The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of remote instructions and scripts from external sources, including GitHub repositories and arbitrary HTTPS URLs. It uses commands like npx skills add and curl to fetch content and place it into the agent's local skills directory (~/.claude/skills/).
[COMMAND_EXECUTION]: The skill executes arbitrary shell commands defined in the [[verification.checks]] section of the agent.toml manifest file. This allows any command present in the repository's configuration to be run when the user invokes the verification mode.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch remote configuration and skill files from external domains, including GitHub and custom Git hosting instances (e.g., Gitea, GitLab), as specified in the repository's manifest.
[DYNAMIC_EXECUTION]: The skill dynamically generates and modifies tool-specific configuration files, such as .claude/settings.json and .mcp.json, based on definitions found in the agent.toml manifest.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the agent.toml file located in the repository root to drive its execution logic and command construction.
Ingestion points: Reads the agent.toml manifest file from the current repository.
Boundary markers: The skill does not implement specific delimiters or safety warnings for content parsed from the manifest.
Capability inventory: Includes file system writes, shell command execution via npx and bash, and network operations via curl.
Sanitization: No explicit sanitization or validation of the commands, sources, or version strings provided in the manifest is described.

agent-bootstrap