valkryhx/google_adk_agent / skills/memory_retrieval_system
memory_retrieval_system
Verified专门针对记忆归档库的 L0(广度扫描) 到 L2(精准读取) 的两级检索引擎。
tools.py
View on GitHub"""
Memory Retrieval System Skill - 记忆检索引擎
基于 ripgrep 和精准文件读取,实现高性能的 L0 (全库扫描) -> L2 (精准读取) 两级检索策略。
专门针对 Swarm 日志结构的归档格式进行优化。
"""
import subprocess
import os
from typing import Optional, List
def search_memory(
pattern: str,
user_id: str,
month: Optional[str] = None,
max_results: int = 100
) -> str:
"""
[L0] 在记忆库中广度扫描线索 (基于 ripgrep)
返回按文件聚合的紧凑索引(文件名 + 匹配行号列表),不返回具体内容。
Agent 根据索引结果,使用 read_memory 精确读取感兴趣的文件和行范围。
Args:
pattern: 搜索模式 (关键词或正则)
user_id: 用户的 ID
month: 可选过滤月份,如 "2026-03"
max_results: 每个文件最大匹配数
Returns:
按文件分组的匹配行号索引
"""
try:
_PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
memory_base = os.path.join(_PROJECT_ROOT, "memory_archive", user_id)
if not os.path.exists(memory_base):
return f"[WARN] 用户 {user_id} 尚无记忆库"
# 使用 ripgrep 只输出 文件名:行号 格式(不输出匹配内容,极度紧凑)
cmd = [
"rg", "--color", "never",
"-n", # 显示行号
"--no-heading", # 每行都带文件路径
"-m", str(max_results),
"-i", # 忽略大小写
]
if month:
cmd.extend(["-g", f"*{month}*"])
cmd.extend([pattern, memory_base])
result = subprocess.run(
cmd,
capture_output=True,
text=True,
encoding='utf-8',
timeout=30
)
if result.returncode == 1:
return f"[L0] 未找到包含 '{pattern}' 的记忆。尝试更换关键词。"
elif result.returncode != 0:
return f"[ERROR] Ripgrep 执行失败: {result.stderr}"
# 聚合:按文件分组,只保留文件名和匹配行号
file_matches = {}
for line in result.stdout.strip().split('\n'):
if not line.strip():
continue
# ripgrep 输出格式 (Windows): d:\path\file.md:42:匹配内容
# 注意 Windows 路径自带一个冒号 (d:)
# 我们应该从右边分割或者用正则提取
import re
# 寻找最后一个路径:行号:内容的模式
match = re.search(r'^(.*?):(\d+):(.*)$', line)
if match:
filepath = match.group(1)
try:
line_no = int(match.group(2))
except ValueError:
continue
basename = os.path.basename(filepath)
if basename not in file_matches:
file_matches[basename] = []
file_matches[basename].append(line_no)
if not file_matches:
return f"[L0] 未找到包含 '{pattern}' 的记忆。"
# 构建紧凑索引输出
total_matches = sum(len(v) for v in file_matches.values())
output = [f"[L0] 搜索 \"{pattern}\" 找到 {len(file_matches)} 个文件,共 {total_matches} 处匹配:\n"]
for i, (fname, lines) in enumerate(file_matches.items(), 1):
line_nums = ','.join(f"L{n}" for n in sorted(lines))
output.append(f" {i}. {fname} ({len(lines)}处匹配: {line_nums})")
output.append(f"\n提示: 使用 read_memory(file_path=\"<文件名>\", start_line=N, end_line=M, user_id=\"{user_id}\") 读取具体内容。")
return '\n'.join(output)
except Exception as e:
return f"[ERROR] 搜索异常: {type(e).__name__}: {str(e)}"
def maybe_truncate(content: str, max_len: int = 15000) -> str:
"""防止 Token 爆炸的截断保护"""
if len(content) > max_len:
return content[:max_len] + f"\n\n... [内容过长已触发自我保护截断,省略后 {len(content)-max_len} 字符。请减小 end_line 重新读取剩余部分]"
return content
def read_memory(
file_path: str,
start_line: int,
end_line: int,
user_id: str
) -> str:
"""
[L2] 精准定位读取指定记忆行范围
Args:
file_path: 从 L0 获取到的目标记忆文件 (相对/绝对路径)
start_line: 起始读取行
end_line: 结束读取行
user_id: 防止越权读取的安全校验
Returns:
包含精确行号的具体记忆内容
"""
try:
# 强制行数限制保护:一次最多读 300 行
if end_line - start_line > 300:
end_line = start_line + 300
_PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
# 处理可能的相对路径
if not os.path.isabs(file_path):
# 文件名格式: 2026-03-09_app_session.md → 月份目录: 2026-03
# 先尝试直接拼接(含月份目录)
basename = os.path.basename(file_path)
month_prefix = basename[:7] if len(basename) >= 7 else ""
candidate = os.path.join(_PROJECT_ROOT, "memory_archive", user_id, month_prefix, basename)
if os.path.exists(candidate):
file_path = candidate
else:
# 兜底:直接拼接不含月份目录
file_path = os.path.join(_PROJECT_ROOT, "memory_archive", user_id, file_path)
# 安全校验: 确保读取的目标文件包含 user_id (极其简化的沙箱)
if user_id not in file_path:
return f"❌ [ERROR] 安全拦截: 无权读取其他用户的记忆档案"
if not os.path.exists(file_path):
return f"❌ [ERROR] 档案不存在: {file_path}"
with open(file_path, 'r', encoding='utf-8') as f:
lines = f.readlines()
total = len(lines)
start = max(0, start_line - 1)
end = min(total, end_line)
selected = lines[start:end]
result_lines = []
for i, line in enumerate(selected, start=start_line):
result_lines.append(f"{i:4d}: {line.rstrip()}")
content = "\n".join(result_lines)
content = maybe_truncate(content) # 字符级硬截断保护
header = f"📖 **[L2 Read] {os.path.basename(file_path)}** (Lines {start_line}-{end}):\n"
if end < end_line:
header += f" *Notice: Auto-capped at 300 lines for safety.*\n"
return header + "-"*40 + "\n" + content + "\n" + "-"*40
except Exception as e:
return f"❌ [ERROR] 读取异常: {str(e)}"
MEMORY_TOOLS = {
"search_memory": search_memory,
"read_memory": read_memory
}
def get_tools(*args, **kwargs) -> List:
return list(MEMORY_TOOLS.values())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |