The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads SKILL.md instruction files from external URLs specified in a manifest. While it enforces HTTPS and requires user confirmation, this mechanism allows the agent to fetch content from remote sources.
[INDIRECT_PROMPT_INJECTION]: The skill is a conduit for instruction-based extensions, creating a vulnerability to malicious input from remote sources.
Ingestion points: Reads .skills.json manifests and the content of downloaded instruction files (SKILL.md).
Boundary markers: The instructions rely on manual user review and confirmation rather than automated boundary delimiters or content sanitization.
Capability inventory: The skill has file-write access to the agent's skills directory and network-read access for fetching remote content.
Sanitization: Includes a blocklist and verifies the use of secure HTTPS protocols.
[COMMAND_EXECUTION]: The bash and PowerShell helper scripts execute local system commands such as git and python3 to resolve project directory paths and parse JSON configuration files.

team-skills