Back to skill

Security audit

Openclaw Agent Skill Evolution

Security checks for vulnerabilities and agentic risk

Overview

This is a transparent guidance-only skill for improving OpenClaw skills, with disclosed safeguards and no hidden code, installer, or credential collection.

Install this only if you want an agent to help evaluate and improve OpenClaw skills. Keep production skill changes under review, especially anything involving shared tools, deletion, credentials, money, security boundaries, or system-wide behavior; require sandbox candidates, diffs, tests, and rollback before accepting deployment.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The usage guidance is effectively self-referential ('use when the conditions in the description are met') and does not provide concrete activation criteria. In an agent skill system, ambiguous triggers can cause the skill to activate too broadly or unpredictably, increasing the chance that high-impact behaviors like benchmarking, red-teaming, and iterative modification are invoked in the wrong context.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

Dokumen ini secara eksplisit mengizinkan AUTO-DEPLOY untuk perubahan yang dinilai low risk, reversible, verifiable, dan lolos test. Meski dibatasi, ini tetap memberi kewenangan otonom pada agen untuk memodifikasi/deploy skill tanpa human approval; jika klasifikasi risiko, verifikasi, atau test set salah/di-bypass, perubahan berbahaya atau regresi dapat masuk ke production.

Content

Scanner excerpt · SKILL.md (reported line 65)May include surrounding context.

md
| Candidate menyebabkan CRITICAL REGRESSION | REJECT | Capability gain ≠ izin merusak fitur lama |
| Motif = HYPE/TRENDING/VIRAL tanpa evidence | BLOCK | No-Hype Rule |
| "Lebih panjang/complex" diklaim = "lebih pintar" | TOLAK klaim | No-Fake-Intelligence Rule |
| Low risk + reversible + verifiable + test passed | Boleh AUTO-DEPLOY | Otomatis-terbatas boundary aman |
| Perubahan kritis | PROPOSE → human approval → deploy | Agent tak ubah boundary sendiri |

## 7. REASONING POLICY

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The example input explicitly constrains output language with "bahasa tetap Indonesia," and the skill file is otherwise written as a fixed Indonesian-language instruction set without offering a language choice. This can violate language/locale policy because it normalizes a mandatory locale rather than documenting user-selectable behavior.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · SKILL.md (reported line 293)May include surrounding context.

md
- **Environment berbeda (Termux/Android)**: dependency desktop gagal → validasi ARM64/permission sebelum adopsi.
- **Skill hanya prompt panjang**: LOW MATURITY → jangan tambal, evaluasi REWRITE dengan decision logic.
- **Tidak ada baseline**: tolak benchmark-banding; buat baseline dulu atau tandai UNKNOWN.
- **Human tidak merespons**: jangan auto-deploy perubahan kritis; status = NEED USER (blocked).
- **Verifikasi tak mungkin**: status = NEED VERIFICATION; jangan klaim sukses.

## Common Mistakes / Anti-Patterns