← 返回 Skills 市场

Agentfarm Finder

Name: Agentfarm Finder
Author: torchesfrms

作者 moer · GitHub ↗ · v1.0.2 · MIT-0

cross-platform ⚠ suspicious

359

总下载

当前安装

版本数

在 OpenClaw 中安装

/install agentfarm-finder

功能描述

Monitors Twitter daily at 16:00 to find and filter new AI Agent, Crypto, and DeFi projects matching specific keywords, outputting raw and curated data.

使用说明 (SKILL.md)

agentfarm-finder

AI Agent 早期项目发现工具。监控 Twitter 上与 AI Agent、Crypto、DeFi 相关的新项目发布。

功能

搜索关键词：bot, agent, auto, openclaw, aiagent, 8004 + mint, farm, yield, puzzle
自动过滤噪音（GTA、游戏、新闻、主播撕逼等）
输出原始数据 + 项目精选
每天 16:00 自动运行

使用方法

# 手动运行
bash ~/.openclaw/workspace/skills/agentfarm-finder/scripts/run.sh

# 或使用 cron
# 0 16 * * * bash /Users/moer/.openclaw/workspace/skills/agentfarm-finder/scripts/run.sh

输出文件

文件	内容
`output/results_YYYY-MM-DD.csv`	原始数据
`output/results_YYYY-MM-DD_projects.csv`	项目精选

配置

修改 scripts/run.sh 中的参数：

HOURS=24 - 时间范围
COUNT=30 - 每词搜索数量

排除列表

搜索排除：nvidia, meta, tesla, slowmist 等过滤排除：grok, nft, 闲聊用户等

安全使用建议

Do not install/run this skill blindly. Before using it: (1) Inspect and remove/clean any unicode-control characters from SKILL.md and README. (2) Verify what 'bird' is (the CLI used) and whether it is authenticated on your machine — this skill will leverage the host's Twitter credentials via that client. (3) Replace hardcoded /Users/moer/... paths with a configurable OUTPUT_DIR (or run in an isolated sandbox) to avoid accidental overwriting or exposing local files. (4) If you need to run it, run first in a sandboxed account or container and review network activity; consider creating a dedicated Twitter API credential for the tool rather than relying on system-wide config. (5) If anything is unclear, ask the author to document required binaries, credentials, and to remove absolute paths and any hidden control characters. If you cannot validate these points, avoid enabling automated scheduling or granting broader privileges.

功能分析

Type: OpenClaw Skill Name: agentfarm-finder Version: 1.0.2 The skill bundle contains multiple scripts (agentfarm.sh, filter_projects.py, and scripts/filter_projects.py) with hardcoded absolute file paths pointing to a specific user's home directory (/Users/moer/), which is a significant configuration flaw that prevents the skill from functioning in other environments. While the core logic of searching Twitter via the 'bird' CLI and filtering results aligns with the stated purpose of discovering AI projects, the reliance on shell execution for external tool interaction and the inclusion of a large volume of historical output data are irregular for a standard skill bundle.

能力标签

cryptorequires-walletcan-make-purchasescan-sign-transactionsrequires-oauth-token

能力评估

ℹ Purpose & Capability

The name/description (monitor Twitter for AI/DeFi projects) align with the code: the scripts run searches and produce CSV/JSON outputs. However the package claims no required binaries/env but the main script calls the 'bird' CLI and expects it to be present and authenticated — that dependency is not declared. The code also assumes a specific user home (/Users/moer/.openclaw/...) which is unrelated to a generic skill install.

⚠ Instruction Scope

SKILL.md directs running scripts/cron which in turn run agentfarm.sh that invokes a Python search via subprocess calling 'bird'. The runtime reads/writes files under a hardcoded /Users/moer/.openclaw/... path and writes many output artifacts. The instructions do not mention that the script uses the host's Twitter credentials via the bird client, nor do they warn about the absolute paths. Prompt-injection detection (unicode-control-chars) was found in SKILL.md, which is suspicious for an instruction-only skill.

✓ Install Mechanism

There is no install spec (instruction-only), so nothing is downloaded or written by an installer. That's lower-risk. However the skill relies on external binary 'bird' which is not declared in the manifest; that mismatch increases operational risk even though there's no install step.

⚠ Credentials

requires.env lists nothing, but the code implicitly depends on (and will use) the host's 'bird' configuration/credentials to query Twitter. config.env exists in the repo with options (ENABLE_TELEGRAM, OUTPUT_DIR, etc.) but the manifest doesn't declare any required env vars. Hardcoded absolute paths to a user's home are present, which may expose or overwrite files in that location if run as that user.

ℹ Persistence & Privilege

The skill is not 'always:true' and does not request elevated platform privileges. It includes a cron wrapper (cron-daily.sh) and instructions to schedule itself, but scheduling is manual. The main concern is that it will use whatever Twitter credentials the local 'bird' client has and will write persistent output into the hardcoded workspace directory.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install agentfarm-finder
安装完成后，直接呼叫该 Skill 的名称或使用 /agentfarm-finder 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.2

- Added daily output files for a wider range of dates, including both raw and filtered project data. - Expanded historical data exports in the output directory. - No changes to README or core functionality.

v1.0.1

- Added daily outputs for filtered and unfiltered results by date, increasing automated data coverage. - Introduced new CSV output files for each run, named by date for better organization and tracking. - Updated scripts to generate and handle these new dated output files automatically. - Minor changes to existing output files and handling logic to support the new file structure.

v1.0.0

agentfarm-finder 1.0.0 – Initial Release - Monitors Twitter for emerging AI Agent, Crypto, and DeFi projects. - Searches with specific keywords and auto-filters out irrelevant content (e.g. games, news, unrelated drama). - Outputs both raw results and a curated project list as CSV files. - Runs automatically every day at 16:00, with manual run and cron support. - Configurable search range, keyword count, and exclusion lists for finer control.

元数据

Slug agentfarm-finder

版本 1.0.2

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 3

常见问题