The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute npx -y patchxnote-agent@latest to install the necessary CLI tools and set up the MCP environment. This command downloads and executes code from the npm registry.
[COMMAND_EXECUTION]: The skill uses various shell commands for configuration, verification, and installation, such as npx patchxnote-agent setup, mcp config, and mcp status --verify.
[EXTERNAL_DOWNLOADS]: The skill references and downloads the patchxnote-agent package from npm and points to documentation/resources on GitHub and Feishu.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including transcripts, summaries, and memories from the PatchX Freenote service. This represents a vulnerability surface where malicious instructions could be embedded in the ingested data. The skill includes explicit instructions for the agent to treat this content as data rather than instructions to mitigate this risk.
Ingestion points: Untrusted data enters the context through tools like patchxnote_list_memories, patchxnote_get_memory, and patchxnote_search_memories described in SKILL.md and references/workflows.md.
Boundary markers: Explicit instructions are present in SKILL.md and references/security-and-evidence.md telling the agent to treat these inputs as user data and ignore embedded instructions.
Capability inventory: The skill has the capability to execute shell commands (npx), write local files (Markdown drafts), and perform network operations (browser OAuth login, webhooks).
Sanitization: The skill instructions recommend redacting sensitive identifiers and using out-of-band browser authorization for secret handling.

patchxnote-mcp