← Back to Skills Marketplace
443
Downloads
0
Stars
1
Active Installs
6
Versions
Install in OpenClaw
/install undersheet
Description
Persistent thread memory for OpenClaw agents across any platform — Moltbook, Hacker News, Reddit, Discord, Twitter. Tracks threads, surfaces only new replies...
Usage Guidance
UnderSheet appears to do what it says: track threads and surface new replies across platforms. Before installing: 1) Review the source files locally (don't blindly run the curl commands). The repo includes executable Python adapters that will be written into ~/.openclaw/skills/undersheet. 2) Be aware it asks you to store credentials and will cache tokens/cookies under ~/.config/undersheet (and ~/.config/moltbook). Those files grant posting power—treat them like secrets. 3) Note the Moltbook adapter contains an automated CAPTCHA solver; this can bypass protections and may violate a platform's terms of service—consider whether you want that capability enabled. 4) If you only need read-only behavior, prefer supplying least-privilege credentials (e.g., bearer tokens for read-only) and avoid placing write-capable keys in the config. 5) Prefer cloning the GitHub repo and inspecting files before use, run initial tests in a sandboxed account or environment, and consider disabling or removing post-comment functionality if you don't want autonomous posting.
Capability Analysis
Type: OpenClaw Skill
Name: undersheet
Version: 1.2.3
The OpenClaw AgentSkills skill bundle 'undersheet' is classified as benign. Its code and documentation consistently align with its stated purpose of providing persistent thread memory and feed tracking across various social platforms. While it handles sensitive credentials (API keys, usernames/passwords, OAuth tokens) by storing them in plaintext JSON files in `~/.config/undersheet/`, this is for its own operation and is a common configuration pattern, not an indicator of malicious intent. The skill makes extensive network calls to legitimate API endpoints and uses `importlib` for a standard plugin architecture. Prompt injection instructions in `SKILL.md` are limited to configuring the skill for persistence (heartbeat) which is part of its stated functionality, without any evidence of deceptive or harmful objectives.
Capability Assessment
Purpose & Capability
Name/description (persistent thread memory across platforms) aligns with the code and instructions: adapters for Discord, HN, Moltbook, Reddit, Twitter are present and the skill reads/writes per-platform state and credentials. The included platform adapters, feed/track/heartbeat commands, and credential file layout are coherent with the stated purpose.
Instruction Scope
SKILL.md tells the agent/user to download files into ~/.openclaw/skills/undersheet and to place credentials under ~/.config/undersheet (and ~/.config/moltbook). The code reads/writes those exact paths (cookie jar, OAuth token cache, state JSONs). This is expected for a local agent skill but expands the scope beyond in-memory operations: it persists tokens, cookies, and state to disk and can post (write) on your behalf. The moltbook adapter includes a built-in CAPTCHA solver that programmatically solves verification challenges—this capability goes beyond simple syncing and can be used to bypass rate-limits/captcha protections, which is higher risk and may violate platform rules.
Install Mechanism
SKILL.md recommends curl'ing raw.githubusercontent.com or cloning the GitHub repo — GitHub raw is a common release host (lower risk than arbitrary URLs). The package as distributed includes many code files; although the registry lists this as instruction-only, the skill bundle contains executable Python files, so the actual install involves writing code to disk (the curl/clone instructions do that). No packaged binary downloads or obscure URLs were used.
Credentials
No environment variables or unrelated credentials are requested by the registry metadata. The code requires per-platform credentials (Discord bot token, Reddit OAuth keys, Hacker News username/password, Twitter bearer + OAuth keys, Moltbook API key) which are all directly relevant to the corresponding adapters. The skill also stores cookies and cached tokens under ~/.config/undersheet which is proportionate to its functionality but worth noting because those files contain auth material.
Persistence & Privilege
The skill is not marked always:true and does not request elevated system-wide privileges. It persistently stores state, cookies, and OAuth tokens in the user's home directory (expected for this use case). It can perform write actions (post_comment) if credentials are provided, which is expected behavior but increases blast radius if misused.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install undersheet - After installation, invoke the skill by name or use
/undersheet - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.2.3
Atomic state writes via os.replace — no more torn JSON on concurrent access or power loss. Corrupt state file recovers cleanly.
v1.2.2
Pure stdlib, zero deps, community-open with CONTRIBUTING.md, dupe-reply guard (mark_replied, get_unanswered_comments)
v1.2.0
**Proxy/VPN support added in v1.2.0**
- Added support for proxy and VPN routing via ~/.config/undersheet/proxy.json (HTTP/SOCKS5), command-line flags, and environment variables.
- Updated README and SKILL documentation to describe proxy setup for agent traffic.
- No changes to dependencies; SOCKS5 uses optional pysocks.
- General improvements to documentation and usage instructions.
v1.1.1
fix: add SKILL.md frontmatter (fixes suspicious flag)
v1.1.0
v1.1: Twitter/X adapter, Discord/Twitter setup docs, status command, verbose mode, recipes, real demo output in README
v1.0.0
Initial release: persistent thread memory for any platform. Moltbook, HN, Reddit, Discord adapters included.
Metadata
Frequently Asked Questions
What is Undersheet?
Persistent thread memory for OpenClaw agents across any platform — Moltbook, Hacker News, Reddit, Discord, Twitter. Tracks threads, surfaces only new replies... It is an AI Agent Skill for Claude Code / OpenClaw, with 443 downloads so far.
How do I install Undersheet?
Run "/install undersheet" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Undersheet free?
Yes, Undersheet is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Undersheet support?
Undersheet is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Undersheet?
It is built and maintained by ubgb (@ubgb); the current version is v1.2.3.
More Skills