The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow that instructs the agent to ingest and process data from external, third-party repositories to inform development decisions, creating a surface for indirect prompt injection.
Ingestion points: The agent is directed to read SKILL.md and reference files from https://github.com/android/skills, https://github.com/skydoves/compose-performance-skills, and https://github.com/chrisbanes/skills as defined in references/current-guidance.md.
Boundary markers: The skill uses instructional boundaries, such as requiring the agent to "Distill only reusable decision rules" and "Report the source surface and residual source docs not loaded," which provides a manual filtering step but lacks automated isolation.
Capability inventory: The skill is scoped to handle high-impact Android tasks, including security configurations, credential management, and architecture decisions.
Sanitization: There is no evidence of automated sanitization, schema validation, or escaping of the content fetched from the external markdown files before the agent incorporates it into its reasoning.

android-external-skill-source-coverage