The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Downloads and configures MCP servers from Anthropic's official registry, which are trusted sources.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it is designed to ingest and act upon untrusted data from external environments.
Ingestion points: Processes data from GitHub notifications, PR bodies, calendar events, and external search results as defined in the 'Example Workflows' section of SKILL.md.
Boundary markers: Absent; the prompts used in scheduled tasks do not include delimiters or instructions to ignore embedded commands in the fetched data.
Capability inventory: The agent possesses significant capabilities including file system writes (memory), GitHub API interactions (posting comments), and browser/desktop control (computer-use MCP).
Sanitization: No evidence of sanitization or validation of external content before it is processed by the agent.
[COMMAND_EXECUTION]: Uses programmatic mode (claude -p) and npx to initialize tasks and load MCP servers. While these are powerful capabilities, they are used here according to standard platform practices.

autonomous-agent-harness