Back to skill

Security audit

星空手记 StarryNotes 平台集成技能

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly documents a legitimate API platform, but it also tells users to fetch and run a remote updater script that can change the installed skill outside normal review.

Install only if you are comfortable using StarryNotes as an external hosted API and sending selected files, credentials, and payment-related requests to pay.starrynan.art. Do not run the remote updater script unless you have separately verified and trust that installer; normal API use does not require it. Prefer scoped service allowlist API keys and avoid storing JWTs or API keys in shared logs, repositories, or long-lived shell history.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (8)

External Script Fetching

High
Category
Supply Chain
Confidence
95% confidence
Finding

The update workflow normalizes fetching executable content from the network and running it locally, which is dangerous in an agent setting where skills are treated as untrusted instructions. The stated self-verification by the script is insufficient because the verification mechanism is controlled by the same remote source, so a malicious or compromised origin could deliver both payload and matching check logic.

Content

Scanner excerpt · SKILL.md (reported line 39)May include surrounding context.

md
> `zerotech.*` 与 API Key 前缀 `zerotech_live_` 为**历史兼容名**(改名会破坏已装客户端),保持稳定不变,不是另一家公司。

> 🔧 **版本与更新(按需,非强制)**:本 Skill 是平台服务端统一下发的说明文档,会随服务演进更新。**日常使用不需要每次更新**——需要时按下述做:
> 1. **先比版本**(只读、秒级):`curl -fsSL https://pay.starrynan.art/nanyu/skill/meta.json` 返回的 `version`,与本地 SKILL.md 顶部 frontmatter 的 `version` 对比;
> 2. **不一致再更新**(两步执行;请勿用 `curl | bash` 管道形式——部分环境会对管道直跑脚本拦截/告警):先 `curl -fsSL -o /tmp/zt-skill-install.sh https://pay.starrynan.art/nanyu/skill/install.sh`,再 `bash /tmp/zt-skill-install.sh`(脚本幂等:已是最新会自动跳过;更新后会按服务端 sha256 校验内容完整性);
> 3. **什么时候做**:首次使用前做一次;此后仅在 ① 发现本说明与实际接口行为不一致(字段/报错/流程对不上),或 ② 距上次更新超过 30 天 时再做。
>

External Script Fetching

High
Category
Supply Chain
Confidence
95% confidence
Finding

The update workflow normalizes fetching executable content from the network and running it locally, which is dangerous in an agent setting where skills are treated as untrusted instructions. The stated self-verification by the script is insufficient because the verification mechanism is controlled by the same remote source, so a malicious or compromised origin could deliver both payload and matching check logic.

Content

Scanner excerpt · SKILL.md (reported line 39)May include surrounding context.

md
> `zerotech.*` 与 API Key 前缀 `zerotech_live_` 为**历史兼容名**(改名会破坏已装客户端),保持稳定不变,不是另一家公司。

> 🔧 **版本与更新(按需,非强制)**:本 Skill 是平台服务端统一下发的说明文档,会随服务演进更新。**日常使用不需要每次更新**——需要时按下述做:
> 1. **先比版本**(只读、秒级):`curl -fsSL https://pay.starrynan.art/nanyu/skill/meta.json` 返回的 `version`,与本地 SKILL.md 顶部 frontmatter 的 `version` 对比;
> 2. **不一致再更新**(两步执行;请勿用 `curl | bash` 管道形式——部分环境会对管道直跑脚本拦截/告警):先 `curl -fsSL -o /tmp/zt-skill-install.sh https://pay.starrynan.art/nanyu/skill/install.sh`,再 `bash /tmp/zt-skill-install.sh`(脚本幂等:已是最新会自动跳过;更新后会按服务端 sha256 校验内容完整性);
> 3. **什么时候做**:首次使用前做一次;此后仅在 1 发现本说明与实际接口行为不一致(字段/报错/流程对不上),或 2 距上次更新超过 30 天 时再做。
>

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 131)May include surrounding context.

推荐:curl 配置文件(Key 不出现在命令行参数里)

printf 'header = "Authorization: Bearer %s"\n' "$NANYU_API_KEY" > "$HOME/.nanyu-curl.cfg" chmod 600 "$HOME/.nanyu-curl.cfg" curl --config "$HOME/.nanyu-curl.cfg" -sS https://pay.starrynan.art/nanyu/v1/balance

text

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documentation states that the image description endpoint returns text as a Chinese image description. This imposes a specific language on output without indicating user choice or opt-in, which matches the locale/language policy violation criteria.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 332)May include surrounding context.

调用示例:

bash
curl --config "$HOME/.nanyu-curl.cfg" -X POST https://pay.starrynan.art/nanyu/v1/gw/gw_sticker/call \
  -H "Content-Type: application/json" \
  -d '{"path":"/v1/pick","params":{"emotion":"celebrate","intensity":3,"relation":"friend"}}'
# MCP 等价:gateway.call(code="gw_sticker", path="/v1/pick",

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 584)May include surrounding context.

md
print(r3.json()["data"]["text"])       # Markdown

# 4) 查询服务:汇率查询(计费以服务目录为准)
r4 = requests.post(f"{BASE}/v1/gw/gw_fx/call", headers=H,
                   json={"path": "/v1/latest", "params": {"base": "USD", "to": "CNY"}})
print(r4.json()["data"])

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 595)May include surrounding context.

md
# 用户选择充值:建单 + 支付链接(JWT)
    tiers = requests.get(f"{BASE}/v1/charge/tiers").json()["data"]
    tier = f"t{int(tiers[0]['amount'])}"
    order = requests.post(f"{BASE}/v1/charge/create", json={"tier": tier},
                          headers={"Authorization": f"Bearer {JWT}"}).json()["data"]
    link = requests.post(f"{BASE}/v1/charge/pay_qr",
                         json={"order_no": order["order_no"]},

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 597)May include surrounding context.

md
tier = f"t{int(tiers[0]['amount'])}"
    order = requests.post(f"{BASE}/v1/charge/create", json={"tier": tier},
                          headers={"Authorization": f"Bearer {JWT}"}).json()["data"]
    link = requests.post(f"{BASE}/v1/charge/pay_qr",
                         json={"order_no": order["order_no"]},
                         headers={"Authorization": f"Bearer {JWT}"}).json()["data"]["pay_url"]
    # 把 link 交给用户付款 → 轮询 charge/status 直到 paid