The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by directing the agent to automate the generation of documentation and changelogs using dynamic placeholders.
Ingestion points: Untrusted data enters the context through external metadata fields such as extension names, descriptions, and manual instructions for changelog entries.
Boundary markers: There are no delimiters or specific warnings to ignore instructions embedded within the data processed for CHANGELOG.md or README.md updates.
Capability inventory: The agent has access to file-write capabilities for project files and can trigger build scripts (bun run, npm run).
Sanitization: No validation or sanitization of external content is specified before it is interpolated into the extension's files.
[COMMAND_EXECUTION]: The instructions mandate the use of the Raycast toolchain (ray lint, ray build) and standard package managers (bun, npm). These are standard procedures for developing within this ecosystem. It also includes the use of npx @raycast/api@latest publish to interact with official services.

raycast