loop-briefing
Warn
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill features "Self-Evolving Skill" instructions that direct the agent to modify the SKILL.md file itself based on runtime results. This self-modification mechanism creates a risk of persistent instruction corruption or behavior alteration if the agent is influenced by malicious data.
- [PROMPT_INJECTION]: The skill incorporates untrusted data from the local environment—including file contents from /tmp/decision_options.txt, git commit messages, and repository status—into the briefing context without explicit boundary markers or sanitization, exposing the agent to indirect prompt injection.
- [DATA_EXFILTRATION]: The skill's execution commands utilize
env -u HTTPS_PROXY -u HTTP_PROXYto intentionally unset network proxy settings, bypassing potential security controls. It is designed to transmit project-sensitive data (branch names, git status, and commit history) to the third-party Pushover API. - [COMMAND_EXECUTION]: The skill uses the
bunruntime to execute a local script utility (pushover_core.ts) located in the plugin's library directory.
Audit Metadata