The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes build logs and artifacts from Buildkite jobs. These external data sources can be influenced by anyone with commit access to a repository, potentially containing instructions designed to manipulate the agent's behavior. * Ingestion points: The scripts/buildkite script fetches logs via the bk job log command, and instructions in SKILL.md and references/build-debugging.md guide the agent to analyze this content. * Boundary markers: The skill does not implement or instruct the agent to use boundary markers to isolate log content from its own instructions. * Capability inventory: The skill enables triggering new builds, canceling active jobs, and accessing the full Buildkite REST API, providing a significant surface for exploitation if the agent is manipulated. * Sanitization: No sanitization is performed on the log content before it is presented to the agent.
[REMOTE_CODE_EXECUTION]: The setup script installs the official Buildkite CLI and the jq utility using the Homebrew package manager. * Evidence: scripts/setup contains commands to brew install jq and brew tap buildkite/buildkite && brew install buildkite/buildkite/bk. * Context: These are standard installations of official tools from a well-known service provider.
[COMMAND_EXECUTION]: The skill provides a wrapper script that passes user-provided arguments directly to the Buildkite CLI. * Evidence: The scripts/buildkite script concludes with exec bk "$@", allowing the agent to execute any supported Buildkite CLI command.

buildkite