The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill configures a tool that ingests data from untrusted external sources (e.g., RSS feeds, Reddit, Mastodon, arXiv) and applies LLM-powered transforms like summarization, filtering, and ranking. Malicious instructions embedded in these external content streams could attempt to influence the LLM's output.
Ingestion points: External content adapters defined in config.yaml (RSS, Reddit, GitHub, Mastodon, etc.).
Boundary markers: No specific delimiters or "ignore instructions" warnings are documented for the LLM transformation pipelines.
Capability inventory: The skill primarily performs file writes for configuration and executes local dashboard management commands.
Sanitization: No explicit sanitization or filtering of external content before LLM processing is mentioned.
[EXTERNAL_DOWNLOADS]: Fetches the official pace Docker image from the author's GitHub Container Registry (ghcr.io/av/pace:latest). It also references well-known technology services for data ingestion (e.g., GitHub API, npm registry, arXiv, Wikipedia).
[COMMAND_EXECUTION]: Executes local commands for the pace CLI tool, docker, docker-compose, and bun for dashboard management and validation. It also includes instructions for interacting with the local dashboard server via curl.

pace-dashboard-configure