The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script (scripts/snapshot.sh) and a repository-specific binary (.opencode/bin/fluxctl) to manage file snapshots and session state. These operations are limited to the local file system and are standard for documentation management tasks.
[PROMPT_INJECTION]: The skill processes untrusted content by reading all markdown files in the project's documentation vault (.flux/brain/). While it uses file delimiters (=== path ===) to separate content, malicious instructions hidden within audited notes could potentially influence the subagents tasked with reviewing the documentation. This represents an indirect prompt injection surface common to tools that process large amounts of user-generated or codebase content.

flux-opencode-meditate