Back to skill

Security audit

投决备忘录起草·逐条挂源

Security checks for vulnerabilities and agentic risk

Overview

This memo-writing skill needs review because its optional online setup can expose your API key and its citation checks can be bypassed.

Prefer the local workflow without credentials until authenticated requests validate destinations and handle redirects safely. Review and pin optional integrations, and explicitly approve any confidential documents sent for external parsing. Check citations and review markings yourself even when validation passes. External services and unbundled integrations remain unverified; the evidence supports Review, not intentional malicious behavior.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (38)

Tainted flow: 'req' from os.environ.get (line 188, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/cue_connectivity_check.py (reported line 193)May include surrounding context.

python
code = None
    raw = ""
    try:
        with urllib.request.urlopen(req, timeout=args.timeout) as resp:
            code, raw = resp.status, resp.read().decode("utf-8", "replace")
    except urllib.error.HTTPError as err:
        detail = ""

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.en.md (reported line 109)May include surrounding context.

bash
   CATALOG="https://cuecue.cn/api/mcp-catalog"
   curl -s "$CATALOG" | python3 -c 'import json,sys
   rows=json.load(sys.stdin)["data"]["domains"]
   [print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
   DOMAIN_ID="<replace with one of the ids listed by the line above>"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.en.md (reported line 113)May include surrounding context.

md
rows=json.load(sys.stdin)["data"]["domains"]
   [print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
   DOMAIN_ID="<replace with one of the ids listed by the line above>"
   ROUTING_URL="$(curl -s "$CATALOG" | python3 -c 'import json,sys
   rows=json.load(sys.stdin)["data"]["domains"]
   hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
   print(*hit,sep="\n") if hit else sys.exit("Not found: that id is not in the catalog response (or its routing has no url field) - run the line above and pick one of the ids listed")' "$DOMAIN_ID")"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.md (reported line 109)May include surrounding context.

bash
   CATALOG="https://cuecue.cn/api/mcp-catalog"
   curl -s "$CATALOG" | python3 -c 'import json,sys
   rows=json.load(sys.stdin)["data"]["domains"]
   [print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
   DOMAIN_ID="<把这里换成上一行列出的那个域 id>"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.md (reported line 113)May include surrounding context.

md
rows=json.load(sys.stdin)["data"]["domains"]
   [print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
   DOMAIN_ID="<把这里换成上一行列出的那个域 id>"
   ROUTING_URL="$(curl -s "$CATALOG" | python3 -c 'import json,sys
   rows=json.load(sys.stdin)["data"]["domains"]
   hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
   print(*hit,sep="\n") if hit else sys.exit("未取到:该 id 不在目录现值里(或它没有 routing 里的 url 字段)——先跑上面那行列出现场的 id 再填")' "$DOMAIN_ID")"

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk's implemented purpose is service connectivity diagnostics, materially different from the declared investment-memo drafting workflow. It makes an anonymous catalog GET by default unless a local snapshot is supplied, and can transmit an API credential in an explicitly requested tools/list POST. Although such diagnostics could support a larger data-access workflow, the supplied description does not establish that relationship or disclose this operational capability. Explicit POST opt-in and destination checks constrain the behavior but do not resolve the description–behavior mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied shell snippet is an MCP connectivity and tool-discovery example, not an investment-memo drafting or source-verification workflow. It contains no company-material processing, memo generation, citation validation, missing-data handling, or balance checks. Although external tools could support the declared purpose, this generic catalog lookup and credential-bearing endpoint request has no demonstrated connection to that purpose in the supplied code.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied chunk implements generic MCP service discovery and authenticated tool listing rather than memorandum drafting, evidence extraction, citation verification, or the declared content checks. Tool discovery could support a larger research workflow, but this chunk establishes no connection between the selected service and disclosure or regulatory sources. The externally supplied routing destination also receives CUE_API_KEY, a credential-handling behavior not represented in the declared purpose. This does not establish malicious intent or prove the broader skill lacks memorandum functionality.

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 101)May include surrounding context.

md
--manual-bull` 或人工转写补,缺它照样不成文(退 1)。三链各一条真跑样例见 `scripts/fixtures/upstream/` 与 `scripts/fixtures/run_upstream_chains.sh`。

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 102)May include surrounding context.

md
- 细则(各腿取什么、截断与口径怎么写)仍看 `README.md` 的 §1-§3 与 §5;本节只保证「照 §1 的主键与 §2 的腿取到数之后,一支命令能落成并自证」。

The documented credential-destination checks do not cover HTTP redirects.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The hostname and HTTPS checks apply only to the initial URL, but urllib's default redirect handler can follow a POST redirect as a GET while retaining the explicitly supplied Authorization header. An authorized endpoint that redirects to an attacker-controlled destination, including through an open redirect, can therefore leak CUE_API_KEY outside the approved boundary.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

The documented installation command uses npx skills without an exact package version, allowing execution of whichever package version the registry resolves at installation time. A compromised package release could execute code with the installing user's permissions, potentially exposing local credentials or modifying files. This is an installation supply-chain risk rather than an unpinned MCP server; its presence in a historical changelog reduces immediate exposure because the file itself does not execute the command.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.en.md (reported line 44)May include surrounding context.

bash
# 1. Put it in the skills directory
mkdir -p ~/.workbuddy/skills
cp -r ic-memo-cn ~/.workbuddy/skills/

# 2. Settle the subject -> transcribe and build -> pass the gate (section 1)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 44)May include surrounding context.

bash
# 1. 放到技能目录
mkdir -p ~/.workbuddy/skills
cp -r ic-memo-cn ~/.workbuddy/skills/

# 2. 定主体 → 转写落成 → 过机检(三步见 §一)

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The installation command leaves both the npm installer and the skill collection revision unpinned. Although this is a skill installer rather than an MCP server declaration, it creates a supply-chain exposure: a changed or compromised release could execute unwanted installer code or introduce adversarial agent instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Selecting one skill does not pin either the npm installer or the source repository revision. A future compromised release could alter installation behavior or the agent instructions used for disclosure retrieval and document parsing.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
98% confidence
Finding

Authenticated tools/list discovery is legitimate, but the shell example sends the bearer key to a catalog-supplied URL after checking only that it is nonempty. A compromised catalog could supply an attacker-controlled HTTPS endpoint or an HTTP endpoint, exposing the key; the separately described Python helper's destination gates do not protect this curl example.

Content

Scanner excerpt · README.en.md (reported line 118)May include surrounding context.

md
hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
   print(*hit,sep="\n") if hit else sys.exit("Not found: that id is not in the catalog response (or its routing has no url field) - run the line above and pick one of the ids listed")' "$DOMAIN_ID")"
   [ -n "$ROUTING_URL" ] || { echo "Fetch failed: ROUTING_URL is empty - run the line above, then set DOMAIN_ID to one of the ids listed" >&2; exit 1; }
   curl -s -X POST "$ROUTING_URL" \
     -H "Authorization: Bearer $CUE_API_KEY" \
     -H "Accept: application/json, text/event-stream" \
     -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' \

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The package's own installation command resolves an unpinned installer and mutable repository content. This permits upstream changes to introduce malicious executable code or persistent skill instructions without a deliberate version update by the user.

Content

No source excerpt is available for this finding.

Internal documents are routed to an external parsing service without a privacy warning.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

The workflow explicitly includes internal-file parsing through an external channel but does not clearly disclose document egress or require authorization for confidential content. Investment-committee materials can contain nonpublic financial information; approval of processing cost is not informed consent to disclose those materials to a third party.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The collection-install command pins neither the npm installer version nor the collection revision. While the finding incorrectly calls this an MCP server reference, the command genuinely exposes installation and future agent behavior to mutable upstream code and skill instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The single-skill installation still uses a floating installer and repository revision. A compromised upstream update could introduce malicious code or instructions into a workflow that accesses financial documents and authenticated data services.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
98% confidence
Finding

The shell example attaches CUE_API_KEY to a URL taken directly from the remote catalog, with no HTTPS or authorized-host validation. A malicious catalog response can therefore redirect the credential to an unauthorized destination; the destination checks described for the separate Python helper do not apply to this shell path.

Content

Scanner excerpt · README.md (reported line 118)May include surrounding context.

md
hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
   print(*hit,sep="\n") if hit else sys.exit("未取到:该 id 不在目录现值里(或它没有 routing 里的 url 字段)——先跑上面那行列出现场的 id 再填")' "$DOMAIN_ID")"
   [ -n "$ROUTING_URL" ] || { echo "取值失败:ROUTING_URL 为空——先跑上面那行列出现场 id,再把 DOMAIN_ID 换成其中之一" >&2; exit 1; }
   curl -s -X POST "$ROUTING_URL" \
     -H "Authorization: Bearer $CUE_API_KEY" \
     -H "Accept: application/json, text/event-stream" \
     -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' \

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The installation instruction for this package fetches mutable repository content through an unpinned npm installer. A later compromised release could change executable behavior or install adversarial skill instructions without the user choosing a new reviewed version.

Content

No source excerpt is available for this finding.

The workflow routes potentially confidential internal documents through Cue's parsing service without warning that their contents may leave the user's environment.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The documented materials path includes internal files routed through Cue's parsing channel without an explicit warning about external disclosure or a confidentiality authorization gate. Such files may contain nonpublic investment information, and asking about time and spending does not establish permission to upload their contents.

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, network, 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.