Back to skill

Security audit

ai-data-delivery

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill benign with medium confidence: the completed AIG scan found no issues and most reviewed behavior aligns with its purpose, but Skillspector coverage was partial and it reported several capability and code-pattern concerns.

The verdict is benign, not risk-free. Before installing, consider that Skillspector inspected 83.9% of components, with some files only partially inspected and some references unresolved; AIG completed with no findings. The reported permission capabilities and YAML-loading pattern remain points to review, and the scan does not establish the host runtime's actual permissions.

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
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (14)

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 43)May include surrounding context.

md
| Connect catalogs or track schema changes | [Read-only metadata](references/readonly-metadata.md); explicit connection and selected tables |

Code capability 'env' detected in scripts/_contract.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Code capability 'file_write' detected in scripts/_contract.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'file_write' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Code capability 'network' detected in scripts/knowledge.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'network' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
85% confidence
Finding

Code accesses a cloud instance metadata endpoint (e.g. 169.254.169.254). A single request can return temporary IAM credentials, making this a high-value SSRF target for credential theft.

Content

Scanner excerpt · workbench/db_metadata.py (reported line 91)May include surrounding context.

python
addresses = socket.getaddrinfo(p["host"], p["port"], type=socket.SOCK_STREAM)
    ips = [ipaddress.ip_address(a[4][0].split('%')[0]) for a in addresses]
    ips = [getattr(a, 'ipv4_mapped', None) or a for a in ips]
    if not ips or any(a.is_link_local or a.is_multicast or a.is_unspecified or str(a) == 'fd00:ec2::254' for a in ips):
        raise MetadataError("Database endpoint is not allowed / 不允许连接该数据库地址")
    secret = os.environ.get(p["password_env"])
    if not secret:

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% 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/query-control.md (reported line 3)May include surrounding context.

md
# Query control · 问数控制单(0.3.4)

Use this after a question selects a metric. Business users describe the question and review meanings; the host Agent/FDE proposes bindings through the existing candidate/edit workflow. Do not ask users to write JSON or confirm invented defaults. This is a reviewed handoff contract, **not a SQL compiler, query gateway or authorization grant**.

## Read a delivered question

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% 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/readonly-metadata.md (reported line 39)May include surrounding context.

python -m pip install -r requirements-connectors.txt

text

Use an administrator-provisioned read-only account with visibility on the explicitly selected base tables. PostgreSQL needs database CONNECT, schema USAGE and table SELECT; MySQL needs table SELECT. The connector issues only fixed catalog queries inside read-only transactions, despite that account's ability to read table rows. It does not accept user SQL, DSNs, connection options or business-row queries. Views and virtual tables are rejected. PostgreSQL partitioned base tables are supported as declared catalog objects; partition data is not read. MySQL uses two matching catalog captures because its catalog is not a transactional DDL snapshot. An absent table can mean lost visibility; it is never automatically deleted.

A PostgreSQL profile (`connection.json`) can be:

Insecure deserialization: yaml.load() with an unsafe Loader

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

Untrusted data is passed to an insecure deserializer (pickle, marshal, dill, jsonpickle, joblib, yaml.load without a safe Loader, or torch.load without weights_only). These deserializers reconstruct arbitrary objects and invoke callables during loading, so deserializing attacker-controlled bytes is equivalent to arbitrary code execution.

Content

Scanner excerpt · workbench/knowledge_parsers.py (reported line 673)May include surrounding context.

python
return obj
        MaterialLoader.add_constructor(yaml.resolver.BaseResolver.DEFAULT_MAPPING_TAG, mapping)
        try:
            value = yaml.load(text, Loader=MaterialLoader)
        except yaml.YAMLError as exc:
            raise ValueError(f'YAML 格式无效:{exc}') from exc
        value = json.loads(json.dumps(value, ensure_ascii=False, default=str, allow_nan=False))

Scope Creep

Low
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · CHANGELOG.md (reported line 36)May include surrounding context.

md
ndustry trials exposed incomplete task dependencies, repetitive per-asset edits and avoidable reader retries.
- Authoring: atomic arrays on the existing `edit --file` path; explicit confirmation, evidence, revision conflicts, rollback and operation receipts remain guarded. `state --summary` discovers identities without full history.
- Handoff: read-only `inspect --id` highlights omitted mappings, relations and join keys, unresolved states and changed source evidence. Suggestions do not silently expand scope or confirm meaning.
- Reader: paged multi-asset `sources` with deduplicated evidence and explicit missing targets; paged, scope-filtered term discovery on no match; clear page identity coverage without claiming business completeness.
- Guidance: one short business-modeling path, source-supported aliases, explicit secondary-table dependencies and reuse of existing confirmation authority. CLI and optional editor retain one canonical graph.
- Scope: no embedded model, business-query ex

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/knowledge.py (reported line 164)May include surrounding context.

python
elif args.command == "connect":
            result = service.metadata_connect({**req, "connection": json.loads(Path(args.file).read_text(encoding="utf-8-sig"))})
        elif args.command in {"scan", "regress"}:
            result = getattr(service, "metadata_" + args.command)({**req, "connection_id": args.connection,
                **({"refresh_review": args.refresh_review, "seed_draft": args.seed_draft} if args.command == "scan" else {})})
        elif args.command == "graph-image":
            from workbench.graph_image import render_graph

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/knowledge.py (reported line 226)May include surrounding context.

python
result = service.import_material({**req, "filename": path.name, "content_base64": base64.b64encode(path.read_bytes()).decode(),
                "domain": args.domain, "scope": args.scope, "language": args.language, "source_id": args.source_id, "distinct_source": args.distinct_source,
                **({"options": {"encoding": args.encoding}} if args.encoding else {}),
                **{key: getattr(args, key) for key in ("effective_from", "effective_to") if getattr(args, key) is not None},
                **({"source_key": args.source_key} if args.source_key is not None else {}),
                "seed_draft": args.seed_draft})
        elif args.command == "task":

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/knowledge.py (reported line 226)May include surrounding context.

python
result = service.import_material({**req, "filename": path.name, "content_base64": base64.b64encode(path.read_bytes()).decode(),
                "domain": args.domain, "scope": args.scope, "language": args.language, "source_id": args.source_id, "distinct_source": args.distinct_source,
                **({"options": {"encoding": args.encoding}} if args.encoding else {}),
                **{key: getattr(args, key) for key in ("effective_from", "effective_to") if getattr(args, key) is not None},
                **({"source_key": args.source_key} if args.source_key is not None else {}),
                "seed_draft": args.seed_draft})
        elif args.command == "task":

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · workbench/static/vendor/elk-LICENSE.md (reported line 54)May include surrounding context.

md
any manner that enables the transfer of a copy.

“Source Code” means the form of a Program preferred for making
modifications, including but not limited to software source code,
documentation source, and configuration files.

“Secondary License” means either the GNU General Public License, Version

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · workbench/static/vendor/elk-LICENSE.md (reported line 186)May include surrounding context.

md
PURPOSE. Each Recipient is solely responsible for determining the
appropriateness of using and distributing the Program and assumes all
risks associated with its exercise of rights under this Agreement,
including but not limited to the risks and costs of program errors,
compliance with applicable laws, damage to or loss of data, programs or
equipment, and unavailability or interruption of operations.