The Agent Skills Directory

Dynamic Command Execution: The sync-mesh.sh script utilizes the eval shell command and unsanitized subshells to perform network requests. The parameters for these operations are sourced from the mesh.json configuration file and environment variables. This pattern presents a potential risk where malicious values in the configuration file (such as shell metacharacters in squad names or URLs) could result in arbitrary command execution on the host system.
External Resource Access: By design, the skill fetches information from remote Git repositories and HTTPS endpoints. This behavior is necessary for its core function of synchronizing state across machines, but it requires users to ensure that the sources configured in mesh.json are trusted and secure.
Indirect Prompt Injection Surface: The skill processes external data which could influence agent behavior.
Ingestion points: The skill reads mesh.json and remote SUMMARY.md files (fetched via sync-mesh.sh or sync-mesh.ps1).
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" markers when reading or processing these files.
Capability inventory: The skill possesses the ability to execute shell commands via git, curl, and eval, and can perform filesystem writes.
Sanitization: The sync-mesh.sh script does not implement sanitization for values retrieved from mesh.json before they are passed to shell execution contexts.
Secure Credential Handling: The synchronization scripts leverage environment variables for authentication tokens. While this avoids hardcoding sensitive credentials, users should be mindful of how these environment variables are populated and protected within their development environment.

distributed-mesh