@swarmdo/cli 1.4.8

package/dist/src/benchmarks/gaia-loader.js

@@ -0,0 +1,326 @@
+/**
+ * GAIA Dataset Loader — ADR-133-PR1
+ *
+ * Authenticates to Hugging Face, downloads the `gaia-benchmark/GAIA`
+ * validation split, caches it under ~/.cache/swarmdo/gaia/, and exposes
+ * a typed `loadGaia()` API consumed by the capability-gaia subcommand.
+ *
+ * Token resolution order (mirrors performance-capability.ts ANTHROPIC_API_KEY pattern):
+ *   1. $HF_TOKEN env var
+ *   2. gcloud secrets versions access latest --secret=huggingface-token
+ *   3. Fail with a clear error message
+ *
+ * This file is deliberately a skeleton / PR-1 checkpoint.  The full
+ * dataset download is gated behind a real HF_TOKEN; a 5-question smoke
+ * fixture is provided for offline / CI-without-HF testing.
+ *
+ * Refs: ADR-133, #2156
+ */
+import { execSync } from 'node:child_process';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import * as os from 'node:os';
+import * as https from 'node:https';
+import * as url from 'node:url';
+// ---------------------------------------------------------------------------
+// HF Token resolution
+// ---------------------------------------------------------------------------
+/**
+ * Resolve a Hugging Face API token using two fallbacks:
+ *   1. $HF_TOKEN env var
+ *   2. `gcloud secrets versions access latest --secret=huggingface-token`
+ *
+ * Throws with a clear error if neither is available.
+ */
+export function resolveHfToken() {
+    const envToken = process.env.HF_TOKEN;
+    if (envToken && envToken.trim())
+        return envToken.trim();
+    try {
+        const out = execSync('gcloud secrets versions access latest --secret=huggingface-token 2>/dev/null', { encoding: 'utf-8', timeout: 10_000 }).trim();
+        if (out)
+            return out;
+    }
+    catch {
+        /* fall through */
+    }
+    throw new Error('HF_TOKEN not found. Set the env var or store it in GCP Secret Manager under the name ' +
+        '"huggingface-token" (e.g. `echo -n "$TOKEN" | gcloud secrets versions add huggingface-token --data-file=-`).');
+}
+// ---------------------------------------------------------------------------
+// Cache helpers
+// ---------------------------------------------------------------------------
+function defaultCacheDir() {
+    return path.join(os.homedir(), '.cache', 'swarmdo', 'gaia');
+}
+/**
+ * Export the default cache directory path (no side effects).
+ */
+export function getDefaultCacheDir() {
+    return defaultCacheDir();
+}
+function ensureDir(dir) {
+    fs.mkdirSync(dir, { recursive: true });
+}
+function revisionCacheKey(revision, level) {
+    return `level${level}-${revision}.json`;
+}
+// ---------------------------------------------------------------------------
+// 5-question smoke fixture (no HF token required)
+// ---------------------------------------------------------------------------
+/**
+ * Minimal offline fixture for smoke tests and CI environments without HF_TOKEN.
+ * Answers are verified by hand against publicly known facts.
+ * All questions are Level 1 (no file attachments, no multi-hop tool use required).
+ *
+ * IMPORTANT: Verify every answer key with `node -e 'console.log(…)'` before
+ * adding entries to this list — three answer-key bugs were caught in session #2156.
+ */
+export const SMOKE_FIXTURE = [
+    {
+        task_id: 'smoke-001',
+        level: 1,
+        question: 'What is the capital of France?',
+        final_answer: 'Paris',
+        file_name: null,
+        file_path: null,
+    },
+    {
+        task_id: 'smoke-002',
+        level: 1,
+        question: 'How many sides does a hexagon have?',
+        final_answer: '6',
+        file_name: null,
+        file_path: null,
+    },
+    {
+        task_id: 'smoke-003',
+        level: 1,
+        question: 'What is 15 multiplied by 4?',
+        final_answer: '60',
+        file_name: null,
+        file_path: null,
+    },
+    {
+        task_id: 'smoke-004',
+        level: 1,
+        question: 'In what year did the Berlin Wall fall?',
+        final_answer: '1989',
+        file_name: null,
+        file_path: null,
+    },
+    {
+        task_id: 'smoke-005',
+        level: 1,
+        question: 'What chemical symbol represents gold on the periodic table?',
+        final_answer: 'Au',
+        file_name: null,
+        file_path: null,
+    },
+];
+// ---------------------------------------------------------------------------
+// HF dataset download (requires HF_TOKEN)
+// ---------------------------------------------------------------------------
+const HF_DATASET_REPO = 'gaia-benchmark/GAIA';
+const HF_API_BASE = 'https://huggingface.co';
+const HF_DATASETS_API = 'https://datasets-server.huggingface.co';
+/**
+ * Fetch JSON from a URL with an Authorization header.
+ * Returns parsed JSON or throws.
+ */
+async function fetchJson(url, token) {
+    return new Promise((resolve, reject) => {
+        const req = https.get(url, {
+            headers: {
+                Authorization: `Bearer ${token}`,
+                'User-Agent': 'swarmdo-gaia-loader/1.0',
+            },
+        }, (res) => {
+            if (res.statusCode === 401 || res.statusCode === 403) {
+                reject(new Error(`HF auth error ${res.statusCode} for ${url}. Check HF_TOKEN permissions (need read access to gaia-benchmark/GAIA).`));
+                res.resume();
+                return;
+            }
+            if (res.statusCode !== 200) {
+                reject(new Error(`HTTP ${res.statusCode} from ${url}`));
+                res.resume();
+                return;
+            }
+            const chunks = [];
+            res.on('data', (c) => chunks.push(c));
+            res.on('end', () => {
+                try {
+                    resolve(JSON.parse(Buffer.concat(chunks).toString('utf-8')));
+                }
+                catch (e) {
+                    reject(new Error(`JSON parse failed for ${url}: ${String(e)}`));
+                }
+            });
+            res.on('error', reject);
+        });
+        req.on('error', reject);
+        req.setTimeout(30_000, () => {
+            req.destroy(new Error(`Timeout fetching ${url}`));
+        });
+    });
+}
+/**
+ * Download the GAIA validation split for a given level from Hugging Face.
+ *
+ * Uses the HF Datasets Server API (paginated parquet/JSON rows endpoint).
+ * Caches the result locally so subsequent runs are instant.
+ *
+ * Uses per-level HF configs (2023_level1, 2023_level2, 2023_level3) so
+ * all questions for a given level fit within the 100-row API limit.
+ */
+async function downloadGaiaLevel(level, token, cacheDir) {
+    ensureDir(cacheDir);
+    // Query a stable revision identifier (the dataset card commit SHA).
+    // For now we use 'main' as the revision key for the cache filename.
+    const revision = 'main';
+    const cacheFile = path.join(cacheDir, revisionCacheKey(revision, level));
+    if (fs.existsSync(cacheFile)) {
+        const cached = JSON.parse(fs.readFileSync(cacheFile, 'utf-8'));
+        return cached;
+    }
+    // HF Datasets Server rows endpoint for the per-level 2023 config.
+    // Using "2023_level{N}" config instead of "2023_all" avoids the pagination
+    // problem: "2023_all" has 165 rows but the API caps at 100 per request,
+    // which silently drops 23 of 53 Level-1 questions. The per-level configs
+    // each have ≤100 rows (L1=53, L2=86, L3=26) and fit in a single request.
+    const levelConfig = `2023_level${level}`;
+    const url = `${HF_DATASETS_API}/rows?dataset=${encodeURIComponent(HF_DATASET_REPO)}` +
+        `&config=${levelConfig}&split=validation&offset=0&length=100`;
+    const data = await fetchJson(url, token);
+    if (!data.rows || !Array.isArray(data.rows)) {
+        throw new Error(`Unexpected HF response shape — missing .rows array. Got: ${JSON.stringify(data).slice(0, 200)}`);
+    }
+    const questions = data.rows
+        .map((r) => r.row)
+        .filter((row) => Number(row['Level']) === level)
+        .map((row) => ({
+        task_id: String(row['task_id'] ?? ''),
+        level,
+        question: String(row['Question'] ?? ''),
+        final_answer: String(row['Final answer'] ?? '').trim(),
+        file_name: row['file_name'] ? String(row['file_name']) : null,
+        file_path: null, // attachment resolution is a PR-2 concern
+        annotator_metadata: row['Annotator Metadata'],
+    }))
+        .filter((q) => q.task_id && q.question && q.final_answer);
+    fs.writeFileSync(cacheFile, JSON.stringify(questions, null, 2), 'utf-8');
+    return questions;
+}
+// ---------------------------------------------------------------------------
+// Attachment download (iter-53b)
+// ---------------------------------------------------------------------------
+/**
+ * Base URL for GAIA validation attachment files.
+ * Individual files are at: HF_FILE_BASE/<task_id>/<file_name>
+ *
+ * HF now serves files through the Xet storage layer; requests will redirect
+ * (301/302/303) to a Xet download URL that requires NO auth header.
+ * We only send the Authorization header to requests targeting huggingface.co.
+ */
+const HF_FILE_BASE = 'https://huggingface.co/datasets/gaia-benchmark/GAIA/resolve/main/2023/validation';
+/**
+ * Download a single GAIA attachment file to the cache directory.
+ * Follows HTTP redirects, sending auth only to huggingface.co domains.
+ * Returns the local file path on success, or null on any error.
+ */
+async function downloadAttachment(taskId, fileName, token, cacheDir) {
+    const destPath = path.join(cacheDir, fileName);
+    if (fs.existsSync(destPath))
+        return destPath; // already cached
+    const fileUrl = `${HF_FILE_BASE}/${taskId}/${encodeURIComponent(fileName)}`;
+    ensureDir(cacheDir);
+    return new Promise((resolve) => {
+        function doGet(requestUrl, depth) {
+            if (depth > 5) {
+                resolve(null);
+                return;
+            }
+            const parsed = new url.URL(requestUrl);
+            const headers = {
+                'User-Agent': 'swarmdo-gaia-loader/1.0',
+            };
+            if (parsed.hostname.includes('huggingface.co')) {
+                headers['Authorization'] = `Bearer ${token}`;
+            }
+            const req = https.get(requestUrl, { headers }, (res) => {
+                if (res.statusCode && [301, 302, 303, 307, 308].includes(res.statusCode)) {
+                    const loc = res.headers['location'];
+                    res.resume();
+                    if (loc)
+                        doGet(loc, depth + 1);
+                    else
+                        resolve(null);
+                    return;
+                }
+                if (res.statusCode !== 200) {
+                    res.resume();
+                    resolve(null);
+                    return;
+                }
+                const chunks = [];
+                res.on('data', (c) => chunks.push(c));
+                res.on('end', () => {
+                    try {
+                        fs.writeFileSync(destPath, Buffer.concat(chunks));
+                        resolve(destPath);
+                    }
+                    catch {
+                        resolve(null);
+                    }
+                });
+                res.on('error', () => resolve(null));
+            });
+            req.on('error', () => resolve(null));
+            req.setTimeout(60_000, () => { req.destroy(); resolve(null); });
+        }
+        doGet(fileUrl, 0);
+    });
+}
+/**
+ * Download all attachment files referenced by the questions list.
+ * Mutates each question's `file_path` field in place.
+ * Skips questions without a file_name.
+ */
+export async function resolveAttachments(questions, token, cacheDir) {
+    const withFiles = questions.filter((q) => q.file_name);
+    await Promise.all(withFiles.map(async (q) => {
+        const localPath = await downloadAttachment(q.task_id, q.file_name, token, cacheDir);
+        q.file_path = localPath;
+    }));
+}
+// ---------------------------------------------------------------------------
+// Public API
+// ---------------------------------------------------------------------------
+/**
+ * Load GAIA validation questions.
+ *
+ * - With `smokeOnly: true` (or when HF_TOKEN is unavailable): returns the 5-question
+ *   smoke fixture — no network call, no token required.
+ * - Otherwise: authenticates to HF, downloads level N questions, caches locally.
+ *
+ * @throws if HF_TOKEN is missing and smokeOnly is false
+ */
+export async function loadGaia(options = {}) {
+    const { level = 1, limit, smokeOnly = false, cacheDir = defaultCacheDir() } = options;
+    if (smokeOnly) {
+        const filtered = SMOKE_FIXTURE.filter((q) => q.level === level);
+        return limit !== undefined ? filtered.slice(0, limit) : filtered;
+    }
+    const token = resolveHfToken();
+    const questions = await downloadGaiaLevel(level, token, cacheDir);
+    // Resolve attachment files in parallel (iter-53b: populates file_path on each question)
+    await resolveAttachments(questions, token, cacheDir);
+    return limit !== undefined ? questions.slice(0, limit) : questions;
+}
+/**
+ * Returns the cache directory path (does not create it).
+ */
+export function getGaiaCacheDir(override) {
+    return override ?? defaultCacheDir();
+}
+//# sourceMappingURL=gaia-loader.js.map

package/dist/src/benchmarks/gaia-loader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-loader.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-loader.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAC9B,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAClC,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAC9B,OAAO,KAAK,KAAK,MAAM,YAAY,CAAC;AACpC,OAAO,KAAK,GAAG,MAAM,UAAU,CAAC;AAoChC,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;GAMG;AACH,MAAM,UAAU,cAAc;IAC5B,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC;IACtC,IAAI,QAAQ,IAAI,QAAQ,CAAC,IAAI,EAAE;QAAE,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC;IAExD,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,QAAQ,CAClB,8EAA8E,EAC9E,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CACvC,CAAC,IAAI,EAAE,CAAC;QACT,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,kBAAkB;IACpB,CAAC;IAED,MAAM,IAAI,KAAK,CACb,uFAAuF;QACvF,8GAA8G,CAC/G,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,gBAAgB;AAChB,8EAA8E;AAE9E,SAAS,eAAe;IACtB,OAAO,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;AAC9D,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,kBAAkB;IAChC,OAAO,eAAe,EAAE,CAAC;AAC3B,CAAC;AAED,SAAS,SAAS,CAAC,GAAW;IAC5B,EAAE,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;AACzC,CAAC;AAED,SAAS,gBAAgB,CAAC,QAAgB,EAAE,KAAgB;IAC1D,OAAO,QAAQ,KAAK,IAAI,QAAQ,OAAO,CAAC;AAC1C,CAAC;AAED,8EAA8E;AAC9E,kDAAkD;AAClD,8EAA8E;AAE9E;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,aAAa,GAAmB;IAC3C;QACE,OAAO,EAAE,WAAW;QACpB,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,gCAAgC;QAC1C,YAAY,EAAE,OAAO;QACrB,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;KAChB;IACD;QACE,OAAO,EAAE,WAAW;QACpB,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,qCAAqC;QAC/C,YAAY,EAAE,GAAG;QACjB,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;KAChB;IACD;QACE,OAAO,EAAE,WAAW;QACpB,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,6BAA6B;QACvC,YAAY,EAAE,IAAI;QAClB,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;KAChB;IACD;QACE,OAAO,EAAE,WAAW;QACpB,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,wCAAwC;QAClD,YAAY,EAAE,MAAM;QACpB,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;KAChB;IACD;QACE,OAAO,EAAE,WAAW;QACpB,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,6DAA6D;QACvE,YAAY,EAAE,IAAI;QAClB,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;KAChB;CACF,CAAC;AAEF,8EAA8E;AAC9E,0CAA0C;AAC1C,8EAA8E;AAE9E,MAAM,eAAe,GAAG,qBAAqB,CAAC;AAC9C,MAAM,WAAW,GAAG,wBAAwB,CAAC;AAC7C,MAAM,eAAe,GAAG,wCAAwC,CAAC;AAEjE;;;GAGG;AACH,KAAK,UAAU,SAAS,CAAC,GAAW,EAAE,KAAa;IACjD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,CACnB,GAAG,EACH;YACE,OAAO,EAAE;gBACP,aAAa,EAAE,UAAU,KAAK,EAAE;gBAChC,YAAY,EAAE,yBAAyB;aACxC;SACF,EACD,CAAC,GAAG,EAAE,EAAE;YACN,IAAI,GAAG,CAAC,UAAU,KAAK,GAAG,IAAI,GAAG,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;gBACrD,MAAM,CAAC,IAAI,KAAK,CAAC,iBAAiB,GAAG,CAAC,UAAU,QAAQ,GAAG,yEAAyE,CAAC,CAAC,CAAC;gBACvI,GAAG,CAAC,MAAM,EAAE,CAAC;gBACb,OAAO;YACT,CAAC;YACD,IAAI,GAAG,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;gBAC3B,MAAM,CAAC,IAAI,KAAK,CAAC,QAAQ,GAAG,CAAC,UAAU,SAAS,GAAG,EAAE,CAAC,CAAC,CAAC;gBACxD,GAAG,CAAC,MAAM,EAAE,CAAC;gBACb,OAAO;YACT,CAAC;YACD,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,GAAG,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9C,GAAG,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE;gBACjB,IAAI,CAAC;oBACH,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC/D,CAAC;gBAAC,OAAO,CAAC,EAAE,CAAC;oBACX,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,GAAG,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;gBAClE,CAAC;YACH,CAAC,CAAC,CAAC;YACH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC1B,CAAC,CACF,CAAC;QACF,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QACxB,GAAG,CAAC,UAAU,CAAC,MAAM,EAAE,GAAG,EAAE;YAC1B,GAAG,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,oBAAoB,GAAG,EAAE,CAAC,CAAC,CAAC;QACpD,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;;GAQG;AACH,KAAK,UAAU,iBAAiB,CAC9B,KAAgB,EAChB,KAAa,EACb,QAAgB;IAEhB,SAAS,CAAC,QAAQ,CAAC,CAAC;IAEpB,oEAAoE;IACpE,oEAAoE;IACpE,MAAM,QAAQ,GAAG,MAAM,CAAC;IACxB,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,gBAAgB,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;IAEzE,IAAI,EAAE,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,SAAS,EAAE,OAAO,CAAC,CAAmB,CAAC;QACjF,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,kEAAkE;IAClE,2EAA2E;IAC3E,wEAAwE;IACxE,yEAAyE;IACzE,yEAAyE;IACzE,MAAM,WAAW,GAAG,aAAa,KAAK,EAAE,CAAC;IACzC,MAAM,GAAG,GACP,GAAG,eAAe,iBAAiB,kBAAkB,CAAC,eAAe,CAAC,EAAE;QACxE,WAAW,WAAW,uCAAuC,CAAC;IAEhE,MAAM,IAAI,GAAG,MAAM,SAAS,CAAC,GAAG,EAAE,KAAK,CAAsD,CAAC;IAE9F,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,4DAA4D,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IACpH,CAAC;IAED,MAAM,SAAS,GAAmB,IAAI,CAAC,IAAI;SACxC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC;SACjB,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC;SAC/C,GAAG,CAAC,CAAC,GAAG,EAAgB,EAAE,CAAC,CAAC;QAC3B,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;QACrC,KAAK;QACL,QAAQ,EAAE,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC;QACvC,YAAY,EAAE,MAAM,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE;QACtD,SAAS,EAAE,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI;QAC7D,SAAS,EAAE,IAAI,EAAE,0CAA0C;QAC3D,kBAAkB,EAAE,GAAG,CAAC,oBAAoB,CAAwC;KACrF,CAAC,CAAC;SACF,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC;IAE5D,EAAE,CAAC,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;IACzE,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,8EAA8E;AAC9E,iCAAiC;AACjC,8EAA8E;AAE9E;;;;;;;GAOG;AACH,MAAM,YAAY,GAChB,kFAAkF,CAAC;AAErF;;;;GAIG;AACH,KAAK,UAAU,kBAAkB,CAC/B,MAAc,EACd,QAAgB,EAChB,KAAa,EACb,QAAgB;IAEhB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;IAC/C,IAAI,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,QAAQ,CAAC,CAAC,iBAAiB;IAE/D,MAAM,OAAO,GAAG,GAAG,YAAY,IAAI,MAAM,IAAI,kBAAkB,CAAC,QAAQ,CAAC,EAAE,CAAC;IAC5E,SAAS,CAAC,QAAQ,CAAC,CAAC;IAEpB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;QAC7B,SAAS,KAAK,CAAC,UAAkB,EAAE,KAAa;YAC9C,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;gBAAC,OAAO,CAAC,IAAI,CAAC,CAAC;gBAAC,OAAO;YAAC,CAAC;YAEzC,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YACvC,MAAM,OAAO,GAA2B;gBACtC,YAAY,EAAE,yBAAyB;aACxC,CAAC;YACF,IAAI,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;gBAC/C,OAAO,CAAC,eAAe,CAAC,GAAG,UAAU,KAAK,EAAE,CAAC;YAC/C,CAAC;YAED,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,UAAU,EAAE,EAAE,OAAO,EAAE,EAAE,CAAC,GAAG,EAAE,EAAE;gBACrD,IAAI,GAAG,CAAC,UAAU,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,CAAC;oBACzE,MAAM,GAAG,GAAG,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;oBACpC,GAAG,CAAC,MAAM,EAAE,CAAC;oBACb,IAAI,GAAG;wBAAE,KAAK,CAAC,GAAG,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;;wBAC1B,OAAO,CAAC,IAAI,CAAC,CAAC;oBACnB,OAAO;gBACT,CAAC;gBACD,IAAI,GAAG,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;oBAC3B,GAAG,CAAC,MAAM,EAAE,CAAC;oBACb,OAAO,CAAC,IAAI,CAAC,CAAC;oBACd,OAAO;gBACT,CAAC;gBACD,MAAM,MAAM,GAAa,EAAE,CAAC;gBAC5B,GAAG,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;gBAC9C,GAAG,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE;oBACjB,IAAI,CAAC;wBACH,EAAE,CAAC,aAAa,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;wBAClD,OAAO,CAAC,QAAQ,CAAC,CAAC;oBACpB,CAAC;oBAAC,MAAM,CAAC;wBACP,OAAO,CAAC,IAAI,CAAC,CAAC;oBAChB,CAAC;gBACH,CAAC,CAAC,CAAC;gBACH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;YACvC,CAAC,CAAC,CAAC;YACH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;YACrC,GAAG,CAAC,UAAU,CAAC,MAAM,EAAE,GAAG,EAAE,GAAG,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAClE,CAAC;QACD,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IACpB,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,SAAyB,EACzB,KAAa,EACb,QAAgB;IAEhB,MAAM,SAAS,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IACvD,MAAM,OAAO,CAAC,GAAG,CACf,SAAS,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC,EAAE,EAAE;QACxB,MAAM,SAAS,GAAG,MAAM,kBAAkB,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,SAAU,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;QACrF,CAAC,CAAC,SAAS,GAAG,SAAS,CAAC;IAC1B,CAAC,CAAC,CACH,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,aAAa;AACb,8EAA8E;AAE9E;;;;;;;;GAQG;AACH,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,UAA2B,EAAE;IAC1D,MAAM,EAAE,KAAK,GAAG,CAAC,EAAE,KAAK,EAAE,SAAS,GAAG,KAAK,EAAE,QAAQ,GAAG,eAAe,EAAE,EAAE,GAAG,OAAO,CAAC;IAEtF,IAAI,SAAS,EAAE,CAAC;QACd,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,CAAC;QAChE,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;IACnE,CAAC;IAED,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,MAAM,SAAS,GAAG,MAAM,iBAAiB,CAAC,KAAK,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;IAClE,wFAAwF;IACxF,MAAM,kBAAkB,CAAC,SAAS,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;IACrD,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACrE,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,eAAe,CAAC,QAAiB;IAC/C,OAAO,QAAQ,IAAI,eAAe,EAAE,CAAC;AACvC,CAAC"}

package/dist/src/benchmarks/gaia-tools/file_read.d.ts

@@ -0,0 +1,35 @@
+/**
+ * GAIA Tool: file_read — ADR-133-PR2 / iter-53b attachment-tools
+ *
+ * Reads a file from the local filesystem and returns its contents as a
+ * UTF-8 string.  Performs content-type dispatch based on extension +
+ * magic bytes:
+ *
+ * Supported extraction formats (iter-53b):
+ *   - Plain text, JSON, CSV, XML, HTML, Markdown, JS/TS, Python, shell, YAML
+ *   - XLSX  — openpyxl subprocess (cell values + fill colours)
+ *   - PPTX  — python-pptx subprocess (per-slide text)
+ *   - PNG / JPEG / GIF / WebP — returns IMAGE_BASE64 marker for vision API
+ *   - MP3 / WAV — OpenAI Whisper (tiny model) subprocess transcript
+ *   - .py source file — returned as UTF-8 text (no execution)
+ *
+ * For PDF / DOCX / other binary: descriptive stub (PR-4 deferred).
+ *
+ * IMAGE_BASE64 marker format:
+ *   [IMAGE_BASE64:{"mediaType":"image/png","base64":"...","path":"/abs/path"}]
+ *
+ * The agent loop in gaia-agent.ts must parse this marker when it appears
+ * in a tool_result and convert it to an Anthropic vision content block.
+ *
+ * Maximum file size: 5 MB.  Paths must be absolute.
+ *
+ * Refs: ADR-133, #2156, iter-53b
+ */
+import { GaiaTool, ToolDefinition } from './types.js';
+export declare class FileReadTool implements GaiaTool {
+    readonly name = "file_read";
+    readonly definition: ToolDefinition;
+    execute(input: Record<string, unknown>): Promise<string>;
+}
+export declare function createFileReadTool(): FileReadTool;
+//# sourceMappingURL=file_read.d.ts.map

package/dist/src/benchmarks/gaia-tools/file_read.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"file_read.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/file_read.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAKH,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AA8RtD,qBAAa,YAAa,YAAW,QAAQ;IAC3C,QAAQ,CAAC,IAAI,eAAe;IAE5B,QAAQ,CAAC,UAAU,EAAE,cAAc,CAqBjC;IAEI,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;CAgF/D;AAMD,wBAAgB,kBAAkB,IAAI,YAAY,CAEjD"}