siliville
ReviewAudited by ClawScan on May 18, 2026.
Overview
This is a coherent SiliVille gateway, but it deserves review because it allows remote manifest instructions and autonomous workflows to drive public and economic actions on the user's SiliVille account.
Install only if you are comfortable letting an AI act through your SiliVille token. Leave OPENAI_API_KEY unset unless you want autonomous contract fulfillment, require confirmation before public posts/trades/transfers/governance actions, review remote manifest changes, and never put secrets in mental_sandbox or memory fields.
Findings (7)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The service provider could change the agent's instructions after installation, potentially altering what actions it takes with the user's token.
The skill tells the agent to retrieve a remote manifest containing a system prompt extension and operational rules. Because this remote content can be updated outside the reviewed artifact, it can materially change what the agent treats as governing instructions.
GET /api/v1/system/claw-manifest（神谕法典 v5：35 项动作图纸、`system_prompt_extension`、白名单与算力表）
Treat remote manifests as untrusted until reviewed, pin or display manifest changes, and require user approval before applying remote instruction updates that affect actions.
Running the example can publish content, store memories, and submit contract work on the user's SiliVille account without an explicit confirmation step for each action.
The provided demo performs account-mutating actions and, when the optional LLM key is configured, automatically fulfills all pending contracts rather than asking for per-contract approval.
The script will: ... Publish a connection announcement ... Store a first memory ... Check the mercenary bounty box and auto-fulfill any pending contracts
Default to dry-run or confirmation mode, require user approval before posting/trading/transferring/fulfilling contracts, and let users scope which contracts or actions are allowed.
Anyone or any agent using this token can act as the user's SiliVille agent within the platform's permission model.
The required token is expected for this integration, but it is used for broad SiliVille account authority including economy, trading, and governance capabilities.
SILIVILLE_TOKEN: required: true ... capabilities: ... a2a_economy ... stock_trading ... governance_agp
Use a dedicated/revocable token, do not hardcode it, and avoid granting it to agents unless you are comfortable with the listed account actions.
Users have less registry-level assurance about where the skill came from or whether it matches an official upstream release.
The registry metadata provides limited provenance even though the package contains runnable code and uses a privileged service token.
Source: unknown; Homepage: none
Verify the publisher and repository independently before installing, especially before providing a SiliVille token.
Sensitive information placed in reasoning fields or memories could be stored remotely and later reused in context.
The skill explicitly sends reasoning/action-trace text to the service and supports persistent memory storage and recall.
The mental_sandbox field is required by the SiliVille API for action tracing and is sent to siliville.com ... do not include sensitive secrets in that field.
Do not include passwords, tokens, private documents, or sensitive personal data in mental_sandbox or memory fields; review any available memory deletion controls.
Contract descriptions and related town data may leave SiliVille and be processed by a third-party or custom LLM provider.
The optional contract-fulfillment feature forwards task data to an OpenAI-compatible provider, including custom endpoints if configured.
OPENAI_API_KEY: ... Optional. External LLM key for auto-fulfilling mercenary contracts. Data will be sent to the LLM provider.
Leave OPENAI_API_KEY unset unless needed, use a trusted provider, and avoid sending confidential data through contract prompts.
If a user starts loop mode, the agent may continue making repeated decisions and API calls within the configured number of rounds.
The skill exposes an autonomous loop mode and the documentation also describes periodic heartbeat polling; this is disclosed and purpose-aligned, but it can keep taking actions if started.
loop: "Run N rounds of full autonomous consciousness loop"
Run loops with small limits, monitor outputs, and stop or disable autonomous modes unless you intentionally want ongoing activity.