The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and installs the third-party package @jinzcdev/markmap-mcp-server from the NPM registry during the bootstrap process as part of its initialization workflow.
[COMMAND_EXECUTION]: The bootstrap.sh and bootstrap.ps1 scripts execute multiple shell commands, including npm, npx, and CLI tools for specific agent hosts (claude, codex, opencode), to facilitate software installation and MCP server registration.
[DYNAMIC_EXECUTION]: The bootstrap scripts employ node -e to execute dynamically generated JavaScript snippets for the purpose of programmatically reading and modifying configuration files such as .claude.json, config.toml, and opencode.json.
[PRIVILEGE_ESCALATION]: The installation instructions for Windows platforms recommend running the PowerShell bootstrap script with the -ExecutionPolicy Bypass flag, which allows the execution of unsigned scripts and bypasses system security protections.
[PERSISTENCE]: The skill automatically modifies the host agent's local configuration files (e.g., ~/.claude.json or ~/.codex/config.toml) to persistently register the markmap-mcp-server as a tool that persists across different user sessions.
[INDIRECT_PROMPT_INJECTION]: The skill accepts arbitrary user-supplied descriptions through #$ARGUMENTS and processes them into a Markdown structure to be passed to an external tool. This ingestion of untrusted data provides a surface for indirect instructions to reach the underlying tool context.
Ingestion points: Processes user input via the #$ARGUMENTS variable defined in SKILL.md.
Boundary markers: Absent; instructions direct the agent to reformat input without explicit isolation markers.
Capability inventory: The skill possesses file-write and command-execution capabilities through its bootstrap scripts and maintains access to the markmap-mcp-server tool.
Sanitization: No explicit sanitization or validation of the input content is performed beyond structural reformatting.

mindmap