The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for processing external, untrusted data (such as test output logs or JSON results) and using that data to generate build annotations or dynamic pipeline steps.
Ingestion points: Content is read from test-output.txt, results.json, and piped from standard input in multiple examples.
Boundary markers: Not explicitly present in the provided bash scripts; data is processed via standard CLI tools like grep and jq.
Capability inventory: The skill leverages pipeline upload to dynamically create new execution steps and annotate to display processed data.
Sanitization: Scripts use structure-aware tools (jq) to parse data, which is appropriate for the intended CI/CD orchestration use-case.
[COMMAND_EXECUTION]: The skill provides instructions for using the buildkite-agent CLI, which is the standard tool for executing operations within Buildkite CI/CD environments.
[EXTERNAL_DOWNLOADS]: Reference recipes include standard examples of fetching assets or configuration from external sources (e.g., releases.example.com and vault.internal) using curl and wget.
[DYNAMIC_EXECUTION]: The pipeline upload subcommand is documented, which is a core feature of Buildkite used to generate and execute pipeline definitions at runtime based on job logic.

buildkite-agent-runtime