The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the npx @agent-native/core@latest utility for tasks such as scanning diffs (recap collect-diff), checking plan validity (plan local check), and serving local previews (plan local serve).
[EXTERNAL_DOWNLOADS]: The skill downloads the @agent-native/core package from the NPM registry and communicates with a hosted service at https://plan.agent-native.com. These are legitimate resources associated with the skill's infrastructure and vendor.
[PROMPT_INJECTION]: The skill processes untrusted external data (git diffs and PR content) which could potentially contain malicious instructions intended to manipulate the agent's output.
Ingestion points: Local files and git diffs accessed via the recap collect-diff and scan helpers.
Boundary markers: The instructions include a 'Grounding Rule' requiring that all structured blocks be derived mechanically from the actual diff, and explicitly forbids 'improvised' or 'invented' content.
Capability inventory: The agent has the capability to execute shell commands, write local MDX files, and call specific MCP tools.
Sanitization: The skill contains a dedicated security section mandating that the agent must never transcribe secrets (API keys, tokens, etc.) from diffs and must redact them instead.

visual-recap