Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
Cortex Memory
v1.1.2Long-term memory for OpenClaw agents — auto-recall before turns, auto-capture after, tools for search/save/forget.
⭐ 0· 869·4 current·4 all-time
byMatthew Schramm@matthewubundi
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The SKILL.md and README explicitly require the openclaw-cortex plugin, a Cortex API key (CORTEX_API_KEY or openclaw.json config), and CLI commands (openclaw cortex ...). The registry metadata, however, declares no required env vars, binaries, or config paths. This mismatch means the skill will expect access to an external API and CLI tools that are not represented in its declared requirements.
Instruction Scope
Runtime instructions direct the agent to auto-recall before turns, auto-capture after turns, always run multi-query searches before hedging, and — when terminal access exists — execute openclaw cortex CLI commands directly. The SKILL.md also mandates calling cortex_save_memory for many technical details. While these behaviors are consistent with a memory integration, they broaden what the agent is allowed/expected to do (execute CLI, persist many concrete details) and could lead to excessive capture of sensitive information if not constrained.
Install Mechanism
There is no install spec (instruction-only), which is low-risk for code execution. However, README includes explicit installation steps for the openclaw-cortex plugin (openclaw plugin install @ubundi/openclaw-cortex@latest) and config edits. The missing install metadata is an incoherence (the skill requires external plugin installation but does not declare it).
Credentials
The documentation and SKILL.md expect a Cortex API key and local OpenClaw plugin config, yet the skill metadata lists no required environment variables or config paths. The skill also instructs the agent to save detailed technical facts (SQL, CLI commands, config values), which could include secrets if filtering fails. The declared 'What is NOT captured' assurances are not enforceable from the instructions alone.
Persistence & Privilege
always:false and default autonomous invocation are appropriate. However, because the SKILL.md explicitly tells the agent to execute openclaw cortex CLI commands when terminal access is available and to auto-capture/save many concrete details, the agent could perform sensitive persistent actions (writes to the memory backend) during normal operation. This combination increases blast radius but is not itself a configuration error.
What to consider before installing
Key things to consider before installing:
- Missing declarations: SKILL.md and README require the openclaw-cortex plugin and a Cortex API key, but the skill metadata lists no env vars or config paths. Ask the author to declare required env vars (e.g., CORTEX_API_KEY) and any required tools so you can assess permissions before enabling.
- Verify the Cortex backend and plugin: only install the openclaw-cortex plugin from a trusted source and confirm where memories will be stored and who can access them. If you don't trust the Cortex backend, do not provide an API key.
- Limit capture and test in a sandbox: start with autoCapture:false and autoRecall:false, or enable audit logging (/audit on or auditLog:true) before enabling auto-capture. Test behavior in an isolated workspace to confirm that sensitive data (credentials, secrets, file contents) are not being saved.
- Review API key scope and retention: use least-privilege API keys and short-lived keys where possible. Confirm retention, deletion, and export policies for stored memories.
- Be cautious about CLI/terminal execution: the skill tells the agent to execute openclaw cortex commands if terminal access exists. Ensure the agent runtime has limited shell privileges and that you are comfortable with it executing those commands.
- Ask for metadata fixes: request that the skill author update the registry metadata to list required env vars, tools, and config paths, and to explain how secrets are filtered from captures. That change would increase transparency and reduce the suspicion.
If you proceed, enable auditing, restrict API key permissions, and test thoroughly in a non-production environment first.Like a lobster shell, security has layers — review code before you run it.
knowledge-graphvk973jjh9zx39s3tv2yvt5708xn81bh8blatestvk97b3x6cn1c2v03f51fda5zcv5836btnmemoryvk973jjh9zx39s3tv2yvt5708xn81bh8bretrievalvk973jjh9zx39s3tv2yvt5708xn81bh8b
License
MIT-0
Free to use, modify, and redistribute. No attribution required.