Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Batch Research 技能 · miantiao-me | DecimalAI

miantiao-me/batch-research

SkillScore86Maintained

批量数据采集技能，负责分批并发调度 researcher agent 抓取所有数据源。

researchby miantiao-me

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/miantiao-me-batch-research/SKILL.md

decimalai skills pull miantiao-me-batch-research --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/miantiao-me-batch-research/SKILL.md https://app.decimal.ai/s/miantiao-me-batch-research/SKILL.md

Raw file: /s/miantiao-me-batch-research/SKILL.md ↗ · machine-readable summary at /s/miantiao-me-batch-research.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 27h ago

—+27% tokens0% turns86% of whole cases pass with skill (19/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	-16%	0%
case-05	✗→✓	▲ Improved	-5%	0%
case-06	✗→✓	▲ Improved	26%	0%
case-07	✗→✓	▲ Improved	42%	0%
case-08	✗→✓	▲ Improved	4%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

miantiao-me / aigc-weekly

agent/.opencode/skills/batch-research/SKILL.md

View on GitHub ↗

License AGPL-3.0Imported 8/3/2026Version v1

Skill contents

SKILL.md

Batch Research 技能

此技能用于指导 /weekly 命令如何高效、分批、并发地从多个数据源采集信息。

核心职责

作为批量调度器，负责：

解析数据源列表
生成所有待抓取 URL
分批并发调用 researcher agent
汇总结果生成报告

工作流程

Step 1: 准备阶段

读取数据源：从 .opencode/REFERENCE.md 获取完整数据源列表
生成 URL 列表：

静态 URL：直接使用
动态 URL（Hacker News）：使用 generateHNUrls(start_date, end_date)

javascript
import { generateHNUrls } from '.opencode/utils.mjs'

// Hacker News - 每天一个 URL
const hnUrls = generateHNUrls(start_date, end_date)
// 返回: [
//   "https://news.ycombinator.com/front?day=2026-03-22",
//   "https://news.ycombinator.com/front?day=2026-03-23",
//   ...
// ]

Step 2: 分批策略

将所有 URL 按优先级分为 3 批，每批 10-12 个 URL：

| 批次 | 数据源类型 | 来源 | | ----------- | ------------------- | ------------------------------------------ | | Batch 1 | Important Resources | REFERENCE.md 中 "Important Resources" 部分 | | Batch 2 | Blogs & Websites | REFERENCE.md 中 "Blogs & Websites" 部分 | | Batch 3 | KOL & Influencers | REFERENCE.md 中 "KOL & Influencers" 部分 |

Step 3: 并发调度

并发配置

| 参数 | 值 | 说明 | | ------------------- | ------------ | -------------- | | max_parallel | 5 | 每轮最大并发数 | | batch_interval | 3s | 批次间等待时间 | | domain_rate_limit | 2 req/domain | 同域名限流 |

调度规则

对每个批次：

分轮执行：将批次内 URL 按 max_parallel 分轮（如 12 个 URL 分 3 轮：5 + 5 + 2）
轮内并发：同时启动该轮所有 researcher agent
等待完成：等待当前轮所有 researcher 返回结果
进入下一轮：当前轮全部完成后，等待 batch_interval，再启动下一轮
批次完成后：进入下一个批次

调用 researcher 的参数格式：

yaml
url: https://news.ycombinator.com/front?day=2026-03-22
source_name: Hacker News
week_id: Y26W12
start_date: 2026-03-22
end_date: 2026-03-28
current_date: 2026-03-25
timezone: UTC+0

并发调用示例（伪代码）：

# Batch 1: Important Resources
并行调用：
  - researcher(url: "https://news.ycombinator.com/front?day=2026-03-22", source_name: "Hacker News")
  - researcher(url: "https://news.ycombinator.com/front?day=2026-03-23", source_name: "Hacker News")
  - researcher(url: "https://drafts.miantiao.me/", source_name: "Miantiao Drafts")
  - researcher(url: "https://www.solidot.org/search?tid=151", source_name: "Solidot")
  - ...

等待 Batch 1 全部完成

# Batch 2: Blogs & Websites
并行调用：
  - researcher(url: "https://www.anthropic.com/engineering", source_name: "Anthropic Engineering")
  - researcher(url: "https://claude.com/blog", source_name: "Claude Blog")
  - ...

等待 Batch 2 全部完成

# Batch 3: KOL & Influencers
并行调用：
  - researcher(url: "https://baoyu.io/", source_name: "Baoyu")
  - ...

等待 Batch 3 全部完成

Step 4: 日志记录

所有日志统一写入 logs/weekly-{week_id}.log，仅用于人类审计，不作为恢复依据。

日志格式：

[2026-03-25T12:34:56Z] [PHASE1] [INFO] 开始抓取 Hacker News
[2026-03-25T12:35:10Z] [PHASE1] [OK] Hacker News - 5 篇文章
[2026-03-25T12:35:15Z] [PHASE1] [FAIL] daily.dev - 429 Too Many Requests (retried 2)

日志级别：

| 级别 | 用途 | | ------ | -------------------------------- | | INFO | 阶段/任务开始 | | OK | 任务成功（含文章数） | | FAIL | 任务失败（含错误原因和重试次数） |

Step 5: 汇总报告

所有批次完成后，在日志末尾生成汇总：

[2026-03-25T12:45:00Z] [PHASE1] [SUMMARY] 总数据源: 30 | 成功: 27 (X篇) | 失败: 3

约束与注意事项

全量抓取：必须抓取所有数据源，不能跳过
批次顺序：必须按批次顺序执行，等待当前批次完成后再进入下一批
错误隔离：单个 researcher 失败不影响其他
重试由 researcher 处理：本技能不负责重试，由 researcher 自行处理
进入下一阶段前：必须完成所有批次的抓取

Want to see if miantiao-me/batch-research would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/19/2026, 19:52 UTC · 22 cases

Measured on 21 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+64%

86% vs 23%

Agent turns

0%

1.0 vs 1.0

Tokens

+27%

2,391.05 vs 1,882

fail → pass · 14pass → pass · 5fail → fail · 3

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 Please execute the `/weekly` content gathering workflow for week Y26W12 (March 22 to March 28, 2026). Fetch all news and article links across our catalog of sources using parallel sub-agents, track execution progress in a log file, and give me a final overview showing total sources processed, successes, and failures once complete.	fail→fail	13,571	6,361	-53%	1	1	0%	2,575	1,891	-27%	0	0	—
case-02 Run the batch information harvesting workflow for the past week. Parse our reference source lists—including dynamic Hacker News dates—and dispatch research tasks concurrently across batches. Make sure all sources are queried through individual researcher agents, write timestamped audit entries, and report the aggregated results.	fail→fail	20,623	8,954	-57%	1	1	0%	3,834	1,857	-52%	0	0	—
case-03 I need to start the `/weekly` research batch job for current date 2026-03-25 covering the latest seven-day window. Please process all configured feeds—ranging from technical blogs to industry influencers—by scheduling parallel researchers, ensuring failure in one source won't stop others, and presenting a final execution tally at the end.	fail→fail	14,486	13,934	-4%	1	1	0%	2,347	4,487	+91%	0	0	—
case-04 We are preparing URL targets for Hacker News coverage during a weekly run spanning 2026-03-01 to 2026-03-07. What project utility script and function should be invoked to construct these daily front-page query targets?	fail→pass	14,213	3,730	-74%	1	1	0%	2,724	2,286	-16%	0	0	—
case-05 Where should the batch engine pull the main catalog of target websites, blogs, and influencer feeds before initiating the extraction steps?	fail→pass	12,111	2,332	-81%	1	1	0%	2,024	1,931	-5%	0	0	—
case-06 We have URLs categorized under 'Important Resources' in our project feed document. Which batch grouping should process these items during execution?	fail→pass	8,088	1,461	-82%	1	1	0%	1,329	1,677	+26%	0	0	—
case-07 When scheduling automated fetch tasks for feeds listed under 'Blogs & Websites', which batch slot should they occupy?	fail→pass	8,109	1,753	-78%	1	1	0%	1,259	1,789	+42%	0	0	—
case-08 Our feed list includes 'KOL & Influencers' entries. Into which batch sequence should these accounts be placed?	fail→pass	12,351	2,491	-80%	1	1	0%	1,796	1,870	+4%	0	0	—
case-09 To avoid overloading system resources during concurrent sub-agent dispatch, what maximum parallel worker count should be configured per dispatch round?	fail→pass	12,358	2,156	-83%	1	1	0%	2,116	1,792	-15%	0	0	—
case-19 Can the workflow proceed to post-processing summary generation if 2 out of 10 sources in Batch 3 encounter persistent connection errors?	fail→pass	10,071	5,279	-48%	1	1	0%	1,605	2,378	+48%	0	0	—
case-10 How long should the execution engine pause between completing one batch and starting the subsequent batch?	fail→pass	12,066	2,242	-81%	1	1	0%	1,865	1,842	-1%	0	0	—
case-11 To prevent triggering web server blockades across target domains, what rate limit rule should be enforced per domain?	fail→pass	11,692	4,366	-63%	1	1	0%	2,041	1,780	-13%	0	0	—
case-12 Where should run execution audit logs be stored for week ID Y26W12?	fail→pass	8,418	1,769	-79%	1	1	0%	1,369	1,774	+30%	0	0	—
case-13 Show an example audit log entry recording the initiation of a fetch task for target 'Solidot' under phase 1.	fail→pass	7,276	2,288	-69%	1	1	0%	1,329	1,906	+43%	0	0	—
case-14 Show how a successful fetch outcome yielding 8 articles from target 'Claude Blog' should be recorded in the system audit stream.	fail→pass	11,157	4,287	-62%	1	1	0%	2,091	2,298	+10%	0	0	—
case-15 If a HTTP 429 rate limit error occurs twice while fetching 'daily.dev', how should this failure be captured in the execution log?	fail→pass	11,729	3,784	-68%	1	1	0%	2,197	2,184	-1%	0	0	—
case-16 Construct the exact YAML payload required when invoking an individual researcher agent for fetching Hacker News on date 2026-03-22 during week Y26W12 with current date 2026-03-25.	fail→pass	4,905	4,044	-18%	1	1	0%	1,029	2,340	+127%	0	0	—
case-17 When an individual web request fails during a concurrent batch run, should the outer batch orchestrator handle HTTP retries or defer them?	pass→pass	14,086	6,789	-52%	1	1	0%	2,184	2,620	+20%	0	0	—
case-18 If one researcher worker crashes while parsing a blog feed in Batch 2, what impact should this have on the remaining active workers in that batch?	pass→pass	8,914	3,897	-56%	1	1	0%	1,379	2,122	+54%	0	0	—
case-20 Write the prompt instructions for an individual researcher agent tasked with scraping and extracting article titles, main takeaways, and outbound links from a single URL.	pass→pass	12,243	11,996	-2%	1	1	0%	2,100	3,531	+68%	0	0	—
case-21 Format a final weekly Markdown report summarizing key AI breakthroughs collected from 15 articles, organizing findings into categories like 'LLM Architecture' and 'Tooling'.	pass→pass	17,870	18,538	+4%	1	1	0%	2,961	4,560	+54%	0	0	—
case-22 Add three new tech news feeds to our project reference document under the appropriate category headings and explain how to maintain valid Markdown lists.	pass→pass	7,126	9,602	+35%	1	1	0%	1,202	3,154	+162%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +64 percentage points is the difference between those two pass rates over the 21 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 22.7%
Pass rate with the skill: 86.4%
Cases attempted: 22
Cases counted toward lift: 21
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/19/2026, 19:52 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We are preparing URL targets for Hacker News coverage during a weekly run spanning 2026-03-01 to 2026-03-07. What project utility script and function should be invoked to construct these daily front-page query targets?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run86

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+63% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested yesterday

Other measured skills in the registry, with their headline benchmark lift.