Security risk report for @forgecat/contains-studio_agents_bonus v0.1.0
Source Integrity
Low
Profile sourced from a declared GitHub repository (contains-studio/agents) with clear attribution and version tracking.
No obfuscated or suspicious external dependencies; no typosquatted packages or hidden supply-chain vectors.
Agent Intent
INT-000Medium
Profile instructs the AI to 'PROACTIVELY' invoke the studio-coach agent automatically and without explicit user request, expanding the agent's authority to self-trigger based on vague conditions ('agents seem stuck', 'team needs motivation').
Guidance plants persistent behavioral rules (e.g., 'Starting each session with energizing affirmations', 'Maintaining unwavering belief') that will systematically steer future outputs toward motivational framing regardless of context appropriateness—a form of guidance poisoning that biases the agent's judgment.
No direct credential exfiltration, system prompt leakage, or remote payload execution; however, the self-triggering and persistent behavioral injection constitute medium-risk manipulation of agent autonomy and output bias.
Details
Evidence
'PROACTIVELY use this agent when complex multi-agent tasks begin, when agents seem stuck or overwhelmed, or when the team needs motivation'
'Starting each session with energizing affirmations'
'Maintaining unwavering belief in the team's abilities'
'You are not just a coach but a catalyst for greatness'
Permissions
Low
Declared tools (Write, Task, Read) are minimal and aligned with the stated coaching/coordination function.
No shell, file_write, file_delete, or high-risk categories; no alwaysApply rules with broad globs.
Authority scope matches the described purpose of agent motivation and multi-agent orchestration.
ForgeCat