Back to skill

Security audit

deep-research

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners report direct network and local-file access beyond its stated host-tool boundary, along with other security weaknesses; the review is incomplete because one referenced path was missing.

Consider the suspicious verdict and the reported network, local-file, and report-rendering weaknesses before installing. The scan covered all listed components but was marked partial because a local path-like reference from SKILL.md did not match a bundled artifact, so evidence is incomplete.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/verify_evidence.py
Finding

Automatic redirects bypass public-address validation

Content
View full analysis

Impact

An attacker can cause GET requests to internal HTTP services reachable by the host, including unauthenticated administrative or metadata endpoints. Responses are processed locally for quote verification; the code does not demonstrate full response exfiltration to the attacker.

Vulnerability Patterns
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The advertised host-mediated access boundary is contradicted by the supplied verifier's direct HTTP requests and unrestricted file:// reads. Attacker-influenced source records can therefore trigger network access outside host-tool controls or read files outside the ordinary local-path boundary, subject to the process's actual permissions.

Content

No source excerpt is available for this finding.

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

The renderer contradicts its promise that all ledger content is HTML-escaped by inserting unknown verification statuses directly into HTML.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Unknown verification statuses become badge labels and are interpolated into HTML without escaping. A tampered verification.jsonl record referenced by a claim can inject active HTML into the generated report, potentially executing script when a reviewer opens it.

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/run_gates.py (reported line 37)May include surrounding context.

python
failed = 0
    for command in commands:
        print(f"\n$ {' '.join(command)}")
        completed = subprocess.run(command, cwd=run_dir.parent)
        failed += completed.returncode != 0
    print(f"\nGATES: {'PASS' if not failed else 'FAIL'} ({failed} failed)")
    return 0 if not failed else 1

The skill performs direct network requests instead of exclusively reusing authorized host tools.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The verifier fetches ledger-supplied URLs directly through urllib rather than an authorized host retrieval adapter, bypassing any controls implemented only in that adapter. Its initial DNS check also does not validate redirect destinations or pin the validated address for the connection, allowing redirects or DNS rebinding to reach otherwise blocked internal services when process networking permits.

Content

No source excerpt is available for this finding.

--require-verified does not enforce the documented requirement for verified evidence on high-impact claims.

Medium
Category
Excessive Agency
Confidence
100% confidence
Finding

check_claims constructs errors for high-impact claims lacking verified evidence but never returns those errors, so --require-verified cannot enforce that requirement. Additionally, main appends mismatch errors without updating result['ok'], allowing a successful exit and misleading gate approval despite verification failures.

Content

No source excerpt is available for this finding.