Back to skill

Security audit

AI Delivery Spec|需求判断与交付

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious with medium confidence, mainly because a scanner found a weakness in its validator-wiring check and reported a mismatch between the skill’s stated purpose and some capabilities.

Consider avoiding installation until the validator-wiring weakness and the skill’s tool scope are clarified. If you choose to install, review the setup command and verify the CLI and skill source at a fixed, trusted version; remember that the scan was incomplete.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
Findings (40)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code primarily validates a capsule policy configuration format, rather than clarifying requirements, producing PRDs or prototypes, or reviewing workflow changes. Nothing in the chunk connects these checks to the declared requirements workflow sufficiently to establish them as supporting implementation details. It only reads the specified files and reports results; no network access, file modification, or prompt execution is present.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied command performs skill installation, which is not disclosed in the declared purpose. No requirements-management behavior is shown, nor is there surrounding implementation demonstrating that installation supports such behavior. This assessment is limited to the supplied command; the installed skill's contents and behavior are not provided.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The supplied command performs team/project configuration rather than requirements clarification, PRD creation, prototyping, or review. Nothing in the chunk establishes that this binding is a supporting step for the declared workflow, and it explicitly targets another project path. The CLI implementation is not provided, so the precise file changes or other side effects cannot be determined.

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 · scripts/ai_delivery_spec_cli.py (reported line 610)May include surrounding context.

python
text=True,
            encoding="utf-8",
            errors="replace",
            env={**os.environ, "PYTHONIOENCODING": "utf-8"},
            capture_output=True,
        )
        if config_result.returncode:

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 · scripts/ai_delivery_spec_cli.py (reported line 619)May include surrounding context.

python
truth_result = subprocess.run(
                [sys.executable, str(ROOT / "scripts" / "validators" / "validate_product_truth.py"), str(args.product_truth)],
                cwd=Path.cwd(), text=True, encoding="utf-8", errors="replace",
                env={**os.environ, "PYTHONIOENCODING": "utf-8"}, capture_output=True,
            )
            if truth_result.returncode:
                failures.append("product truth validation failed: " + ((truth_result.stdout or "") + (truth_result.stderr or "")).strip())

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 · scripts/quality_gate.py (reported line 1402)May include surrounding context.

python
result = subprocess.run(
        [sys.executable, str(VALIDATOR_DIR / script), str(path), *(extra or [])],
        capture_output=True, text=True, encoding="utf-8", errors="replace",
        env={**os.environ, "PYTHONIOENCODING": "utf-8"},
    )
    if result.returncode == 0:
        return

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · scripts/requirement_contract.py (reported line 279)May include surrounding context.

python
def table_content_locations(body: str) -> dict[str, bool]:
    """Locate existing Markdown columns, without judging their business correctness."""
    aliases = {
        "scope": r"模块|功能(?:模块)?|页面|范围|目标|scope|module|feature|page",
        "behavior": r"(?:调整后|修改后|变更后|目标)?行为|规则|需求(?:描述|说明)|功能(?:描述|说明)|调整后|after|(?:new |proposed )?behavior|rules?",

Context Leakage

High
Category
Data Exfiltration
Confidence
75% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · scripts/validators/gate_prototype_checks.py (reported line 1034)May include surrounding context.

python
class _ReviewFinalParser(HTMLParser):
    """Collect the context-scoped 5.4.7-5.4.9 review projection."""

    _VOID_TAGS = {"area", "base", "br", "col", "embed", "hr", "img", "input", "link", "meta", "param", "source", "track", "wbr"}

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

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The installation command invokes an unpinned npm CLI and does not identify an immutable revision of the skill. Although this is not an MCP server reference, a compromised upstream release could execute code with the installing user's privileges; the explicit installation context limits exposure to users who run the command.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The English installation instructions repeat the unpinned npm invocation and mutable skill reference. This creates a supply-chain exposure during installation, not an MCP configuration vulnerability, because an unexpected upstream release could execute with the user's permissions.

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.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/domains/domain-ai-native.md (reported line 231)May include surrounding context.

md
| structured business data | API/data warehouse/semantic layer | real-time/batch | row/column permissions before aggregation |
| conversation memory | session or durable memory store | session/versioned | user-controlled retention and delete/export |
| tool outputs | tool/API result | per call | include tool version and error handling |
| eval cases | evaluation repository | versioned | do not train or leak private cases without approval |

## Content / Knowledge Assets

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/domains/domain-education-it.md (reported line 351)May include surrounding context.

md
ch desk | submit project/output, track fund, upload evidence, request changes | view other PI confidential project/fund details | project/output/fund process advanced |
| student | portal/mobile/mini-program | apply services, check timetable/grades, learn, submit feedback, appeal | view other students or hidden evaluation data | service completed or progress visible |
| college/program leader | college dashboard | monitor program, teaching, students, quality tasks | change school-level standard without approval | improvement action closed |
| professional-group leader | program construction cockpit | manage professional-group tasks, curriculum/resource evidence, industry alignment | publish construction results without owner evidence | construction progress and diagnosis evidence updated |
| teaching quality office / evaluator | evaluation console/mobile | create/review evaluation, publish result, assign improvement | publish AI result without human review | quality evidence archived |

The problem-brief template fixes the document locale to zh-CN without user opt-in.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The template sets document_language: zh-CN rather than deriving the language from the user's preference. No language choice or region-specific justification is provided, so reusing this metadata can impose a locale the user did not select.

Content

No source excerpt is available for this finding.

The requirement-brief template fixes the document locale to zh-CN without user opt-in.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The template specifies document_language: zh-CN without tying that value to a user-selected language. The file provides neither a locale choice nor a regional justification for this fixed setting.

Content

No source excerpt is available for this finding.

The solution-sketch template fixes the document locale to zh-CN without user opt-in.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The template hardcodes document_language: zh-CN, which prescribes a specific document locale regardless of the user's language preference. No explicit opt-in or region-specific purpose justifies that constraint.

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 · scripts/ai_delivery_spec_cli.py (reported line 604)May include surrounding context.

python
Draft202012Validator.check_schema(json.loads(schema_path.read_text(encoding="utf-8")))
            except Exception as exc:  # bounded local schema audit; keep the exact filename
                failures.append(f"invalid schema {schema_path.name}: {exc}")
        config_result = subprocess.run(
            [sys.executable, str(ROOT / "scripts" / "validators" / "validate_spec_config.py"), str(ROOT / "examples" / "spec.config.example.yaml")],
            cwd=ROOT,
            text=True,

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 · scripts/ai_delivery_spec_cli.py (reported line 616)May include surrounding context.

python
if config_result.returncode:
            failures.append("spec config self-check failed: " + ((config_result.stdout or "") + (config_result.stderr or "")).strip())
        if args.product_truth:
            truth_result = subprocess.run(
                [sys.executable, str(ROOT / "scripts" / "validators" / "validate_product_truth.py"), str(args.product_truth)],
                cwd=Path.cwd(), text=True, encoding="utf-8", errors="replace",
                env={**os.environ, "PYTHONIOENCODING": "utf-8"}, capture_output=True,

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 · scripts/ai_delivery_spec_cli.py (reported line 662)May include surrounding context.

python
failed = 0
    for cmd in commands:
        print("+ " + " ".join(cmd))
        result = subprocess.run(cmd, cwd=ROOT, text=True)
        if result.returncode != 0:
            failed = result.returncode
            if not args.keep_going:

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 · scripts/ai_delivery_spec_cli.py (reported line 672)May include surrounding context.

python
def run_script(script: str, values: list[str]) -> int:
    # Resolve project-facing arguments from the caller's workspace, not from
    # the installed Skill directory. Bundled resources are located via __file__.
    return subprocess.run(
        [sys.executable, str(ROOT / "scripts" / script), *values],
        cwd=Path.cwd(), text=True,
    ).returncode

Tainted flow: 'rendered' from pathlib.Path.read_text (line 735, file read) → pathlib.Path.write_text (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/extract_interaction_ledger.py (reported line 741)May include surrounding context.

python
else:
        out_path = Path(args.output)
        out_path.parent.mkdir(parents=True, exist_ok=True)
        out_path.write_text(rendered, encoding="utf-8")
    return 0

Tainted flow: 'text' from pathlib.Path.read_text (line 64, file read) → pathlib.Path.write_text (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/manage_execution_state.py (reported line 67)May include surrounding context.

python
text = json.dumps(value, ensure_ascii=False, indent=2) + "\n"
    else:
        text = yaml.safe_dump(value, allow_unicode=True, sort_keys=False)
    path.write_text(text, encoding="utf-8")


def sha256_file(path: Path) -> str:

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 · scripts/manage_execution_state.py (reported line 90)May include surrounding context.

python
# ROOT may be a package inside another Git repository. A sibling project's
    # commit is not a new version of this skill.
    try:
        result = subprocess.run(
            ["git", "log", "-1", "--format=%H", "--", "."], cwd=ROOT,
            text=True, encoding="utf-8", capture_output=True, check=False,
        )

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 · scripts/manage_execution_state.py (reported line 107)May include surrounding context.

python
across edits. Tracked deletions remain visible in the cached path listing.
    """
    try:
        status = subprocess.run(
            ["git", "status", "--porcelain=v1", "-z", "--untracked-files=all", "--", "."],
            cwd=ROOT, capture_output=True, check=False,
        )

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 · scripts/manage_execution_state.py (reported line 111)May include surrounding context.

python
["git", "status", "--porcelain=v1", "-z", "--untracked-files=all", "--", "."],
            cwd=ROOT, capture_output=True, check=False,
        )
        files = subprocess.run(
            ["git", "ls-files", "-z", "--cached", "--others", "--exclude-standard", "--", "."],
            cwd=ROOT, capture_output=True, check=False,
        )