forge-party
Fail
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
forge-memoryCLI tool and interpolates the user-provided{TOPIC}directly into shell commands. - Evidence:
forge-memory search "<topic>" --limit 3andforge-memory log "Party terminée : {TOPIC}, ..."in the Workflow section. - Risk: If the
<topic>argument contains shell metacharacters (e.g.,;,&,|,`), it could lead to arbitrary command execution on the host machine. - [PROMPT_INJECTION]: The skill accepts a raw
topicargument which is used both for agent instructions and as a parameter in shell commands, creating a surface for both direct and indirect prompt injection. - [DATA_EXPOSURE]: The skill reads and writes to local project files (
.forge/memory/MEMORY.md) and system paths (~/.claude/skills/forge/references/agents/orchestrator.md). While these are within the expected 'FORGE' ecosystem, the combination with command injection increases the risk of unauthorized data access.
Recommendations
- AI detected serious security threats
Audit Metadata