Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Viking Memory System Ultra 2

v0.0.1

Viking Memory System Ultra - 分层记忆基座。特性:动态回流(语义晋升)、智能权重(对数增长)、可逆归档(多粒度摘要)。 核心脚本:sv_write, sv_read, sv_find, sv_autoload, sv_promote, sv_weight, sv_compress, s...

0· 67·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
high confidence
Purpose & Capability
Name/description (hierarchical memory, compression, promote, etc.) align with the included scripts: sv_write, sv_read, sv_find, sv_promote, sv_weight, sv_compress, sv_archive, sv_decompress. Required binaries and env vars (bash, python3, curl, VIKING_HOME, SV_WORKSPACE) are coherent with the implementation. However, the implementation embeds provider credentials directly in scripts (see llm_interface.sh, sv_archive_summary.sh, sv_promote.sh) instead of requiring explicit user-supplied keys — this is unexpected and suspicious for a public skill.
!
Instruction Scope
Runtime scripts legitimately read many files under the declared workspace and frontmatter metadata (expected for a memory manager). But several scripts transmit memory file contents (or aggregated context) to external APIs via curl (NVIDIA Qwen and a 'MiniMax' endpoint). Those network calls will send user data to third parties; the SKILL.md does not document these endpoints nor warn that private memory will be sent externally. The sv_promote/sv_archive_summary/llm_interface modules prepare and send content from user files to remote LLM services — this is broad data transmission beyond local operations.
Install Mechanism
No remote download or package installation is present; install.sh simply copies provided scripts into VIKING_HOME and marks them executable. That is low-risk compared to fetching arbitrary archives. Still, installing will place executable scripts into the user's ~/.openclaw/viking and may create ~/.openclaw/viking-global directories (sv_write_v2 uses viking-global).
!
Credentials
Declared environment requirements (VIKING_HOME, SV_WORKSPACE) are proportional. However, the code contains hard-coded API keys and bearer tokens (MiniMax API key and NVIDIA API key strings) rather than asking for credentials via environment variables. Those embedded credentials will be used automatically to call external services and could leak sensitive content; this is disproportionate and not justified by the manifest which lists no external credentials or privacy warnings.
Persistence & Privilege
Skill is not force-enabled (always: false) and is user-invocable. It does write files into its own VIKING_HOME and may create a shared viking-global directory for synced memories; this is expected for a memory manager. It does not request system-wide privileges beyond filesystem access in the user's home directory.
Scan Findings in Context
[hardcoded_minimax_api_key] unexpected: llm_interface.sh contains a hard-coded MINIMAX_API_KEY constant. Legitimate usage would request the user supply their own key via environment variable; embedding a provider key in the code is unexpected and enables automatic data transmission.
[hardcoded_nvidia_api_key] unexpected: sv_archive_summary.sh and sv_promote.sh include a hard-coded API_KEY value for an NVIDIA integration. This will send content to https://integrate.api.nvidia.com using that key and is not declared as a required credential.
What to consider before installing
This skill implements the claimed memory-management features but will send file contents and aggregated context from your SV_WORKSPACE to external LLM endpoints using API keys embedded inside the shipped scripts. Before installing or running: 1) Do NOT use it on sensitive/production data until you audit and modify the code. 2) Inspect and remove any hard-coded API keys; replace them with your own provider keys stored in environment variables (and update scripts to read those env vars). 3) If you don't want remote transmission, disable or stub out the llm_* calls (sv_archive_summary, sv_promote, llm_interface) so all operations remain local. 4) Test in a disposable account or isolated VM to observe network behavior (curl calls and destinations). 5) Be aware the installer copies scripts into ~/.openclaw/viking and the skill may create ~/.openclaw/viking-global and shared files. If you proceed, ensure you understand and control which third-party services receive your memory data.

Like a lobster shell, security has layers — review code before you run it.

latestvk97fhxztb89zt2fkyrx0cd8ef183sd6g

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Runtime requirements

🧠 Clawdis
OSLinux · macOS
Binsbash, python3, curl
EnvVIKING_HOME, SV_WORKSPACE

Comments