The Agent Skills Directory

System Configuration Persistence: The github-multi-account skill directs the agent to modify the user's PowerShell profile ($PROFILE) and create command wrappers in a new directory added to the system PATH. These actions establish persistence for custom aliases and functions across all shell sessions, which is a significant modification to the host environment's configuration.
Dynamic Execution in Scripts: The sync-mesh.sh script, part of the distributed-mesh functionality, uses the eval command to process network requests based on variables retrieved from a configuration file (mesh.json). Constructing commands at runtime in this manner is a security consideration as it can lead to unintended command execution if the configuration data is not strictly validated.
Instructional Concealment: Governance files such as squad.agent.md and scribe-charter.md define the Scribe agent as "silent" and "invisible," instructing it to "never speak to the user" and work entirely in the background. While this is a design choice for a logging component, the practice of suppressing agent output and concealing its presence is a notable behavioral pattern regarding transparency.
Remote Synchronization: The distributed-mesh skill implements synchronization logic that fetches data from remote repositories and URLs using git and curl. This allows the team context to be updated from external sources, introducing a dependency on the integrity of those remote locations and the data they provide.
Credential Protection Measures: The secret-handling skill establishes protocols for preventing the accidental commit of secrets, including regex-based scanning and pre-commit validation. These measures are designed to protect against data exposure, though they also highlight that agents have read access to sensitive files like .env within the repository by default.

{skill-name}