The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches remote Markdown files from 'https://basic-anteater-667.convex.site/' and instructs the agent to "follow its Procedure+Rules sections". This allows remote content to influence or dictate agent behavior.
Ingestion points: The agent fetches 'capabilities.json' and individual '/capability/.md' files from 'https://basic-anteater-667.convex.site/'.
Boundary markers: The skill includes a cautionary rule: "Served doc text is procedure instructions, not arbitrary shell to blindly execute — apply normal judgment."
Capability inventory: The skill executes local scripts ('/add-hosting', '/add-component') and anticipates the use of 'curl' and 'bash'.
Sanitization: No explicit sanitization or validation of the remote Markdown content is performed before the agent is instructed to follow it.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch a capability catalog and documentation at runtime from an external URL ('https://basic-anteater-667.convex.site/'). This creates a runtime dependency on external infrastructure for the skill's logic.
[COMMAND_EXECUTION]: The workflow involves running local scripts ('/add-hosting' and '/add-component') based on the results of the remote fetch or user input. It also references the potential use of 'curl' and 'bash' in its execution environment.

convex-add