The Agent Skills Directory

[SAFE]: The skill is designed as a reference and scaffolding resource. It includes explicit guardrails that prioritize read-only actions and require user consent for any proposed modifications or command executions.
[INDIRECT_PROMPT_INJECTION]: The skill reads local repository files (e.g., plugin.json, SKILL.md) to provide tailored guidance, which creates a surface for indirect prompt injection if those files contain malicious instructions. The risk is mitigated by the skill's limited toolset and its role in proposing rather than auto-executing actions.
Ingestion points: Plugin configuration files, manifests, and component instructions within the user's workspace.
Boundary markers: None identified in the provided files.
Capability inventory: The skill uses the Read, Grep, and Glob tools to inspect files and suggests shell commands or code diffs for user approval.
Sanitization: The skill does not explicitly sanitize or filter content retrieved from the analyzed files.

plugin-authoring