Huahua Spirits

v1.4.7

灵兽陪伴系统 - 每个用户拥有独特的灵兽伙伴,有性格、有成长、有互动,让 AI 更有温度

0· 105·1 current·1 all-time
bybaiye@baiye1997
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Pending
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description (companionship, deterministic spirit generation) matches the included scripts and files. The scripts implement generation (generate.js), rendering/display (render.js, display.js), soul prompt/save (soul.js), and passive/active behaviors (buddy.js). No unexpected cloud credentials, binaries, or unrelated dependencies are requested.
Instruction Scope
SKILL.md instructs the agent to run local node scripts and to use the LLM to turn the soul.js prompt into a name/personality. That LLM call is explicit in the workflow but not implemented in the shipped code (expected: agent/model call). The skill also instructs the agent to use a user identifier as the seed—this is necessary for deterministic generation but is a privacy-relevant detail. Everything the instructions ask the agent to read/write is limited to the skill directory (assets/companion.json, assets/sprites.json) and provided scripts.
Install Mechanism
No install spec; this is instruction + script based. All code is provided in the package. There are no downloads from external URLs and no extraction/install steps that pull arbitrary binaries.
Credentials
The skill requests no environment variables or external credentials, which is proportional. However, it persistently stores companion data (assets/companion.json) and records short snippets of user messages (hook-react appends up to 50 interactionHistory entries). The skill also requires the agent to provide a user ID (e.g., Feishu open_id, Telegram ID) as a seed for deterministic generation — expected for this feature but privacy-sensitive because the seed links identities to generated artifacts.
Persistence & Privilege
always:false (normal). The skill writes persistent state to its own assets/companion.json (saving name, personality, stats, bond, lastInteraction and an interactionHistory). This is expected for a companion system, but it means the skill will retain user message snippets and bond state on disk across agent runs.
Assessment
This skill appears to do exactly what it claims: local, deterministic ASCII 'spirit' companions. Key points to consider before installing: - Privacy: the skill persistently stores companion data in assets/companion.json inside the skill folder, including bond, lastInteraction timestamps and short snippets of user messages (hook-react records up to 50 items). If you operate in a shared environment, these saved snippets could be visible to others with access to the agent's filesystem. - Identity seed: the recommended seed is a platform user ID (Feishu open_id, Telegram/Discord ID). Passing platform IDs will deterministically bind that identity to a spirit. If you do not want platform IDs saved or linked, use a less-sensitive seed (display name or a hashed identifier under your control). - LLM call: soul.js prompt prints a JSON prompt the agent should send to an LLM to generate name/personality; the package does not perform that network call itself. Ensure your agent sends only the intended prompt and does not include extra secrets when invoking the model. - No network exfiltration: the included scripts (generate.js, render.js, display.js, buddy.js, soul.js) perform only local file I/O and random/stat generation; there are no outgoing network calls or child_process/exec uses in the provided code. If you modify the skill, re-check for external calls. - Per-user isolation: if multiple users share a single agent instance, consider how companion.json will be managed (the skill stores one companion.json in its assets folder). You may want to maintain per-user companion files or ensure the agent uses per-user baseDir separation. If these behaviors (local persistent storage, deterministic binding to a seed ID, and silent recording of short positive-message snippets) are acceptable for your deployment, the skill is coherent and reasonable to install. If you need stronger privacy guarantees, request changes (per-user files, encryption, disable hook-react recording, or use ephemeral seeds).

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

latestvk97c2wt4bk9sah0mg37twb4bs184936v

License

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

Runtime requirements

🐱 Clawdis
OSmacOS · Linux · Windows

Comments