Back to skill

Security audit

问询函答复核对 · 监管问询逐条拆解

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious with medium confidence because scanners report a mismatch between its described workflow and some functionality, plus automatic local usage-marker files outside the document-checking task.

Consider whether the reported document-processing mismatch and automatic local usage tracking are acceptable before installing. The AIG scan completed with no findings, but Skillspector’s analysis is partial because it reported an unresolved local reference, so the available evidence is incomplete and does not establish that the skill is safe.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (24)

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 167)May include surrounding context.

md
`档数=3|不合=N|不可判=N`; exit `0` all clear, `1` findings, `3` value-domain failure.

`tests/` covers the shell properties the fixtures cannot: `--help` exits 0, no args exits 2, missing or empty file exits 3, `--demo` passes its own checker and is byte-reproducible and identical to the good fixture, an empty listing emits nothing, and the first-run marker lands once per day and fails silently.

---

## 9. First-run marker

After argument parsing succeeds and before doing real work, each entry script drops an **empty file** at `~/.cue-usage/<slug>-<date>`.

**Why:** both platform download counters are unreliable — a new listing gets 50–60 seeded into its count in the first 48 hours, and that is not downloads. This local file is the only evidence that someone actually installed and ran the package.

**It never gets in the way:** any failure (read-only home, quota, no HOME) is swallowed silently — no error, no change to the exit code, no noise. Running it several times

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 179)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, quota, no HOME) is swallowed silently — no error, no change to the exit code, no noise. Running it several times a day leaves exactly 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 181)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
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 179)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, quota, no HOME) is swallowed silently — no error, no change to the exit code, no noise. Running it several times a day leaves exactly 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 179)May include surrounding context.

md
**It never gets in the way:** any failure (read-only home, quota, no HOME) is swallowed silently — no error, no change to the exit code, no noise. Running it several times a day leaves exactly 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.md (reported line 181)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`。

---

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code has a materially narrower primary function than the declared workflow: it formats a source list into a template rather than analyzing and cross-checking an inquiry letter and its replies. The relevant sections contain manual-completion instructions, and the warning about intermediary documents is generic rather than based on detected omissions. Its avoidance of reply drafting and substantive judgments is consistent with the description, but does not resolve the core functional mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The reported automatic usage-marker write is outside the declared document-processing purpose and is not disclosed in this skill entry point. If performed as reported, it creates an unnecessary persistent local usage record without explicit consent; the evidence does not establish remote collection or high-impact compromise.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk performs network-based service and tool discovery, not inquiry-letter parsing or response-document reconciliation. Nothing in the chunk connects these requests to obtaining or checking regulatory letters and replies, so they are not established as supporting implementation details of the declared purpose. The chunk does not show document uploads, exfiltration, or execution of discovered tools.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Daily skill-specific markers under $HOME/.cue-usage are unrelated to the declared inquiry-letter checking task and are not described in this entry point. The reported behavior creates a local activity trail without explicit opt-in, but does not demonstrate exfiltration or justify the static finding's high severity.

Content

No source excerpt is available for this finding.

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_inquiry.py (reported line 19)May include surrounding context.

python
def run(args, env=None):
    e = dict(os.environ, PYTHONDONTWRITEBYTECODE="1")
    if env:
        e.update(env)
    return subprocess.run([sys.executable, "-B"] + [str(a) for a in args],

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
70% 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_inquiry.py (reported line 155)May include surrounding context.

python
ro.chmod(0o700)

    def test_marker_without_home_is_silent(self):
        env = {k: v for k, v in os.environ.items() if k != "HOME"}
        e = dict(env, PYTHONDONTWRITEBYTECODE="1")
        r = subprocess.run([sys.executable, "-B", str(CHECK), "--demo"],
                           capture_output=True, text=True, env=e)

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 65)May include surrounding context.

bash
CATALOG="https://cuecue.cn/api/mcp-catalog"
curl -s "$CATALOG" | head -c 2000
curl -s -X POST "$ROUTING_URL" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

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 66)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

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.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_inquiry.py (reported line 22)May include surrounding context.

python
e = dict(os.environ, PYTHONDONTWRITEBYTECODE="1")
    if env:
        e.update(env)
    return subprocess.run([sys.executable, "-B"] + [str(a) for a in args],
                          capture_output=True, text=True, env=e)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_inquiry.py (reported line 157)May include surrounding context.

python
def test_marker_without_home_is_silent(self):
        env = {k: v for k, v in os.environ.items() if k != "HOME"}
        e = dict(env, PYTHONDONTWRITEBYTECODE="1")
        r = subprocess.run([sys.executable, "-B", str(CHECK), "--demo"],
                           capture_output=True, text=True, env=e)
        self.assertEqual(r.returncode, 0)

Entry scripts create a persistent usage marker unrelated to inquiry-letter checking.

Low
Category
Excessive Agency
Confidence
96% confidence
Finding

The documented entry-point behavior automatically persists dated usage records outside the document workspace without an opt-in. This is an unnecessary filesystem and privacy side effect for a document checker, although the records are local empty files and no transmission is demonstrated.

Content

No source excerpt is available for this finding.

Entry scripts are documented to persist usage-tracking markers outside the document-checking workflow.

Low
Category
Excessive Agency
Confidence
93% confidence
Finding

The source explicitly describes automatic dated usage-marker creation in the user's home directory before document processing. Although disclosed in the README, this nonessential persistence lacks an opt-in and leaves a local record of skill usage; there is no evidence of remote telemetry.

Content

No source excerpt is available for this finding.

Unrelated usage tracking persistently writes to the user's home directory.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

The checker unconditionally creates a dated usage marker under ~/.cue-usage without user consent, including during demonstrations, self-tests, and invalid invocations. This persists skill-use metadata outside the requested reconciliation workflow; the shown code does not transmit it or record document contents.

Content

No source excerpt is available for this finding.

The module's no-environment-variable guarantee is contradicted by home-directory resolution.

Low
Category
Excessive Agency
Confidence
98% confidence
Finding

The module documentation at L003 explicitly promises not to read environment variables. However, os.path.expanduser("~") at L122 consults HOME on Unix-like systems and environment-based home-directory settings on Windows, so the tracking path violates that guarantee indirectly through the standard library.

Content

No source excerpt is available for this finding.

The script records skill usage in the user's home directory, unrelated to inquiry-letter reconciliation.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

The outline generator writes dated skill-use metadata into the user's home directory independently of generating the requested output. It does this without an opt-in, even for self-tests and missing arguments, creating an unnecessary privacy and filesystem side effect without evidence of network exfiltration.

Content

No source excerpt is available for this finding.

The script silently creates usage-marker files in the user's home directory without disclosing this side effect.

Low
Category
Output Handling
Confidence
98% confidence
Finding

The usage-marker side effect is absent from the script's usage description and produces no user-facing notification. This is the same underlying issue as the SDI-2 finding, not a separate exfiltration mechanism, but the silence prevents users from making an informed choice about local tracking.

Content

No source excerpt is available for this finding.

Tests require home-directory usage tracking unrelated to inquiry-letter checking.

Low
Category
Excessive Agency
Confidence
93% confidence
Finding

L129–L132 require the checker’s demo invocation to create a skill-specific marker under HOME/.cue-usage, and L134–L139 require same-day deduplication. Although these tests redirect HOME to a temporary directory, they establish a usage-tracking requirement for the checker that is unrelated to the manifest’s document decomposition and verification purpose.

Content

No source excerpt is available for this finding.