The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses specific directives to constrain the agent's output, requiring one-line responses and prohibiting summaries or additional questions unless a step fails. These instructions override default agent behavior to enforce a specific user experience.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it analyzes the user's codebase to discover workflows.
Ingestion points: Codebase analysis in rules/discover-workflows.md
Boundary markers: Absent
Capability inventory: Command execution (npm, git, sed), file writing, and vendor-specific MCP calls
Sanitization: Absent
[PROMPT_INJECTION]: The instructions in rules/connect-knock-cli.md contain a directive to not mention or configure MCP connectors when using the CLI path. While this logic limits user information, it functions as a guard to prevent redundant or conflicting configurations.
[COMMAND_EXECUTION]: The skill utilizes shell commands like npm install, git branch, sed, and grep to perform environment setup and manage local configuration files such as ~/.cursor/mcp.json and .env.local.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of official vendor tools, including the Knock CLI (@knocklabs/cli) and supplementary skills (knocklabs/skills), using standard package managers.
[SAFE]: The skill correctly handles API keys by directing the agent to prompt for user input and save it to a local .env.local file, with explicit instructions to avoid echoing the secret or committing it to version control.

knock-setup