The Agent Skills Directory

[PROMPT_INJECTION]: The instructions explicitly direct the agent to avoid summarizing its actions or listing installed tools and skills (e.g., 'Do not summarize or list installed tools/skills' and 'keep every response to one short line'). While intended to reduce verbosity, these constraints conceal automated modifications to the user's environment from the chat history.\n- [INDIRECT_PROMPT_INJECTION]: The skill analyzes local codebase files to discover product workflows and research data structures. It lacks explicit boundary markers (delimiters) or instructions to ignore embedded prompts in the ingested files.\n
Ingestion points: rules/discover-workflows.md (codebase examination for product understanding) and rules/import-users.md (codebase research for user data mapping).\n
Boundary markers: Absent. No delimiters or 'ignore' instructions are used for external content.\n
Capability inventory: Global package installation (npm install -g), file writing (.env.local), git branch creation, and network-enabled MCP tool calls.\n
Sanitization: Absent. Data from the codebase is directly used to generate workflow proposals and implementation scripts.\n- [COMMAND_EXECUTION]: The skill executes various shell commands, including the global installation of the Knock CLI, the creation of git branches, and the automated configuration of local environment files (.env.local) using shell redirection and read commands.\n- [EXTERNAL_DOWNLOADS]: Downloads vendor-specific tools and dependencies, including the @knocklabs/cli and @knocklabs/node packages, as well as additional agent skills from the npm registry. These are official vendor resources related to the skill's primary purpose.

knock-setup