The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to bypass human oversight before performing an external write operation.
Evidence: In SKILL.md, Step 7 (Create GitHub issue) states: "Do NOT ask the user to review before creating — just create it and share the URL."
[COMMAND_EXECUTION]: The skill uses the GitHub command-line interface to perform write operations on remote repositories.
Evidence: SKILL.md invokes gh issue create in Step 7.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by processing untrusted data (the codebase) and using it to influence external actions.
Ingestion points: Codebase exploration occurs in Step 1 of SKILL.md using the Explore tool.
Boundary markers: Absent; there are no instructions to delimit or ignore potentially malicious instructions embedded within the analyzed codebase.
Capability inventory: The skill can create GitHub issues (gh issue create) and spawn multiple sub-agents (Agent tool) to process data.
Sanitization: Absent; the skill does not define validation or escaping procedures for content extracted from the codebase before it is posted to GitHub.

improve-codebase-architecture