The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx @agent-native/core@latest to download and execute utility packages from the npm registry. These tools are used for generating local recaps, fetching block schemas, and re-authenticating clients.
[COMMAND_EXECUTION]: It instructs the agent to execute various shell commands for managing visual plans, such as plan local check, plan local serve, and plan local verify. These are intended for local development and previewing recap artifacts.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of git diffs, pull request descriptions, and commit history, which represents a surface for indirect prompt injection.
Ingestion points: Data is ingested from PRs, branches, commits, and git diffs as described in the SKILL.md file.
Boundary markers: The skill relies on structured MDX blocks for output and includes explicit documentation to prevent the transcription of secrets.
Capability inventory: The skill can execute CLI commands via npx and interact with the Plan MCP connector for network and database operations.
Sanitization: The skill contains mandatory instructions to redact API keys, tokens, and other sensitive literals found in the source diffs (e.g., using sk-••• or <redacted>).

visual-recap