bermuda-install
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to run
npx skills add bon5co/bermuda. This involves downloading and executing content from an external registry. The resource matches the vendor naming patterns for the skill author. - [COMMAND_EXECUTION]: The skill performs file system operations to modify persistent agent configuration files, specifically
~/.codex/AGENTS.mdand~/.claude/CLAUDE.md. It also directs the agent to execute shell commands to verify installations and manage links. - [PERSISTENCE]: By modifying the standing instruction files (
AGENTS.mdorCLAUDE.md), the skill ensures that the Bermuda harness instructions are loaded into the agent's context for every future session, establishing a persistent behavioral change. - [INDIRECT_PROMPT_INJECTION]: The skill configures a target agent to ingest data from external logs, lessons, and memory files (e.g.,
bermuda alog list,bermuda improve write,bermuda memory path). - Ingestion points: The agent is instructed to read from log files, shared records, and lessons at the start of tasks.
- Boundary markers: The skill uses specific comment markers (
<!-- bermuda-skill:begin -->) to delimit the managed instruction block. - Capability inventory: The instructions provide the agent with capabilities to execute a variety of
bermudaCLI commands, including writing to logs, posting to forums, and running flows. - Sanitization: No explicit sanitization of the ingested data is described in the provided setup instructions.
Audit Metadata