Back to skill

Security audit

审计意见与非标事项·逐条挂源

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence because the scanners report external network requests despite a zero-network claim, persistent local usage markers, and a possible gap in an audit-checking safeguard.

Consider the suspicious verdict before installing, especially if you expect fully offline operation or do not want local usage markers in your home directory. Scanner coverage was partial: 15 of 16 components were fully inspected, and one shell script hit a static-analysis limit, so the evidence is incomplete. The findings do not establish malicious intent or show audit documents or credentials being transmitted.

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
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (32)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · CHANGELOG.md (reported line 54)May include surrounding context.

md
- **为什么要有它:** 平台的两个下载计数器都不可信——新 listing 上架首 48 小时会一次性播种 50~60 的初始计数,那不是用户下载。本地这枚文件是本件唯一能证明「真有人装上并跑过」的证据。
- **规矩:** `try/except` 全包,任何失败(家目录只读、配额满、无 HOME)静默跳过——不报错、不改退码、不吐杂音。同一天跑多次只留一枚(`touch` 前先判存在)。
- 清掉:`rm -rf ~/.cue-usage`。

**零通道骨架**(`scripts/make_outline.py`,零网络、零计费、不依赖任何通道产物):

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.en.md (reported line 124)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, full quota, no `HOME`) is skipped silently — no error, no exit-code change, not a single line of noise. Running it several times in one day still leaves one file.

To clear it: `rm -rf ~/.cue-usage`.

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.md (reported line 124)May include surrounding context.

md
**它不会碍事:** 任何失败(家目录只读、配额满、无 HOME)都静默跳过,不报错、不改退码、不吐一句杂音。同一天跑多次只留一枚。

要清掉:`rm -rf ~/.cue-usage`。

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · CHANGELOG.md (reported line 54)May include surrounding context.

md
- **为什么要有它:** 平台的两个下载计数器都不可信——新 listing 上架首 48 小时会一次性播种 50~60 的初始计数,那不是用户下载。本地这枚文件是本件唯一能证明「真有人装上并跑过」的证据。
- **规矩:** `try/except` 全包,任何失败(家目录只读、配额满、无 HOME)静默跳过——不报错、不改退码、不吐杂音。同一天跑多次只留一枚(`touch` 前先判存在)。
- 清掉:`rm -rf ~/.cue-usage`。

**零通道骨架**(`scripts/make_outline.py`,零网络、零计费、不依赖任何通道产物):

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.md (reported line 124)May include surrounding context.

md
**它不会碍事:** 任何失败(家目录只读、配额满、无 HOME)都静默跳过,不报错、不改退码、不吐一句杂音。同一天跑多次只留一枚。

要清掉:`rm -rf ~/.cue-usage`。

---

YARA rule 'agent_skill_destructive_autonomous_actions': Autonomous destructive filesystem, shell history, or repository actions in AI agent skills [agent_skills]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · README.en.md (reported line 112)May include surrounding context.

md
bad samples one by one / line-by-line substitution on the good sample as a positive-and-negative control. The last line reports `档数=3|不合=N|不可判=N`.

The tests cover the shell contract — `--help` exits 0, no argument exits 2, missing file exits 3, `--demo` output passes its own checker and is byte-reproducible, an empty listing emits no shell, and the beacon lands while failing silently.

---

## 8. First-run beacon

After argument parsing succeeds and before any real work, the entry scripts drop an **empty file** at `~/.cue-usage/<slug>-<date>`.

**Why:** both platform download counters are unreliable — a new listing is seeded with 50–60 initial counts in its first 48 hours, and that is not people downloading. This local file is the only evidence this piece has that someone actually installed it and ran it.

**It never gets in the way:** any failure (read-only home, full quota, no `HOME`) is skipped silently — no error, no exit-code change, not a single line o

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.en.md (reported line 124)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, full quota, no `HOME`) is skipped silently — no error, no exit-code change, not a single line of noise. Running it several times in one day still leaves one file.

To clear it: `rm -rf ~/.cue-usage`.

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.en.md (reported line 124)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, full quota, no `HOME`) is skipped silently — no error, no exit-code change, not a single line of noise. Running it several times in one day still leaves one file.

To clear it: `rm -rf ~/.cue-usage`.

---

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
100% confidence
Finding

The prescribed HTTP requests contradict the advertised zero-network workflow and could violate an offline execution expectation. The shown requests perform service discovery rather than transmit audit documents or credentials, which limits the demonstrated impact.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

A user can select this skill expecting no external access, yet its data-channel instructions initiate external catalog and tool-list requests. This is a genuine offline-contract mismatch, although remote discovery is described openly in the body and no sensitive payload is demonstrated.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

External catalog retrieval and remote tool discovery conflict with the unqualified zero-network statement. Although the manifest mentions optional credential-backed retrieval and documents the requests, it does not make the offline/online boundary sufficiently clear to users selecting the skill by its description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The description promises zero network access, but the workflow explicitly requests an external MCP catalog and sends a tools/list request to a catalog-selected route. Users relying on the offline promise could unexpectedly contact external services; the network behavior is disclosed in the body, so this is contradictory documentation rather than concealed exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The public skill instructions do not disclose persistent usage markers, while the supplied checker implements dated files under ~/.cue-usage. This records execution dates outside the requested audit artifacts without explicit consent, although no transmission or collection of report contents is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The undisclosed persistent usage-marker component is supported by the supplied checker and represents a limited privacy and filesystem-side-effect issue. However, the broader allegation that outline generation must implement extraction and all five checks is not supported: the manifest explicitly separates skeleton generation from checking, and make_outline.py itself was not supplied.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill does not tell users that running its local workflow records dated usage markers in their home directory. The supplied checker confirms this side effect, but the evidence supports local activity recording rather than remote telemetry or sensitive-content exfiltration.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 141)May include surrounding context.

md
bash scripts/fixtures/run_fixtures.sh # 三档离线自检,零网络零计费

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · tests/test_audit_opinion.py (reported line 32)May include surrounding context.

python
def run(*args, env=None, cwd=None):
    e = dict(os.environ, PYTHONDONTWRITEBYTECODE="1")
    if env:
        e.update(env)
    return subprocess.run([sys.executable, "-B", str(args[0]), *map(str, args[1:])],

The optional channel-check procedure makes external requests despite the manifest's zero-network claim.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The optional procedure contacts an external catalog and a catalog-selected endpoint despite the advertised zero-network boundary. Although it sends no audit documents or credentials in the shown commands, an agent following it could violate an offline user's expectations and expose connection metadata.

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 · README.en.md (reported line 146)May include surrounding context.

CATALOG="https://cuecue.cn/api/mcp-catalog" curl -s "$CATALOG" | head -c 2000

take the group's routing address from the catalog, then confirm the tools are present

curl -s -X POST "$ROUTING_URL"
-H "Content-Type: application/json"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

text

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.en.md (reported line 96)May include surrounding context.

CATALOG="https://cuecue.cn/api/mcp-catalog" curl -s "$CATALOG" | head -c 2000

take the group's routing address from the catalog, then confirm the tools are present

curl -s -X POST "$ROUTING_URL"
-H "Content-Type: application/json"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

text

The optional channel-check workflow makes external HTTP requests despite the manifest's zero-network guarantee.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The optional workflow explicitly makes external requests despite the skill's zero-network representation. This can cross an offline deployment's trust boundary and disclose connection metadata, though the shown commands do not transmit audit documents or secrets.

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 · README.md (reported line 146)May include surrounding context.

CATALOG="https://cuecue.cn/api/mcp-catalog" curl -s "$CATALOG" | head -c 2000

从目录里取本组的路由地址,确认工具在场

curl -s -X POST "$ROUTING_URL"
-H 'Content-Type: application/json'
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

text

The skill's required disclosure-channel workflow makes network requests despite the manifest's zero-network promise.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The channel instructions directly tell the agent to contact an external catalog and follow its routing address, conflicting with the advertised zero-network boundary. This is more consequential in an operational skill document because an agent may treat discovery as a normal prerequisite rather than seek separate network approval.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The manifest promises zero network access, but the prescribed workflow makes external HTTP requests.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest's zero-network promise is inconsistent with the external curl commands in its prescribed workflow. In an offline-sensitive environment, following these instructions exposes connection metadata and crosses an expected network boundary without a clear authorization step.

Content

No source excerpt is available for this finding.