33god-projects
Warn
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPERSISTENCEREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
misehooks to automatically execute multiple scripts includinglink-agentfiles.sh,provision-packs.py,sync-skills.py, andsync.pywhenever a user enters a project directory. - [PERSISTENCE]: The skill configures
systemdservices and timers such ashermes-gatewayandhermes-heartbeatto maintain active agents across system restarts. - [PERSISTENCE]: The skill injects project-specific hooks into global user configuration files for multiple AI agent CLIs, specifically
~/.codex/hooks.jsonand~/.kimi-code/config.toml. - [REMOTE_CODE_EXECUTION]: The
provision-packs.pyscript dynamically fetches and materializes skill packs from remote Git repositories into the project's.agents/skills/directory. - [EXTERNAL_DOWNLOADS]: The skill uses
npx -y bmad-method@latest installto download and execute tooling from the npm registry. - [DATA_EXFILTRATION]: The Hindsight memory feature sends agent context and learnings to the external API
https://api.hs.delo.sh. - [DATA_EXFILTRATION]: The Bloodbank integration publishes events to a NATS event bus, which defaults to
127.0.0.1:4222. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Plane ticket boards and Hindsight memories, which could contain malicious instructions. Ingestion points: Hindsight recall results and Bloodbank event data. Boundary markers: Uses
<hindsight-memory>tags to delimit memory content. Capability inventory: Includes shell command execution viamise, file system writes for configurations, and network operations for API and NATS communication. Sanitization: Employs HMAC signature verification for incoming Plane webhooks.
Audit Metadata