The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes existing README.md files which are untrusted external inputs. It lacks explicit boundary markers or instructions to ignore embedded commands within the content being compacted. * Ingestion points: The skill reads the entire README.md file in the diagnostic step (SKILL.md). * Boundary markers: There are no instructions provided to the agent to distinguish between the content of the README and the skill's own instructions. * Capability inventory: The skill utilizes Bash, Write, and Edit tools (SKILL.md). * Sanitization: No sanitization or filtering is performed on the ingested text before processing or potentially using it in shell-based formatting commands.
[COMMAND_EXECUTION]: The skill instructs the agent to execute code formatters (e.g., oxfmt, Prettier) via the Bash tool if a configuration is detected in the repository (SKILL.md). While intended for standard development workflows, this execution path could be exploited if a repository's configuration files are maliciously crafted.

compact-readme