The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves technical preferences from an external source (the author's GitHub repository), creating a surface for indirect prompt injection where instructions embedded in the guidance could attempt to influence the agent.
Ingestion points: Technical preference documents and README files located in the sjunepark/mytech repository.
Boundary markers: The instructions in SKILL.md explicitly state that local project requirements take precedence and that the skill should not expand its underlying task authority.
Capability inventory: The skill uses git and standard file system commands (cat, ls) to interact with the repository and local files, as evidenced by commands like git rev-parse HEAD and git status in the provided evaluation logs.
Sanitization: No explicit technical sanitization or filtering of the external content is described, though the skill is instructed to report coverage limits and missing sources.
[COMMAND_EXECUTION]: Evaluation records (e.g., evals/cli-b.json and evals/missing-b.json) show the agent performing shell commands including git rev-parse HEAD, git status --short, and ls -ld to manage and verify the state of the preference repository.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch data from a remote GitHub repository (https://github.com/sjunepark/mytech). This is documented as a vendor-owned resource belonging to the skill author (sjunepark).

consult-mytech