Back to skill

Security audit

defect-detective

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because it may send diagnostic-derived search terms to an issue tracker without requiring sensitive-data redaction or warning about disclosure; its review is incomplete.

Before installing, consider whether diagnostic log details could be sensitive and whether the configured tracker is approved to receive them. The scan covered 6 of 7 components fully; one file was only partially inspected because obfuscated instruction text could not be fully evaluated, so the evidence is 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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · references/triage-examples.en.md (reported line 95)May include surrounding context.

text
Environment: macOS + zsh, AI agent running in a sandboxed execution environment
Symptom: agent fails to write skill files:
  [Errno 13] Permission denied: '~/.trae-cn/skills/fast-commit/SKILL.md'
  os.makedirs('~/.trae-cn/skills/fast-commit') also Permission denied
The same process writes to the project directory without issue; only ~/.trae-cn/skills/ is denied. Reproduces every time.

Tracker searches transmit log-derived fingerprint data without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow sends log-derived message keywords and endpoint details to the configured tracker without requiring sensitive-data redaction or checking whether that destination is approved for the data. Logs can contain credentials, personal information, or confidential URL parameters, which could consequently enter tracker query logs even though the operation is read-only. Using an engineer-configured tracker reduces the concern compared with an arbitrary destination, but does not eliminate this disclosure path.

Content

No source excerpt is available for this finding.

The workflow sends diagnostic-derived search terms to the configured issue tracker without warning about potential disclosure of sensitive log content.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The normalization step removes volatile fields but does not require removing secrets or personal information before diagnostic-derived search terms are transmitted to the tracker. Sensitive message fragments or endpoint values could therefore be disclosed to the tracker and retained in its query logs. Read-only credentials prevent ticket mutations, not outbound disclosure; an internal, approved tracker would reduce the practical risk.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/triage-examples.en.md (reported line 96)May include surrounding context.

text
Environment: macOS + zsh, AI agent running in a sandboxed execution environment
Symptom: agent fails to write skill files:
  [Errno 13] Permission denied: '~/.trae-cn/skills/fast-commit/SKILL.md'
  os.makedirs('~/.trae-cn/skills/fast-commit') also Permission denied
The same process writes to the project directory without issue; only ~/.trae-cn/skills/ is denied. Reproduces every time.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/triage-examples.md (reported line 96)May include surrounding context.

text
环境:macOS + zsh,AI agent 运行在沙盒执行环境
现象:agent 写入 skill 文件失败:
  [Errno 13] Permission denied: '~/.trae-cn/skills/fast-commit/SKILL.md'
  os.makedirs('~/.trae-cn/skills/fast-commit') 同样 Permission denied
同一进程对项目目录的写入完全正常,只有写 ~/.trae-cn/skills/ 被拒。每次必现。