when: {
+        kind: 'tool_invoked',
+        label: 'Forge invokes a file-read or manifest verb',
+        payload: { verb_id: 'yujin.fs.read-file' },
+    },
+};
+const CASUAL_OUT_OF_SCOPE = {
+    slug: 'casual-out-of-scope',
+    name: 'Casual / out-of-scope conversation',
+    persona: 'You are someone who opened Forge by accident and want to chat '
+        + 'casually -- ask for a song, a poem, or just say hi. You have '
+        + 'no plans to build software in this session.',
+    goal: 'See whether Forge respects out-of-scope intent (PND-022) or '
+        + 'whether it intrusively asks the target/mode warmup despite '
+        + 'your clearly non-dev prompt.',
+    constraints: [
+        'Open with a clearly non-dev request (song, poem, greeting).',
+        'Do NOT mention any dev keywords (deploy, react, code, etc).',
+        'If Forge asks the warmup anyway, push back politely and continue '
+            + 'with your original request.',
+    ],
+    max_turns: 4,
+    assertions: [
+        {
+            kind: 'reply_not_contains',
+            label: 'Forge does not ask the target/mode warmup',
+            payload: { text: 'responsive', case_insensitive: true },
+            blocker: false,
+        },
+        {
+            kind: 'reply_not_contains',
+            label: 'Forge does not ask didactico/tecnico mode warmup',
+            payload: { text: 'didactico', case_insensitive: true },
+            blocker: false,
+        },
+    ],
+    success_when: {
+        kind: 'reply_matches_regex',
+        label: 'Forge engages with the non-dev intent at least once',
+        payload: { regex: '(canci[oó]n|cancion|poema|hola|saludo|m[uú]sica|verso|haiku)', flags: 'i' },
+    },
+};
+const PROFILES = [
+    GREENFIELD_SPEC_WRITER,
+    BROWNFIELD_BUG_FIXER,
+    CASUAL_OUT_OF_SCOPE,
+];
+const PROFILE_BY_SLUG = new Map(PROFILES.map((p) => [p.slug, p]));
+export function listProfiles() {
+    return PROFILES;
+}
+export function getProfile(slug) {
+    return PROFILE_BY_SLUG.get(slug) ?? null;
+}
+//# sourceMappingURL=profiles.js.map

package/dist/test_agent/profiles.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"profiles.js","sourceRoot":"","sources":["../../src/test_agent/profiles.ts"],"names":[],"mappings":"AAqBA,MAAM,sBAAsB,GAAY;IACtC,IAAI,EAAE,wBAAwB;IAC9B,IAAI,EAAE,wBAAwB;IAC9B,OAAO,EACL,gEAAgE;UAC9D,iEAAiE;UACjE,iEAAiE;UACjE,uBAAuB;IAC3B,IAAI,EACF,4DAA4D;UAC1D,mEAAmE;UACnE,wDAAwD;IAC5D,WAAW,EAAE;QACX,yBAAyB;QACzB,+CAA+C;QAC/C,qEAAqE;QACrE,uEAAuE;cACnE,sDAAsD;KAC3D;IACD,SAAS,EAAE,CAAC;IACZ,UAAU,EAAE;QACV;YACE,IAAI,EAAE,oBAAoB;YAC1B,KAAK,EAAE,oDAAoD;YAC3D,OAAO,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,gBAAgB,EAAE,IAAI,EAAE;SACvD;QACD;YACE,IAAI,EAAE,oBAAoB;YAC1B,KAAK,EAAE,2CAA2C;YAClD,OAAO,EAAE,EAAE,GAAG,EAAE,CAAC,EAAE;SACpB;KACF;IACD,YAAY,EAAE;QACZ,IAAI,EAAE,qBAAqB;QAC3B,KAAK,EAAE,yDAAyD;QAChE,OAAO,EAAE;YACP,KAAK,EAAE,gHAAgH;YACvH,KAAK,EAAE,GAAG;SACX;KACF;CACF,CAAC;AAEF,MAAM,oBAAoB,GAAY;IACpC,IAAI,EAAE,sBAAsB;IAC5B,IAAI,EAAE,sBAAsB;IAC5B,OAAO,EACL,+DAA+D;UAC7D,4DAA4D;UAC5D,8DAA8D;IAClE,IAAI,EACF,+DAA+D;UAC7D,+DAA+D;UAC/D,+BAA+B;IACnC,WAAW,EAAE;QACX,6BAA6B;QAC7B,qEAAqE;QACrE,8DAA8D;KAC/D;IACD,SAAS,EAAE,CAAC;IACZ,UAAU,EAAE;QACV;YACE,IAAI,EAAE,oBAAoB;YAC1B,KAAK,EAAE,oDAAoD;YAC3D,OAAO,EAAE,EAAE,GAAG,EAAE,CAAC,EAAE;YACnB,OAAO,EAAE,KAAK;SACf;KACF;IACD,YAAY,EAAE;QACZ,IAAI,EAAE,cAAc;QACpB,KAAK,EAAE,4CAA4C;QACnD,OAAO,EAAE,EAAE,OAAO,EAAE,oBAAoB,EAAE;KAC3C;CACF,CAAC;AAEF,MAAM,mBAAmB,GAAY;IACnC,IAAI,EAAE,qBAAqB;IAC3B,IAAI,EAAE,oCAAoC;IAC1C,OAAO,EACL,gEAAgE;UAC9D,+DAA+D;UAC/D,6CAA6C;IACjD,IAAI,EACF,8DAA8D;UAC5D,6DAA6D;UAC7D,8BAA8B;IAClC,WAAW,EAAE;QACX,6DAA6D;QAC7D,6DAA6D;QAC7D,mEAAmE;cAC/D,6BAA6B;KAClC;IACD,SAAS,EAAE,CAAC;IACZ,UAAU,EAAE;QACV;YACE,IAAI,EAAE,oBAAoB;YAC1B,KAAK,EAAE,2CAA2C;YAClD,OAAO,EAAE,EAAE,IAAI,EAAE,YAAY,EAAE,gBAAgB,EAAE,IAAI,EAAE;YACvD,OAAO,EAAE,KAAK;SACf;QACD;YACE,IAAI,EAAE,oBAAoB;YAC1B,KAAK,EAAE,kDAAkD;YACzD,OAAO,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,gBAAgB,EAAE,IAAI,EAAE;YACtD,OAAO,EAAE,KAAK;SACf;KACF;IACD,YAAY,EAAE;QACZ,IAAI,EAAE,qBAAqB;QAC3B,KAAK,EAAE,qDAAqD;QAC5D,OAAO,EAAE,EAAE,KAAK,EAAE,8DAA8D,EAAE,KAAK,EAAE,GAAG,EAAE;KAC/F;CACF,CAAC;AAEF,MAAM,QAAQ,GAAc;IAC1B,sBAAsB;IACtB,oBAAoB;IACpB,mBAAmB;CACpB,CAAC;AAEF,MAAM,eAAe,GAAyB,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AAExF,MAAM,UAAU,YAAY;IAC1B,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAY;IACrC,OAAO,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC;AAC3C,CAAC"}

package/dist/test_agent/runner.d.ts

@@ -0,0 +1,56 @@
+/**
+ * Agente de Test multi-turn runner (PND-024 Etapa 1).
+ *
+ * Drives a synthetic conversation between a Profile-driven user
+ * (Gemini Flash Lite by default, anything ChatLLMClient-compatible
+ * via dependency injection) and a target Forge chat endpoint
+ * (any HTTP-reachable Forge `/api/chat`).
+ *
+ * The runner is profile-agnostic -- it loads the assertion + success
+ * criterion logic from src/test_agent/types.ts and walks the profile
+ * through max_turns iterations, stopping early when success_when
+ * matches or a blocker assertion fails.
+ *
+ * ASCII-only.
+ */
+import type { Profile, ToolRoundLite, ProfileRunReport } from './types.js';
+export interface SyntheticUserClient {
+    /** Compose the next user message given the conversation so far +
+     *  the profile guidance. Implementations call Gemini / Anthropic /
+     *  whatever to actually generate the text. */
+    nextTurn(opts: {
+        profile: Profile;
+        history: Array<{
+            role: 'user' | 'assistant';
+            content: string;
+        }>;
+        turn_index: number;
+    }): Promise<{
+        text: string;
+    }>;
+}
+export interface TargetChatClient {
+    /** POST to the target Forge's /api/chat endpoint (or any
+     *  surface that takes the conversation history + returns a
+     *  reply + tool rounds). */
+    chat(opts: {
+        messages: Array<{
+            role: 'user' | 'assistant';
+            content: string;
+        }>;
+    }): Promise<{
+        reply: string;
+        tool_rounds: ToolRoundLite[];
+    }>;
+}
+export interface RunProfileOpts {
+    profile: Profile;
+    synthetic: SyntheticUserClient;
+    target: TargetChatClient;
+    /** Optional cap override on profile.max_turns. */
+    max_turns?: number;
+    /** Optional clock injection (tests). Defaults to Date.now. */
+    now?: () => number;
+}
+export declare function runProfile(opts: RunProfileOpts): Promise<ProfileRunReport>;
+//# sourceMappingURL=runner.d.ts.map

package/dist/test_agent/runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/test_agent/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AACH,OAAO,KAAK,EACV,OAAO,EAA0C,aAAa,EAC9D,gBAAgB,EACjB,MAAM,YAAY,CAAC;AAEpB,MAAM,WAAW,mBAAmB;IAClC;;kDAE8C;IAC9C,QAAQ,CAAC,IAAI,EAAE;QACb,OAAO,EAAM,OAAO,CAAC;QACrB,OAAO,EAAM,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC;YAAC,OAAO,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QACpE,UAAU,EAAG,MAAM,CAAC;KACrB,GAAG,OAAO,CAAC;QAAE,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAC/B;;gCAE4B;IAC5B,IAAI,CAAC,IAAI,EAAE;QAAE,QAAQ,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC;YAAC,OAAO,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;KAAE,GAAG,OAAO,CAAC;QACxF,KAAK,EAAQ,MAAM,CAAC;QACpB,WAAW,EAAE,aAAa,EAAE,CAAC;KAC9B,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAS,OAAO,CAAC;IACxB,SAAS,EAAO,mBAAmB,CAAC;IACpC,MAAM,EAAU,gBAAgB,CAAC;IACjC,kDAAkD;IAClD,SAAS,CAAC,EAAM,MAAM,CAAC;IACvB,8DAA8D;IAC9D,GAAG,CAAC,EAAY,MAAM,MAAM,CAAC;CAC9B;AA8DD,wBAAsB,UAAU,CAAC,IAAI,EAAE,cAAc,GAAG,OAAO,CAAC,gBAAgB,CAAC,CA8EhF"}

package/dist/test_agent/runner.js

@@ -0,0 +1,134 @@
+const DEFAULT_MAX_TURNS = 8;
+function evaluateAssertion(a, ctx, index) {
+    switch (a.kind) {
+        case 'reply_contains': {
+            const text = String(a.payload['text'] ?? '');
+            const ci = a.payload['case_insensitive'] === true;
+            const needle = ci ? text.toLowerCase() : text;
+            const hay = ci ? ctx.reply.toLowerCase() : ctx.reply;
+            const passed = hay.includes(needle);
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'reply missing "' + text + '"' }) };
+        }
+        case 'reply_not_contains': {
+            const text = String(a.payload['text'] ?? '');
+            const ci = a.payload['case_insensitive'] === true;
+            const needle = ci ? text.toLowerCase() : text;
+            const hay = ci ? ctx.reply.toLowerCase() : ctx.reply;
+            const passed = !hay.includes(needle);
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'reply contains "' + text + '"' }) };
+        }
+        case 'reply_matches_regex': {
+            const regex = String(a.payload['regex'] ?? '');
+            const flags = String(a.payload['flags'] ?? '');
+            let re;
+            try {
+                re = new RegExp(regex, flags);
+            }
+            catch {
+                return { index, kind: a.kind, label: a.label, passed: false,
+                    detail: 'invalid regex ' + JSON.stringify(regex) };
+            }
+            const passed = re.test(ctx.reply);
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'reply did not match ' + JSON.stringify(regex) }) };
+        }
+        case 'tool_invoked': {
+            const verb = String(a.payload['verb_id'] ?? '');
+            const passed = ctx.tool_rounds.some((r) => r.verb_id === verb || r.tool === verb);
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'verb ' + verb + ' not invoked' }) };
+        }
+        case 'tool_not_invoked': {
+            const verb = String(a.payload['verb_id'] ?? '');
+            const passed = !ctx.tool_rounds.some((r) => r.verb_id === verb || r.tool === verb);
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'verb ' + verb + ' invoked' }) };
+        }
+        case 'turn_count_at_most': {
+            const max = Number(a.payload['max'] ?? Number.MAX_SAFE_INTEGER);
+            const passed = ctx.turn <= max;
+            return { index, kind: a.kind, label: a.label, passed,
+                ...(passed ? {} : { detail: 'turn ' + ctx.turn + ' > max ' + max }) };
+        }
+    }
+}
+export async function runProfile(opts) {
+    const profile = opts.profile;
+    const max = opts.max_turns ?? profile.max_turns ?? DEFAULT_MAX_TURNS;
+    const now = opts.now ?? Date.now;
+    const startedMs = now();
+    const startedAt = new Date(startedMs).toISOString();
+    const history = [];
+    const turns = [];
+    let successMet = false;
+    let stoppedByBlocker = false;
+    for (let i = 1; i <= max; i++) {
+        const turnStart = now();
+        const userTurn = await opts.synthetic.nextTurn({
+            profile, history: history.slice(), turn_index: i,
+        });
+        const userMessage = userTurn.text.trim();
+        history.push({ role: 'user', content: userMessage });
+        const targetReply = await opts.target.chat({ messages: history.slice() });
+        history.push({ role: 'assistant', content: targetReply.reply });
+        const evalCtx = {
+            reply: targetReply.reply,
+            tool_rounds: targetReply.tool_rounds,
+            turn: i,
+        };
+        const assertion_results = profile.assertions.map((a, idx) => evaluateAssertion(a, evalCtx, idx));
+        let recordedSuccess = false;
+        if (profile.success_when) {
+            const r = evaluateAssertion(profile.success_when, evalCtx, -1);
+            assertion_results.push(r);
+            if (r.passed) {
+                successMet = true;
+                recordedSuccess = true;
+            }
+        }
+        void recordedSuccess;
+        const turnEnd = now();
+        turns.push({
+            turn: i,
+            user_message: userMessage,
+            assistant_reply: targetReply.reply,
+            tool_rounds: targetReply.tool_rounds,
+            latency_ms: turnEnd - turnStart,
+            assertion_results,
+        });
+        if (successMet)
+            break;
+        const blockerFailed = assertion_results.some((r) => !r.passed && profile.assertions[r.index]?.blocker === true);
+        if (blockerFailed) {
+            stoppedByBlocker = true;
+            break;
+        }
+    }
+    const endedMs = now();
+    const allResults = turns.flatMap((t) => t.assertion_results);
+    const passes = allResults.filter((r) => r.passed).length;
+    const failures = allResults.filter((r) => !r.passed).length;
+    let status;
+    if (successMet)
+        status = 'passed';
+    else if (stoppedByBlocker || failures > 0)
+        status = 'failed';
+    else
+        status = 'incomplete';
+    return {
+        profile_slug: profile.slug,
+        profile_name: profile.name,
+        status,
+        turns_used: turns.length,
+        success_criterion_met: successMet,
+        failures,
+        passes,
+        duration_ms: endedMs - startedMs,
+        turns,
+        started_at: startedAt,
+        ended_at: new Date(endedMs).toISOString(),
+    };
+}
+//# sourceMappingURL=runner.js.map

package/dist/test_agent/runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/test_agent/runner.ts"],"names":[],"mappings":"AAmDA,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B,SAAS,iBAAiB,CACxB,CAAY,EACZ,GAAkE,EAClE,KAAa;IAEb,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;QACf,KAAK,gBAAgB,CAAC,CAAC,CAAC;YACtB,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;YAC7C,MAAM,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,kBAAkB,CAAC,KAAK,IAAI,CAAC;YAClD,MAAM,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;YAC9C,MAAM,GAAG,GAAG,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC;YACrD,MAAM,MAAM,GAAG,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YACpC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,iBAAiB,GAAG,IAAI,GAAG,GAAG,EAAE,CAAC,EAAE,CAAC;QAC3E,CAAC;QACD,KAAK,oBAAoB,CAAC,CAAC,CAAC;YAC1B,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;YAC7C,MAAM,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,kBAAkB,CAAC,KAAK,IAAI,CAAC;YAClD,MAAM,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;YAC9C,MAAM,GAAG,GAAG,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC;YACrD,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YACrC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,kBAAkB,GAAG,IAAI,GAAG,GAAG,EAAE,CAAC,EAAE,CAAC;QAC5E,CAAC;QACD,KAAK,qBAAqB,CAAC,CAAC,CAAC;YAC3B,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;YAC/C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;YAC/C,IAAI,EAAU,CAAC;YACf,IAAI,CAAC;gBAAC,EAAE,GAAG,IAAI,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAAC,CAAC;YACtC,MAAM,CAAC;gBAAC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK;oBAClD,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC;YAAC,CAAC;YACtE,MAAM,MAAM,GAAG,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAClC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,sBAAsB,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,CAAC;QAC3F,CAAC;QACD,KAAK,cAAc,CAAC,CAAC,CAAC;YACpB,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;YAChD,MAAM,MAAM,GAAG,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CACxC,CAAC,CAAC,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;YACzC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,GAAG,IAAI,GAAG,cAAc,EAAE,CAAC,EAAE,CAAC;QAC5E,CAAC;QACD,KAAK,kBAAkB,CAAC,CAAC,CAAC;YACxB,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;YAChD,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CACzC,CAAC,CAAC,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;YACzC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,GAAG,IAAI,GAAG,UAAU,EAAE,CAAC,EAAE,CAAC;QACxE,CAAC;QACD,KAAK,oBAAoB,CAAC,CAAC,CAAC;YAC1B,MAAM,GAAG,GAAG,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,gBAAgB,CAAC,CAAC;YAChE,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC;YAC/B,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,MAAM;gBAC3C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,GAAG,GAAG,CAAC,IAAI,GAAG,SAAS,GAAG,GAAG,EAAE,CAAC,EAAE,CAAC;QACjF,CAAC;IACH,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,IAAoB;IACnD,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;IAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,IAAI,OAAO,CAAC,SAAS,IAAI,iBAAiB,CAAC;IACrE,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC;IACjC,MAAM,SAAS,GAAG,GAAG,EAAE,CAAC;IACxB,MAAM,SAAS,GAAG,IAAI,IAAI,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;IAEpD,MAAM,OAAO,GAA2D,EAAE,CAAC;IAC3E,MAAM,KAAK,GAAiB,EAAE,CAAC;IAC/B,IAAI,UAAU,GAAG,KAAK,CAAC;IACvB,IAAI,gBAAgB,GAAG,KAAK,CAAC;IAE7B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9B,MAAM,SAAS,GAAG,GAAG,EAAE,CAAC;QACxB,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC;YAC7C,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,EAAE,EAAE,UAAU,EAAE,CAAC;SACjD,CAAC,CAAC;QACH,MAAM,WAAW,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;QACzC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;QACrD,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC1E,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC;QAChE,MAAM,OAAO,GAAG;YACd,KAAK,EAAE,WAAW,CAAC,KAAK;YACxB,WAAW,EAAE,WAAW,CAAC,WAAW;YACpC,IAAI,EAAE,CAAC;SACR,CAAC;QACF,MAAM,iBAAiB,GAAsB,OAAO,CAAC,UAAU,CAAC,GAAG,CACjE,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,iBAAiB,CAAC,CAAC,EAAE,OAAO,EAAE,GAAG,CAAC,CAC/C,CAAC;QAEF,IAAI,eAAe,GAAG,KAAK,CAAC;QAC5B,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;YACzB,MAAM,CAAC,GAAG,iBAAiB,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;YAC/D,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YAC1B,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC;gBAAC,UAAU,GAAG,IAAI,CAAC;gBAAC,eAAe,GAAG,IAAI,CAAC;YAAC,CAAC;QAC9D,CAAC;QACD,KAAK,eAAe,CAAC;QAErB,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,KAAK,CAAC,IAAI,CAAC;YACT,IAAI,EAAE,CAAC;YACP,YAAY,EAAE,WAAW;YACzB,eAAe,EAAE,WAAW,CAAC,KAAK;YAClC,WAAW,EAAE,WAAW,CAAC,WAAW;YACpC,UAAU,EAAE,OAAO,GAAG,SAAS;YAC/B,iBAAiB;SAClB,CAAC,CAAC;QAEH,IAAI,UAAU;YAAE,MAAM;QACtB,MAAM,aAAa,GAAG,iBAAiB,CAAC,IAAI,CAC1C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,OAAO,KAAK,IAAI,CAClE,CAAC;QACF,IAAI,aAAa,EAAE,CAAC;YAAC,gBAAgB,GAAG,IAAI,CAAC;YAAC,MAAM;QAAC,CAAC;IACxD,CAAC;IAED,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;IACtB,MAAM,UAAU,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC;IAC7D,MAAM,MAAM,GAAK,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAC3D,MAAM,QAAQ,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAE5D,IAAI,MAA0C,CAAC;IAC/C,IAAI,UAAU;QAAE,MAAM,GAAG,QAAQ,CAAC;SAC7B,IAAI,gBAAgB,IAAI,QAAQ,GAAG,CAAC;QAAE,MAAM,GAAG,QAAQ,CAAC;;QACxD,MAAM,GAAG,YAAY,CAAC;IAE3B,OAAO;QACL,YAAY,EAAE,OAAO,CAAC,IAAI;QAC1B,YAAY,EAAE,OAAO,CAAC,IAAI;QAC1B,MAAM;QACN,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,qBAAqB,EAAE,UAAU;QACjC,QAAQ;QACR,MAAM;QACN,WAAW,EAAE,OAAO,GAAG,SAAS;QAChC,KAAK;QACL,UAAU,EAAE,SAAS;QACrB,QAAQ,EAAE,IAAI,IAAI,CAAC,OAAO,CAAC,CAAC,WAAW,EAAE;KAC1C,CAAC;AACJ,CAAC"}

package/dist/test_agent/target_http.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Target HTTP chat client for the Agente de Test (PND-024 Etapa 1).
+ *
+ * POSTs to a Forge `/api/chat` (or any chat surface that follows the
+ * same shape -- `{messages: [{role, content}]}` request, response
+ * with `text` + `tool_rounds`).
+ *
+ * ASCII-only.
+ */
+import type { TargetChatClient } from './runner.js';
+export interface TargetHttpOpts {
+    /** Full URL, e.g. http://127.0.0.1:8089/api/chat. */
+    url: string;
+    /** Optional headers (Authorization, tenant id, etc). */
+    headers?: Record<string, string>;
+    /** Optional fetch impl injection for tests. */
+    fetchImpl?: typeof fetch;
+    /** Per-request timeout in milliseconds. Default 60_000. */
+    timeoutMs?: number;
+}
+export declare function makeHttpTarget(opts: TargetHttpOpts): TargetChatClient;
+//# sourceMappingURL=target_http.d.ts.map

package/dist/test_agent/target_http.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"target_http.d.ts","sourceRoot":"","sources":["../../src/test_agent/target_http.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAGpD,MAAM,WAAW,cAAc;IAC7B,qDAAqD;IACrD,GAAG,EAAE,MAAM,CAAC;IACZ,wDAAwD;IACxD,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,+CAA+C;IAC/C,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,2DAA2D;IAC3D,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAYD,wBAAgB,cAAc,CAAC,IAAI,EAAE,cAAc,GAAG,gBAAgB,CA8CrE"}

package/dist/test_agent/target_http.js

@@ -0,0 +1,51 @@
+export function makeHttpTarget(opts) {
+    const fetchImpl = opts.fetchImpl ?? fetch;
+    const headers = { 'content-type': 'application/json', ...(opts.headers ?? {}) };
+    const timeoutMs = opts.timeoutMs ?? 60_000;
+    return {
+        async chat(req) {
+            const ctrl = new AbortController();
+            const timer = setTimeout(() => ctrl.abort(), timeoutMs);
+            try {
+                const r = await fetchImpl(opts.url, {
+                    method: 'POST',
+                    headers,
+                    signal: ctrl.signal,
+                    body: JSON.stringify({ messages: req.messages }),
+                });
+                clearTimeout(timer);
+                if (!r.ok) {
+                    return {
+                        reply: '(target returned HTTP ' + r.status + ')',
+                        tool_rounds: [],
+                    };
+                }
+                const body = (await r.json().catch(() => ({})));
+                const reply = typeof body.text === 'string' ? body.text
+                    : typeof body.reply === 'string' ? body.reply
+                        : '';
+                const tool_rounds = Array.isArray(body.tool_rounds)
+                    ? body.tool_rounds.map((t) => {
+                        const verb = t.input && typeof t.input.verb_id === 'string'
+                            ? t.input.verb_id : undefined;
+                        const entry = { tool: String(t.tool ?? 'tool') };
+                        if (verb !== undefined)
+                            entry.verb_id = verb;
+                        if (t.is_error === true)
+                            entry.is_error = true;
+                        return entry;
+                    })
+                    : [];
+                return { reply, tool_rounds };
+            }
+            catch (err) {
+                clearTimeout(timer);
+                return {
+                    reply: '(target error: ' + (err instanceof Error ? err.message : String(err)) + ')',
+                    tool_rounds: [],
+                };
+            }
+        },
+    };
+}
+//# sourceMappingURL=target_http.js.map

package/dist/test_agent/target_http.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"target_http.js","sourceRoot":"","sources":["../../src/test_agent/target_http.ts"],"names":[],"mappings":"AAiCA,MAAM,UAAU,cAAc,CAAC,IAAoB;IACjD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC;IAC1C,MAAM,OAAO,GAAK,EAAE,cAAc,EAAE,kBAAkB,EAAE,GAAG,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;IAClF,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,MAAM,CAAC;IAC3C,OAAO;QACL,KAAK,CAAC,IAAI,CAAC,GAAG;YACZ,MAAM,IAAI,GAAG,IAAI,eAAe,EAAE,CAAC;YACnC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;YACxD,IAAI,CAAC;gBACH,MAAM,CAAC,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,GAAG,EAAE;oBAClC,MAAM,EAAE,MAAM;oBACd,OAAO;oBACP,MAAM,EAAE,IAAI,CAAC,MAAM;oBACnB,IAAI,EAAI,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,CAAC;iBACnD,CAAC,CAAC;gBACH,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;oBACV,OAAO;wBACL,KAAK,EAAQ,wBAAwB,GAAG,CAAC,CAAC,MAAM,GAAG,GAAG;wBACtD,WAAW,EAAE,EAAE;qBAChB,CAAC;gBACJ,CAAC;gBACD,MAAM,IAAI,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAsB,CAAC;gBACrE,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI;oBAC3C,CAAC,CAAC,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK;wBAC7C,CAAC,CAAC,EAAE,CAAC;gBACjB,MAAM,WAAW,GAAoB,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC;oBAClE,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;wBACzB,MAAM,IAAI,GAAG,CAAC,CAAC,KAAK,IAAI,OAAO,CAAC,CAAC,KAAK,CAAC,OAAO,KAAK,QAAQ;4BACzD,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;wBAChC,MAAM,KAAK,GAAkB,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,IAAI,MAAM,CAAC,EAAE,CAAC;wBAChE,IAAI,IAAI,KAAK,SAAS;4BAAM,KAAK,CAAC,OAAO,GAAI,IAAI,CAAC;wBAClD,IAAI,CAAC,CAAC,QAAQ,KAAK,IAAI;4BAAK,KAAK,CAAC,QAAQ,GAAG,IAAI,CAAC;wBAClD,OAAO,KAAK,CAAC;oBACf,CAAC,CAAC;oBACJ,CAAC,CAAC,EAAE,CAAC;gBACP,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC;YAChC,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,OAAO;oBACL,KAAK,EAAE,iBAAiB,GAAG,CAAC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,GAAG;oBACnF,WAAW,EAAE,EAAE;iBAChB,CAAC;YACJ,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC"}

package/dist/test_agent/types.d.ts

@@ -0,0 +1,117 @@
+/**
+ * Agente de Test (PND-024 Etapa 1 local).
+ *
+ * The test agent drives a multi-turn conversation against a target
+ * Forge chat server, acting as a synthetic user. The synthetic user
+ * follows a Profile: an instruction set + intent + assertions list.
+ *
+ * Etapa 1 lives entirely client-side -- the runner reuses Forge's
+ * own LLM clients (Gemini Flash Lite BYOK is the default for the
+ * synthetic-user side; the target Forge can be any provider). Etapa
+ * 2 (Platform managed) is a separate slice.
+ *
+ * The harness is intentionally minimal: each turn the runner calls
+ * the synthetic user to produce a new prompt, POSTs it to the target
+ * `/api/chat`, captures the response + tool rounds, then evaluates
+ * assertions. The loop stops when one of {success criterion met,
+ * max_turns reached, blocker assertion failed} triggers.
+ *
+ * ASCII-only.
+ */
+export type AssertionKind = 'reply_contains' | 'reply_not_contains' | 'tool_invoked' | 'tool_not_invoked' | 'reply_matches_regex' | 'turn_count_at_most';
+export interface Assertion {
+    kind: AssertionKind;
+    /** Free-text label surfaced in the report so the user can spot
+     *  which assertion fired without re-reading the kind. */
+    label: string;
+    /** Payload shape varies per kind:
+     *    reply_contains            -> { text: string, case_insensitive?: boolean }
+     *    reply_not_contains        -> { text: string, case_insensitive?: boolean }
+     *    tool_invoked              -> { verb_id: string }
+     *    tool_not_invoked          -> { verb_id: string }
+     *    reply_matches_regex       -> { regex: string, flags?: string }
+     *    turn_count_at_most        -> { max: number }
+     */
+    payload: Record<string, unknown>;
+    /** When true, a failure aborts the run early. Otherwise the
+     *  failure is recorded and the run continues. */
+    blocker?: boolean;
+}
+export interface Profile {
+    /** Stable identifier (kebab-case). Surfaced in the report + on
+     *  the CLI as the profile to run. */
+    slug: string;
+    /** Pretty name for the report header. */
+    name: string;
+    /** What kind of user this profile simulates. The synthetic-user
+     *  LLM gets this as system context. */
+    persona: string;
+    /** Initial goal the user is trying to achieve. The synthetic-user
+     *  LLM steers each turn toward this goal. */
+    goal: string;
+    /** Free-text constraints (one per line) the synthetic user must
+     *  honour -- e.g. "speak Spanish", "do not paste code". */
+    constraints: string[];
+    /** Hard cap on turns. Default 8. */
+    max_turns?: number;
+    /** Assertions evaluated after every turn. */
+    assertions: Assertion[];
+    /** Optional success criterion: when met, the run terminates with
+     *  status='passed' even if max_turns has room. Same shape as an
+     *  Assertion but kept separate so the report makes the win visible. */
+    success_when?: Assertion;
+}
+export interface ToolRoundLite {
+    /** Tool name as the chat server reported it. */
+    tool: string;
+    /** Verb id pulled from input.verb_id when present. */
+    verb_id?: string;
+    /** Whether the round errored. */
+    is_error?: boolean;
+}
+export interface TurnRecord {
+    /** Index, starting at 1. */
+    turn: number;
+    /** Synthetic user's input for the turn. */
+    user_message: string;
+    /** Target's assistant reply text. */
+    assistant_reply: string;
+    /** Tool rounds the target server reported. */
+    tool_rounds: ToolRoundLite[];
+    /** Latency end-to-end in milliseconds. */
+    latency_ms: number;
+    /** Assertion outcomes evaluated after this turn. */
+    assertion_results: AssertionResult[];
+}
+export interface AssertionResult {
+    /** Index into Profile.assertions or -1 when this is the success
+     *  criterion. */
+    index: number;
+    kind: AssertionKind;
+    label: string;
+    passed: boolean;
+    /** When false, the detail explains why. */
+    detail?: string;
+}
+export interface ProfileRunReport {
+    profile_slug: string;
+    profile_name: string;
+    status: 'passed' | 'failed' | 'incomplete';
+    /** Total turns the run produced. */
+    turns_used: number;
+    /** True when success_when fired. */
+    success_criterion_met: boolean;
+    /** Total assertion failures (blocker or not). */
+    failures: number;
+    /** Total assertion successes. */
+    passes: number;
+    /** Wall-clock duration of the whole run. */
+    duration_ms: number;
+    /** Per-turn detail. */
+    turns: TurnRecord[];
+    /** Started at ISO timestamp. */
+    started_at: string;
+    /** Ended at ISO timestamp. */
+    ended_at: string;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/test_agent/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/test_agent/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,MAAM,MAAM,aAAa,GACrB,gBAAgB,GAChB,oBAAoB,GACpB,cAAc,GACd,kBAAkB,GAClB,qBAAqB,GACrB,oBAAoB,CAAC;AAEzB,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,aAAa,CAAC;IACpB;6DACyD;IACzD,KAAK,EAAE,MAAM,CAAC;IACd;;;;;;;OAOG;IACH,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACjC;qDACiD;IACjD,OAAO,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,OAAO;IACtB;yCACqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,yCAAyC;IACzC,IAAI,EAAE,MAAM,CAAC;IACb;2CACuC;IACvC,OAAO,EAAE,MAAM,CAAC;IAChB;iDAC6C;IAC7C,IAAI,EAAE,MAAM,CAAC;IACb;+DAC2D;IAC3D,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,oCAAoC;IACpC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,6CAA6C;IAC7C,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB;;2EAEuE;IACvE,YAAY,CAAC,EAAE,SAAS,CAAC;CAC1B;AAED,MAAM,WAAW,aAAa;IAC5B,gDAAgD;IAChD,IAAI,EAAE,MAAM,CAAC;IACb,sDAAsD;IACtD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,UAAU;IACzB,4BAA4B;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,2CAA2C;IAC3C,YAAY,EAAE,MAAM,CAAC;IACrB,qCAAqC;IACrC,eAAe,EAAE,MAAM,CAAC;IACxB,8CAA8C;IAC9C,WAAW,EAAE,aAAa,EAAE,CAAC;IAC7B,0CAA0C;IAC1C,UAAU,EAAE,MAAM,CAAC;IACnB,oDAAoD;IACpD,iBAAiB,EAAE,eAAe,EAAE,CAAC;CACtC;AAED,MAAM,WAAW,eAAe;IAC9B;qBACiB;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,aAAa,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,OAAO,CAAC;IAChB,2CAA2C;IAC3C,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,gBAAgB;IAC/B,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,QAAQ,GAAG,QAAQ,GAAG,YAAY,CAAC;IAC3C,oCAAoC;IACpC,UAAU,EAAE,MAAM,CAAC;IACnB,oCAAoC;IACpC,qBAAqB,EAAE,OAAO,CAAC;IAC/B,iDAAiD;IACjD,QAAQ,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,4CAA4C;IAC5C,WAAW,EAAE,MAAM,CAAC;IACpB,uBAAuB;IACvB,KAAK,EAAE,UAAU,EAAE,CAAC;IACpB,gCAAgC;IAChC,UAAU,EAAE,MAAM,CAAC;IACnB,8BAA8B;IAC9B,QAAQ,EAAE,MAAM,CAAC;CAClB"}