Assumption: only the stated behavior is known; prompt handling, logging, plugins, storage, and execution limits are unverified. - **LLM01 — Prompt Injection: ⚠️ needs review.** Ensure user/file content is passed as a separate `user` turn, never concatenated into a system prompt. Treat file contents as untrusted. - **LLM02 — Sensitive Information Disclosure: ⚠️ needs review.** Prevent credentials or PII from entering prompts; redact agent responses and logs. - **LLM03 — Supply Chain: ⚠️ needs ...
Installs into .claude/skills of the current project.
Are you the author of Common Llm Security?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/hoangnguyen0403-common-llm-security-a9e81dac)
Assumption: only the stated behavior is known; prompt handling, logging, plugins, storage, and execution limits are unverified.
- **LLM01 — Prompt Injection: ⚠️ needs review.** Ensure user/file content is passed as a separate `user` turn, never concatenated into a system prompt. Treat file contents as untrusted.
- **LLM02 — Sensitive Information Disclosure: ⚠️ needs review.** Prevent credentials or PII from entering prompts; redact agent responses and logs.
- **LLM03 — Supply Chain: ⚠️ needs review.** Review all models, plugins, and package resources; pin source revisions and hashes.
- **LLM04 — Data & Model Poisoning: ⚠️ needs review.** Validate and sanitize file-derived data before persistence, training, or embedding.
- **LLM05 — Improper Output Handling: ⚠️ needs review.** Do not use model output directly as a filesystem path or command. Validate against an explicit path policy and sanitize outputs.
- **LLM06 — Excessive Agency: 🔴 confirmed finding (P0).** Autonomous deletion without human-in-the-loop confirmation is unsafe. Require confirmation immediately before deletion, enforce least-privilege read/delete permissions, restrict operations to an allowlisted workspace, support dry-run and audit logging, and block recursive or broad paths.
- **LLM07 — System Prompt Leakage: ⚠️ needs review.** Keep system instructions out of tool errors, file responses, and API output.
- **LLM08 — Vector & Embedding Weaknesses: ⚠️ needs review.** If files are indexed, validate content and enforce tenant/workspace namespace isolation.
- **LLM09 — Misinformation: ✅ not affected based on the stated scope.** No critical medical, financial, or legal decision is described.
- **LLM10 — Unbounded Consumption: ⚠️ needs review.** Set `max_tokens`, invocation rate limits, and an agent loop depth/iteration cap.
The confirmed **LLM06 P0** issue caps the security score at **40/100** until deletion requires explicit confirmation and host-enforced filesystem boundaries.
