The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several CLI tools including curl for status notifications to a local service (localhost:31337), bun for running TypeScript-based logic, and gh for GitHub API interactions. It also uses yt-dlp to fetch metadata from video sources.
[EXTERNAL_DOWNLOADS]: The skill periodically fetches information from external sources such as Anthropic's engineering blog, changelogs, and documentation. It also monitors RSS feeds from well-known AI practitioners (e.g., Simon Willison, Chip Huyen) and official GitHub repositories for the Model Context Protocol and Claude Code. These downloads are used to identify new architectural patterns and feature updates.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests and processes untrusted data from external websites and transcripts to generate recommendations for the system's core configuration.
Ingestion points: Content is retrieved via Tools/Anthropic.ts (web fetching) and Workflows/Upgrade.md (YouTube transcripts via GetTranscript.ts).
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the external content during the extraction phase.
Capability inventory: The skill proposes modifications to the agent's internal "Algorithm" specification, security hooks, and skill definitions based on the ingested data.
Sanitization: The workflow mitigates risk by producing a "LifeOS Upgrade Report" that requires user review and manual implementation of recommendations, rather than auto-applying changes.

Upgrade