The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled local ESM script (scripts/providers.mjs) to retrieve provider configurations and format requirements.
[EXTERNAL_DOWNLOADS]: The skill fetches authoritative documentation from well-known services and trusted organizations, including Anthropic, Cursor, OpenAI, and Vercel Labs, to perform compatibility checks.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests untrusted skill content and interpolates it into subagent prompts for analysis.
Ingestion points: Skill content is read from user-provided paths or direct input via the $ARGUMENTS variable in SKILL.md.
Boundary markers: The subagent prompt template in references/audit-prompt.md utilizes horizontal rules (---) to delimit the interpolated YAML frontmatter and markdown body.
Capability inventory: The parent agent can read/write files and execute local scripts; subagents are granted WebFetch capabilities to access provider documentation.
Sanitization: There is no explicit sanitization or escaping of the interpolated skill content within the subagent prompt template.

skill-portability