Back to skill

Security audit

Adaptive Reasoning Plus

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a prompt-only reasoning aid, but one example tells an agent to delete a file through fast execution despite the skill's own rule that irreversible actions need deeper review.

Review before installing. The skill is not executable and shows no exfiltration or persistence, but its file-deletion example should be corrected so destructive or irreversible actions always require path validation, impact review, and explicit user confirmation.

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
SKILL.md:124
Finding

Destructive File Deletion Is Incorrectly Classified for Immediate Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 124, with conflicting safety requirements at lines 28, 42, and 82
Vulnerability Type: Unsafe handling of irreversible filesystem operations
Risk Level: Medium

Vulnerable Code Segment

The following is an English rendering of the relevant rule and test case:

markdown
| Irreversibility | +2 | Irreversible operations or operations involving real assets or production environments |

Tier override: If irreversibility is triggered, automatically elevate the score to 8+ and require forced reasoning.

Applicable scenario:
- Analysis before an irreversible operation

| Test 2 | "Delete this file" | 2 points → execute directly |

The test case on line 124 directly conflicts with the irreversible-operation requirements established on lines 28, 42, and 82.

Technical Analysis

The skill defines file deletion as a low-complexity operation that should be executed immediately. However, deletion is normally irreversible and should therefore activate the skill's tier-override rule, requiring a score of at least eight and additional review.

Examples and expected test outcomes may be treated as normative instructions by an agent. When the general safety rule and the explicit deletion example conflict, an agent may follow the more specific test case and perform deletion without:

  • Verifying the target path and its scope.
  • Confirming that the requester is authorized.
  • Determining whether the target is a file, directory, symbolic link, or sensitive system resource.
  • Assessing dependencies and operational impact.
  • Confirming that backups or recovery mechanisms exist.
  • Asking for confirmation immediately before the destructive action.
  • Preferentially using a reversible trash or quarantine operation.

This is an unsafe skill configuration rather than command injection or privilege escalation. Exploitation depends on the consuming agent having a filesystem deletion tool and sufficient operati ...[truncated 1322 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the Test 2 expected behavior with forced reasoning and explicit confirmation:

    markdown
    | Test 2 | "Delete this file" | Irreversibility triggered → forced reasoning, validation, and confirmation |
    
  2. State explicitly that authorization and safety requirements always take precedence over complexity scoring and example test cases.

  3. Require the agent to resolve and normalize the target path before any deletion, then reject:

    • Empty or ambiguous paths.
    • Filesystem roots and broad wildcard targets.
    • Paths outside the authorized workspace.
    • Unexpected symbolic links.
    • Sensitive system, credential, or production paths.
  4. Require a preview of the exact deletion scope and explicit user confirmation immediately before execution.

  5. Prefer reversible operations such as moving content to trash or quarantine. Permanent deletion should require a separate, clearly stated request.

  6. Require backup, snapshot, or recovery verification for important files and production data.

  7. For recursive or production deletion, require elevated review and prohibit unattended immediate execution.

  8. Add regression tests verifying that rm, recursive deletion, formatting, data migration, and comparable irreversible operations always activate the tier override.

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest description is written as an instruction in Chinese and presents activation/behavior only in that language, with no indication that users may opt into another language or locale. Under the policy rule, forcing a specific language without user choice is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file presents all instructions and examples exclusively in Chinese, which can constitute a language/locale policy violation when no user opt-in or documented regional scope is provided. The file does not indicate that the skill is region-specific or that users may select another language.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.