@bgicli/bgicli 2.1.1 → 2.2.1

package/data/workflows/spatial-transcriptomics/scripts/load_example_data.py

@@ -0,0 +1,175 @@
+"""
+============================================================================
+LOAD SPATIAL TRANSCRIPTOMICS DATA
+============================================================================
+
+Functions:
+  - load_visium_heart(): Load V1_Human_Heart example dataset from 10x Genomics
+  - load_visium_data(): Load user-provided Visium data (h5ad or Space Ranger)
+
+Usage:
+  from load_example_data import load_visium_heart
+  adata = load_visium_heart()
+"""
+
+from pathlib import Path
+from typing import Optional, Union
+
+
+def load_visium_heart() -> 'anndata.AnnData':
+    """
+    Load V1_Human_Heart Visium dataset from 10x Genomics.
+
+    Downloads the processed dataset via scanpy's built-in datasets module.
+    Contains ~4,247 spots and ~33,538 genes with H&E tissue image.
+
+    Returns
+    -------
+    AnnData
+        Visium spatial transcriptomics data with:
+        - .X: count matrix (spots x genes)
+        - .obs: spot metadata
+        - .obsm['spatial']: spatial coordinates
+        - .uns['spatial']: tissue image and scalefactors
+    """
+    import scanpy as sc
+
+    print("Loading V1_Human_Heart Visium dataset...")
+    print("  Source: 10x Genomics Spatial Gene Expression")
+    print("  Description: Human heart tissue section")
+    print("  Platform: Visium Spatial Gene Expression")
+    print("  (First run downloads ~50 MB, subsequent runs use cache)")
+
+    try:
+        adata = sc.datasets.visium_sge(sample_id="V1_Human_Heart")
+    except AttributeError:
+        # Fallback for Python <3.12 (tarfile.data_filter not available)
+        print("  Using manual download fallback...")
+        adata = _download_visium_manual("V1_Human_Heart")
+
+    # Make variable names unique (critical for some Visium datasets)
+    adata.var_names_make_unique()
+
+    # Report dataset summary
+    has_image = 'spatial' in adata.uns and len(adata.uns['spatial']) > 0
+
+    print(f"\n✓ Data loaded successfully!")
+    print(f"  Spots: {adata.n_obs}")
+    print(f"  Genes: {adata.n_vars}")
+    print(f"  Spatial image: {'Yes' if has_image else 'No'}")
+    if 'spatial' in adata.obsm:
+        print(f"  Spatial coordinates: {adata.obsm['spatial'].shape}")
+
+    return adata
+
+
+def _download_file(url: str, dest: Path) -> None:
+    """Download a file with proper User-Agent header."""
+    import urllib.request
+
+    req = urllib.request.Request(url, headers={
+        'User-Agent': 'Mozilla/5.0 (compatible; scanpy-downloader)'
+    })
+    with urllib.request.urlopen(req) as response, open(dest, 'wb') as out:
+        while True:
+            chunk = response.read(8192)
+            if not chunk:
+                break
+            out.write(chunk)
+
+
+def _download_visium_manual(sample_id: str) -> 'anndata.AnnData':
+    """Manual Visium dataset download for Python <3.12 compatibility."""
+    import scanpy as sc
+    import tarfile
+
+    base_url = "https://cf.10xgenomics.com/samples/spatial-exp/1.1.0"
+
+    cache_dir = Path(sc.settings.datasetdir) / sample_id
+    cache_dir.mkdir(parents=True, exist_ok=True)
+
+    # Download H5 file (required by sc.read_visium)
+    h5_path = cache_dir / "filtered_feature_bc_matrix.h5"
+    if not h5_path.exists():
+        h5_url = f"{base_url}/{sample_id}/{sample_id}_filtered_feature_bc_matrix.h5"
+        print(f"  Downloading filtered feature-barcode matrix (H5)...")
+        _download_file(h5_url, h5_path)
+
+    # Download and extract spatial data
+    spatial_dir = cache_dir / "spatial"
+    if not spatial_dir.exists():
+        spatial_url = f"{base_url}/{sample_id}/{sample_id}_spatial.tar.gz"
+        print(f"  Downloading spatial data...")
+        tar_path = cache_dir / "spatial.tar.gz"
+        _download_file(spatial_url, tar_path)
+        with tarfile.open(tar_path, 'r:gz') as tf:
+            tf.extractall(cache_dir)
+        tar_path.unlink()
+
+    adata = sc.read_visium(cache_dir)
+    return adata
+
+
+def load_visium_data(
+    path: Union[str, Path],
+    library_id: Optional[str] = None
+) -> 'anndata.AnnData':
+    """
+    Load user-provided Visium data.
+
+    Supports:
+    - .h5ad files (AnnData format)
+    - Space Ranger output directory (with filtered_feature_bc_matrix/)
+    - .h5 files (10x HDF5 format)
+
+    Parameters
+    ----------
+    path : str or Path
+        Path to data file or Space Ranger output directory.
+    library_id : str, optional
+        Library ID for Space Ranger output.
+
+    Returns
+    -------
+    AnnData
+        Loaded spatial data with spatial coordinates.
+    """
+    import scanpy as sc
+
+    path = Path(path)
+
+    if not path.exists():
+        raise FileNotFoundError(f"Path not found: {path}")
+
+    if path.suffix == '.h5ad':
+        print(f"Loading AnnData from {path}...")
+        adata = sc.read_h5ad(path)
+    elif path.suffix == '.h5':
+        print(f"Loading 10x HDF5 from {path}...")
+        adata = sc.read_10x_h5(path)
+    elif path.is_dir():
+        print(f"Loading Space Ranger output from {path}...")
+        adata = sc.read_visium(path, library_id=library_id)
+    else:
+        raise ValueError(
+            f"Unsupported format: {path.suffix}. "
+            "Expected .h5ad, .h5, or Space Ranger output directory."
+        )
+
+    adata.var_names_make_unique()
+
+    has_image = 'spatial' in adata.uns and len(adata.uns['spatial']) > 0
+
+    print(f"\n✓ Data loaded successfully!")
+    print(f"  Spots: {adata.n_obs}")
+    print(f"  Genes: {adata.n_vars}")
+    print(f"  Spatial image: {'Yes' if has_image else 'No'}")
+    if 'spatial' in adata.obsm:
+        print(f"  Spatial coordinates: {adata.obsm['spatial'].shape}")
+
+    return adata
+
+
+if __name__ == "__main__":
+    adata = load_visium_heart()
+    print("\nExample data ready for spatial analysis!")

package/data/workflows/spatial-transcriptomics/scripts/spatial_workflow.py

@@ -0,0 +1,206 @@
+"""
+============================================================================
+SPATIAL TRANSCRIPTOMICS ANALYSIS WORKFLOW
+============================================================================
+
+Complete Visium spatial analysis: QC -> filtering -> normalization ->
+clustering -> spatial neighbors -> SVGs -> neighborhood enrichment ->
+co-occurrence.
+
+Functions:
+  - run_spatial_analysis(): Execute complete spatial workflow
+
+Usage:
+  from spatial_workflow import run_spatial_analysis
+  adata = run_spatial_analysis(adata)
+"""
+
+import numpy as np
+import pandas as pd
+from typing import Optional, List
+
+
+def run_spatial_analysis(
+    adata: 'anndata.AnnData',
+    min_genes: int = 200,
+    min_cells: int = 10,
+    max_pct_mito: float = 50.0,
+    n_top_genes: int = 2000,
+    n_pcs: int = 30,
+    resolution: float = 0.8,
+    n_neighbors: int = 15,
+    svgs_n_perms: int = 100,
+    svgs_n_jobs: int = 1,
+    output_dir: str = "visium_results"
+) -> 'anndata.AnnData':
+    """
+    Run complete spatial transcriptomics analysis.
+
+    Pipeline:
+    1. QC metrics (mitochondrial %, total counts, genes per spot)
+    2. Filter spots (min_genes, max_mito)
+    3. Normalize + log1p
+    4. HVG selection (seurat_v3 on raw counts)
+    5. PCA + Neighbors + UMAP + Leiden clustering
+    6. Spatial neighbors graph
+    7. Spatially variable genes (Moran's I)
+    8. Neighborhood enrichment
+    9. Co-occurrence analysis
+
+    Parameters
+    ----------
+    adata : AnnData
+        Raw Visium data (from load_example_data).
+    min_genes : int
+        Minimum genes per spot (default: 200).
+    min_cells : int
+        Minimum spots per gene (default: 10).
+    max_pct_mito : float
+        Maximum mitochondrial percentage (default: 50.0; heart tissue is MT-rich).
+    n_top_genes : int
+        Number of highly variable genes (default: 2000).
+    n_pcs : int
+        Number of PCA components (default: 30).
+    resolution : float
+        Leiden clustering resolution (default: 0.8).
+    n_neighbors : int
+        Number of neighbors for expression graph (default: 15).
+    svgs_n_perms : int
+        Permutations for Moran's I (default: 100).
+    svgs_n_jobs : int
+        Parallel jobs for Moran's I (default: 1).
+    output_dir : str
+        Output directory (default: "visium_results").
+
+    Returns
+    -------
+    AnnData
+        Processed adata with clustering, SVGs, enrichment results attached.
+    """
+    import scanpy as sc
+    import squidpy as sq
+    import os
+
+    os.makedirs(output_dir, exist_ok=True)
+
+    print("\n=== Step 2: Spatial Transcriptomics Analysis ===\n")
+    n_spots_initial = adata.n_obs
+
+    # --- 1. QC metrics ---
+    print("1. Calculating QC metrics...")
+    adata.var['mt'] = adata.var_names.str.startswith('MT-')
+    sc.pp.calculate_qc_metrics(
+        adata, qc_vars=['mt'], percent_top=None, log1p=False, inplace=True
+    )
+    print(f"   Median genes/spot: {adata.obs['n_genes_by_counts'].median():.0f}")
+    print(f"   Median counts/spot: {adata.obs['total_counts'].median():.0f}")
+    print(f"   Median %MT: {adata.obs['pct_counts_mt'].median():.1f}%")
+
+    # --- 2. Filter spots ---
+    print(f"\n2. Filtering spots (min_genes={min_genes}, max_mito={max_pct_mito}%)...")
+
+    # Store raw counts BEFORE any processing
+    adata.layers['counts'] = adata.X.copy()
+
+    sc.pp.filter_cells(adata, min_genes=min_genes)
+    sc.pp.filter_genes(adata, min_cells=min_cells)
+    adata = adata[adata.obs['pct_counts_mt'] < max_pct_mito, :].copy()
+
+    n_spots_after = adata.n_obs
+    retention = 100 * n_spots_after / n_spots_initial
+    print(f"   Spots: {n_spots_initial} -> {n_spots_after} ({retention:.1f}% retained)")
+
+    # --- 3. Normalize ---
+    print("\n3. Normalizing (target sum + log1p)...")
+    sc.pp.normalize_total(adata, target_sum=1e4)
+    sc.pp.log1p(adata)
+
+    # Store normalized counts for SVG computation later
+    adata.layers['normalized'] = adata.X.copy()
+
+    # --- 4. HVG selection ---
+    print(f"\n4. Selecting {n_top_genes} highly variable genes...")
+    sc.pp.highly_variable_genes(
+        adata, n_top_genes=n_top_genes, flavor='seurat_v3',
+        layer='counts'
+    )
+    n_hvgs = adata.var['highly_variable'].sum()
+    print(f"   Highly variable genes: {n_hvgs}")
+
+    # --- 5. PCA + Neighbors + UMAP + Leiden ---
+    print("\n5. Dimensionality reduction and clustering...")
+    sc.pp.scale(adata, max_value=10)
+    sc.tl.pca(adata, n_comps=n_pcs, svd_solver='arpack')
+    sc.pp.neighbors(adata, n_neighbors=n_neighbors, n_pcs=n_pcs)
+    sc.tl.umap(adata)
+    sc.tl.leiden(adata, resolution=resolution, key_added='leiden')
+
+    n_clusters = adata.obs['leiden'].nunique()
+    print(f"   PCA: {n_pcs} components")
+    print(f"   Leiden clusters: {n_clusters} (resolution={resolution})")
+
+    # --- 6. Spatial neighbors ---
+    print("\n6. Building spatial neighbors graph...")
+    sq.gr.spatial_neighbors(adata, coord_type='grid', n_neighs=6)
+    n_edges = adata.obsp['spatial_connectivities'].nnz
+    print(f"   Spatial graph: {n_edges} edges")
+
+    # --- 7. Spatially variable genes (Moran's I) ---
+    print(f"\n7. Computing spatially variable genes (Moran's I, n_perms={svgs_n_perms})...")
+
+    # Restore normalized expression for SVG computation (scale modifies X)
+    adata.X = adata.layers['normalized'].copy()
+
+    sq.gr.spatial_autocorr(
+        adata,
+        mode='moran',
+        n_perms=svgs_n_perms,
+        n_jobs=svgs_n_jobs
+    )
+
+    # Extract significant SVGs
+    moranI = adata.uns['moranI'].copy()
+    svgs = moranI[moranI['pval_norm_fdr_bh'] < 0.05].sort_values('I', ascending=False)
+    n_svgs = len(svgs)
+    print(f"   Spatially variable genes (FDR < 0.05): {n_svgs}")
+    if n_svgs > 0:
+        top5 = svgs.index[:5].tolist()
+        print(f"   Top SVGs: {', '.join(top5)}")
+
+    # Store SVG results
+    adata.uns['svg_results'] = svgs
+
+    # --- 8. Neighborhood enrichment ---
+    print("\n8. Computing neighborhood enrichment...")
+    sq.gr.nhood_enrichment(adata, cluster_key='leiden')
+    print(f"   Enrichment matrix: {n_clusters}x{n_clusters}")
+
+    # --- 9. Co-occurrence ---
+    print("\n9. Computing co-occurrence scores...")
+    sq.gr.co_occurrence(adata, cluster_key='leiden')
+    print(f"   Co-occurrence computed for {n_clusters} clusters")
+
+    # --- Store analysis parameters ---
+    adata.uns['spatial_analysis_params'] = {
+        'min_genes': min_genes,
+        'min_cells': min_cells,
+        'max_pct_mito': max_pct_mito,
+        'n_top_genes': n_top_genes,
+        'n_pcs': n_pcs,
+        'resolution': resolution,
+        'n_neighbors': n_neighbors,
+        'svgs_n_perms': svgs_n_perms,
+        'n_spots_initial': int(n_spots_initial),
+        'n_spots_after': int(n_spots_after),
+        'n_clusters': int(n_clusters),
+        'n_svgs': int(n_svgs),
+    }
+
+    print("\n" + "=" * 50)
+    print("✓ Spatial analysis completed successfully!")
+    print("=" * 50)
+    print(f"\n  Spots analyzed: {n_spots_after}")
+    print(f"  Clusters found: {n_clusters}")
+    print(f"  SVGs identified: {n_svgs}")
+
+    return adata

package/dist/bgi.js

@@ -13930,6 +13930,33 @@ async function streamOnce(client, messages, model) {
     finishReason
   };
 }
+async function compactMessages(messages, config) {
+  const prov = PROVIDERS[config.provider];
+  if (!prov) throw new Error(`Unknown provider: ${config.provider}`);
+  const baseURL = config.provider === "custom" ? config.customUrl : prov.baseURL;
+  const apiKey = getApiKey(config);
+  const client = new openai_default({ apiKey: apiKey || "none", baseURL });
+  const transcript = messages.filter((m2) => m2.role === "user" || m2.role === "assistant").map((m2) => `[${m2.role === "user" ? "\u7528\u6237" : "AI"}]: ${String(m2.content ?? "").slice(0, 2e3)}`).join("\n\n");
+  const resp = await client.chat.completions.create({
+    model: config.model,
+    messages: [
+      {
+        role: "system",
+        content: "\u4F60\u662F\u4E00\u4E2A\u5BF9\u8BDD\u6458\u8981\u52A9\u624B\u3002\u8BF7\u5C06\u4EE5\u4E0B\u5BF9\u8BDD\u5386\u53F2\u538B\u7F29\u4E3A\u7B80\u6D01\u7684\u4E2D\u6587\u6458\u8981\uFF0C\u4FDD\u7559\u6240\u6709\u5173\u952E\u6280\u672F\u4FE1\u606F\uFF1A\u6587\u4EF6\u8DEF\u5F84\u3001\u547D\u4EE4\u3001\u5206\u6790\u7ED3\u679C\u3001\u7528\u6237\u51B3\u7B56\u3001\u5DF2\u6FC0\u6D3B\u7684\u5DE5\u4F5C\u6D41/\u6280\u80FD\u3002\u6458\u8981\u5E94\u8BA9\u5BF9\u8BDD\u80FD\u591F\u65E0\u7F1D\u7EE7\u7EED\u3002"
+      },
+      {
+        role: "user",
+        content: `\u8BF7\u538B\u7F29\u4EE5\u4E0B\u5BF9\u8BDD\u5386\u53F2\uFF1A
+
+${transcript}
+
+\u8F93\u51FA\u683C\u5F0F\uFF1A\u76F4\u63A5\u8F93\u51FA\u6458\u8981\u6587\u672C\uFF0C\u4E0D\u9700\u8981\u4EFB\u4F55\u524D\u7F00\u3002`
+      }
+    ],
+    stream: false
+  });
+  return resp.choices[0]?.message?.content ?? "\uFF08\u5BF9\u8BDD\u5386\u53F2\u5DF2\u538B\u7F29\uFF09";
+}
 function getApiKey(cfg) {
   const prov = PROVIDERS[cfg.provider];
   if (prov?.envKey && process.env[prov.envKey]) return process.env[prov.envKey];
@@ -14657,7 +14684,33 @@ function routeSkill(message) {
 }
 
 // src/index.ts
-var VERSION2 = "2.1.0";
+var VERSION2 = "2.2.0";
+function installBundledData() {
+  const bundledData = (0, import_path4.join)(__dirname, "..", "data");
+  if (!(0, import_fs4.existsSync)(bundledData)) return;
+  ensureDirs();
+  const targets = [
+    { src: (0, import_path4.join)(bundledData, "workflows"), dest: WORKFLOWS_DIR, name: "\u5DE5\u4F5C\u6D41" },
+    { src: (0, import_path4.join)(bundledData, "skills"), dest: SKILLS_DIR, name: "OpenClaw Skills" },
+    { src: (0, import_path4.join)(bundledData, "tools"), dest: TOOLS_DIR, name: "\u5DE5\u5177" }
+  ];
+  let installed = false;
+  for (const { src, dest, name } of targets) {
+    if (!(0, import_fs4.existsSync)(src)) continue;
+    const isEmpty = !(0, import_fs4.existsSync)(dest) || (0, import_fs4.readdirSync)(dest).length === 0;
+    if (isEmpty) {
+      (0, import_fs4.mkdirSync)(dest, { recursive: true });
+      (0, import_fs4.cpSync)(src, dest, { recursive: true });
+      if (!installed) {
+        process.stdout.write(source_default.dim("\u6B63\u5728\u521D\u59CB\u5316\u5185\u7F6E\u6570\u636E...\n"));
+        installed = true;
+      }
+      process.stdout.write(source_default.green(`  \u2713 ${name} \u5DF2\u5B89\u88C5
+`));
+    }
+  }
+  if (installed) console.log();
+}
 function printBanner() {
   console.log(source_default.cyan.bold(`
   \u2588\u2588\u2588\u2588\u2588\u2588\u2557  \u2588\u2588\u2588\u2588\u2588\u2588\u2557 \u2588\u2588\u2557     \u2588\u2588\u2588\u2588\u2588\u2588\u2557\u2588\u2588\u2557     \u2588\u2588\u2557
@@ -14681,6 +14734,7 @@ function printHelp() {
   console.log(source_default.bold.cyan("\u2500\u2500\u2500 \u5BF9\u8BDD\u7BA1\u7406 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500"));
   console.log(`  ${source_default.cyan("/clear")}              \u6E05\u7A7A\u5BF9\u8BDD\u5386\u53F2`);
   console.log(`  ${source_default.cyan("/history")}            \u67E5\u770B\u5BF9\u8BDD\u7EDF\u8BA1\uFF08\u8F6E\u6B21 / Token \u4F30\u7B97\uFF09`);
+  console.log(`  ${source_default.cyan("/compact")}            \u7ACB\u5373\u538B\u7F29\u5BF9\u8BDD\u5386\u53F2\uFF08\u8D85 60k token \u81EA\u52A8\u89E6\u53D1\uFF09`);
   console.log(`  ${source_default.cyan("/save")} [\u6587\u4EF6\u540D]      \u4FDD\u5B58\u5BF9\u8BDD\u4E3A Markdown \u6587\u4EF6`);
   console.log(`  ${source_default.cyan("/think")} [on|off]     \u5207\u6362\u601D\u8003\u6A21\u5F0F (Qwen3 /think \u524D\u7F00)`);
   console.log();
@@ -14908,6 +14962,45 @@ ${msg.content}
   (0, import_fs4.writeFileSync)(outPath, lines.join("\n"), "utf8");
   console.log(source_default.green(`\u2713 \u5BF9\u8BDD\u5DF2\u4FDD\u5B58: ${outPath}`));
 }
+var COMPACT_TOKEN_THRESHOLD = 6e4;
+var COMPACT_KEEP_RECENT = 8;
+function estimateTokens(messages) {
+  const chars = messages.reduce((n2, m2) => n2 + String(m2.content ?? "").length, 0);
+  return Math.round(chars / 3.5);
+}
+async function maybeCompact(history, cfg) {
+  const tokens = estimateTokens(history);
+  if (tokens < COMPACT_TOKEN_THRESHOLD) return history;
+  const recent = history.slice(-COMPACT_KEEP_RECENT);
+  const old = history.slice(0, -COMPACT_KEEP_RECENT);
+  if (old.length === 0) return history;
+  process.stdout.write(source_default.dim(`
+[\u4E0A\u4E0B\u6587\u5DF2\u8FBE ~${Math.round(tokens / 1e3)}k tokens\uFF0C\u6B63\u5728\u81EA\u52A8\u538B\u7F29...]
+`));
+  try {
+    const summary = await compactMessages(old, cfg);
+    const compacted = [
+      {
+        role: "user",
+        content: `[\u5BF9\u8BDD\u5386\u53F2\u6458\u8981 \u2014 \u8BF7\u5728\u6B64\u57FA\u7840\u4E0A\u7EE7\u7EED]
+
+${summary}`
+      },
+      {
+        role: "assistant",
+        content: "\u2713 \u5DF2\u7406\u89E3\u4E4B\u524D\u7684\u5BF9\u8BDD\u6458\u8981\uFF0C\u8BF7\u7EE7\u7EED\u3002"
+      },
+      ...recent
+    ];
+    const saved = estimateTokens(history) - estimateTokens(compacted);
+    process.stdout.write(source_default.dim(`[\u538B\u7F29\u5B8C\u6210\uFF0C\u91CA\u653E\u7EA6 ~${Math.round(saved / 1e3)}k tokens]
+
+`));
+    return compacted;
+  } catch {
+    return history.slice(-COMPACT_KEEP_RECENT * 2);
+  }
+}
 async function handleCommand(input, rl, history, thinkMode) {
   const [cmd, ...rest] = input.slice(1).trim().split(/\s+/);
   const arg = rest.join(" ");
@@ -15011,6 +15104,37 @@ async function handleCommand(input, rl, history, thinkMode) {
       saveConversation(history, arg || void 0);
       break;
     }
+    case "compact": {
+      const tokens = estimateTokens(history);
+      if (history.length < 4) {
+        console.log(source_default.dim("\u5BF9\u8BDD\u592A\u77ED\uFF0C\u65E0\u9700\u538B\u7F29"));
+        break;
+      }
+      console.log(source_default.dim(`\u5F53\u524D\u5BF9\u8BDD\u7EA6 ~${Math.round(tokens / 1e3)}k tokens\uFF0C\u6B63\u5728\u538B\u7F29...`));
+      try {
+        const currentCfg = loadConfig();
+        const recent = history.slice(-COMPACT_KEEP_RECENT);
+        const old = history.slice(0, -COMPACT_KEEP_RECENT);
+        if (old.length === 0) {
+          console.log(source_default.dim("\u8FD1\u671F\u6D88\u606F\u4E0D\u8DB3\uFF0C\u65E0\u9700\u538B\u7F29"));
+          break;
+        }
+        const summary = await compactMessages(old, currentCfg);
+        const newHistory = [
+          { role: "user", content: `[\u5BF9\u8BDD\u5386\u53F2\u6458\u8981 \u2014 \u8BF7\u5728\u6B64\u57FA\u7840\u4E0A\u7EE7\u7EED]
+
+${summary}` },
+          { role: "assistant", content: "\u2713 \u5DF2\u7406\u89E3\u4E4B\u524D\u7684\u5BF9\u8BDD\u6458\u8981\uFF0C\u8BF7\u7EE7\u7EED\u3002" },
+          ...recent
+        ];
+        const after = estimateTokens(newHistory);
+        console.log(source_default.green(`\u2713 \u538B\u7F29\u5B8C\u6210: ${history.length} \u6761\u6D88\u606F \u2192 ${newHistory.length} \u6761\uFF0C~${Math.round(after / 1e3)}k tokens`));
+        return { injectHistory: newHistory };
+      } catch (err) {
+        console.error(source_default.red(`\u538B\u7F29\u5931\u8D25: ${err instanceof Error ? err.message : String(err)}`));
+      }
+      break;
+    }
     case "think": {
       const val = arg.toLowerCase();
       if (val === "on" || val === "1" || val === "true") {
@@ -15105,6 +15229,7 @@ async function handleCommand(input, rl, history, thinkMode) {
   return {};
 }
 async function main() {
+  installBundledData();
   printBanner();
   const rl = (0, import_readline.createInterface)({
     input: process.stdin,
@@ -15155,6 +15280,7 @@ async function main() {
         history = [];
         injectedSkills.clear();
       }
+      if (result.injectHistory) history = result.injectHistory;
       if (result.thinkMode !== void 0) thinkMode = result.thinkMode;
       continue;
     }
@@ -15183,7 +15309,7 @@ ${expanded}` : expanded;
       const currentCfg = loadConfig();
       const reply = await chat(history, currentCfg, systemPrompt);
       history.push({ role: "assistant", content: reply });
-      if (history.length > 40) history = history.slice(-40);
+      history = await maybeCompact(history, currentCfg);
     } catch (err) {
       const msg = err instanceof Error ? err.message : String(err);
       console.error(source_default.red(`

package/package.json

@@ -1,12 +1,13 @@
 {
   "name": "@bgicli/bgicli",
-  "version": "2.1.1",
+  "version": "2.2.1",
   "description": "BGI CLI — Bioinformatics AI terminal for Chinese researchers (百炼/DeepSeek/Kimi/Qwen)",
   "bin": {
     "bgi": "dist/bgi.js"
   },
   "files": [
     "dist/bgi.js",
+    "data/",
     "README.md"
   ],
   "publishConfig": {