The Agent Skills Directory

[DATA_EXPOSURE]: The skill explicitly forbids the agent from running commands that expose environment variables (e.g., env, printenv, echo $CIRCLE_TOKEN), demonstrating a proactive security posture regarding credential leakage.
[COMMAND_EXECUTION]: The skill manages remote execution through the chunk CLI, routing commands to isolated CircleCI sidecar environments as intended by the tool's design. This includes syncing local files and running validation tests.
[CREDENTIALS_UNSAFE]: Authentication is handled via secure methods such as OAuth with PKCE or system keychain storage through the vendor's official CLI. The skill avoids hardcoding tokens and provides instructions for users to perform interactive logins when necessary.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided inputs such as snapshot IDs and command strings. While these are executed on the sidecar, the risk is localized to the remote environment, and the skill includes state probes (chunk --version, chunk auth status) to verify the environment before performing operations.

chunk-sidecar