The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection as it is designed to process external data sources.
Ingestion points: Reads user context, project files, and external evidence (scraped text, reviews, comments) as defined in SKILL.md and references/skill-contract.md.
Boundary markers: Explicitly instructs the agent to treat external inputs as untrusted and to ignore embedded instructions.
Capability inventory: No scripts or command execution capabilities are included in the skill.
Sanitization: Relies on instructional constraints to prevent the agent from following instructions found in data.
[SAFE]: The skill documentation mandates best practices for security and privacy, such as keeping API keys out of source control and requiring user authorization for any side effects.
[NO_CODE]: The skill is composed of configuration and instructional markdown files with no executable scripts or system-level commands.

channel-registry