The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting and processing untrusted data from external sources.
Ingestion points: During Phase 1 (Investigation), the skill instructs the agent to retrieve external data using gh issue view <number> and to search for error messages using the Grep tool. This content is directly used to define the problem location and root cause.
Boundary markers: The instructions do not specify any delimiters or warnings to ignore embedded instructions within the GitHub issue content or error logs, which could allow malicious text to influence the agent's behavior.
Capability inventory: The skill has access to powerful tools such as Bash, Edit, Write, and Skill, which could be abused if the agent inadvertently follows instructions found in untrusted external data.
Sanitization: There are no defined steps for sanitizing, escaping, or validating the input from external tools before it is interpolated into the agent's reasoning process.

bug-fix