The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions designed to influence the agent.
Ingestion points: The skill ingests data from external CI providers, specifically reading logs using gh run view and downloading build artifacts with gh run download (SKILL.md).
Boundary markers: The instructions do not define boundary markers or delimiters to separate the retrieved log content from the agent's instructions, nor do they advise ignoring potential commands embedded in the logs.
Capability inventory: The skill allows for the execution of shell commands through the gh CLI and local reproduction of build commands, providing an execution path for instructions that might be extracted from failing logs.
Sanitization: There is no requirement or guidance for sanitizing, escaping, or validating the content of the CI logs or artifacts before the agent processes them or uses them to generate reproduction commands.

ci-debugging