SSkillEngineering
Prompt Injection Review (Claude Code)
claude codeAIsecurityagentsfree
In short
Map what untrusted text reaches your model, and what it can then do.
⛨Free, and yours to run in your own tools. No account needed.
TypeSkillCategoryEngineeringRuns onClaude CodeLicenseStandard license
⛨Human-reviewedEvery listing checked before it goes live
⛨Source-stampedCarries a visible line saying where it came from
⛨Real reviewsOnly verified buyers can rate it
A Claude Code skill for LLM features that read anything a stranger wrote. An LLM cannot reliably separate instructions from data, so injection is an architecture problem, not a prompt problem: telling a model to ignore instructions is a request, not a boundary. Crosses every untrusted input path against every capability the model holds, because untrusted text plus capability IS the vulnerability and neither alone is. Covers the most-missed source (tool results), the exfiltration path through rendered markdown that needs no write tool at all, and which defences are real versus decorative.
- ✓1 .md (SKILL.md) file (7 KB), instant download
- ✓Runs on Claude Code
- ✓Version 1.0.0, all v1.x updates included free
