claudeception
Fail
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The activation hook script (
scripts/claudeception-activator.sh) utilizes imperative and authoritative keywords such as "CRITICAL", "MUST", "NON-NEGOTIABLE", and "NOT optional". This language is designed to override the agent's default operational constraints and force a specific behavior (skill evaluation) after every interaction, bypassing normal instruction prioritization. - [COMMAND_EXECUTION]: The installation process establishes global persistence by adding a shell script hook to the agent's environment (
~/.claude/hooks/claudeception-activator.sh) and modifying the configuration file (~/.claude/settings.json). This ensures the skill's logic and injected prompts execute automatically during every user session. - [PROMPT_INJECTION]: The skill's core function—generating and saving new instruction files (skills) based on session activity—creates a surface for indirect prompt injection where malicious content from external sources can become permanently codified in the agent's behavior.
- Ingestion points: Untrusted data enters the agent context via the
WebFetch,WebSearch,Read, andGreptools during the
Recommendations
- AI detected serious security threats
Audit Metadata