The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation describes patterns for ingesting untrusted external data, such as webhook payloads (via the buildkite:webhook metadata key) or user-supplied fields from block steps, and interpolating them into build annotations or dynamically generated pipelines.
Ingestion points: SKILL.md (Meta-data section) and references/patterns-and-recipes.md (Block step and webhook examples) describe reading external data.
Boundary markers: Examples typically show direct interpolation of variables into shell scripts or YAML templates without explicit boundary markers or instructions to ignore embedded commands.
Capability inventory: The skill enables pipeline upload (dynamic command execution), meta-data set (state sharing), and annotate (displaying content).
Sanitization: While jq is used in examples to parse JSON payloads, there is no explicit instruction provided for sanitizing these values before they are interpolated into executable pipeline YAML or Markdown annotations.
[DYNAMIC_EXECUTION]: The skill documents the pipeline upload subcommand, which allows for the runtime generation and execution of pipeline steps. This is a standard and intended feature of the Buildkite platform for creating dynamic workflows based on build state.

buildkite-agent-runtime