The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions explicitly direct users to store Zapier webhook URLs in plain text files within the references/zaps.md file. As noted in the documentation, these URLs serve as authentication tokens that allow anyone with the URL to trigger the associated workflows.
[EXTERNAL_DOWNLOADS]: The skill facilitates the connection of external MCP servers via mcp.zapier.com. This is a well-known service used for its intended purpose of workflow automation.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute curl commands, enabling it to perform POST requests to external webhook URLs for workflow orchestration.
[INDIRECT_PROMPT_INJECTION]: The skill implements a 'Self-Improvement Protocol' where the agent is instructed to use the Edit tool to modify its own reference files (references/zaps.md and references/mcp-patterns.md) based on user feedback.
Ingestion points: Untrusted user instructions provided during conversation and the content of the reference files themselves.
Boundary markers: Absent; the skill relies on the agent's interpretation of natural language instructions to determine what to persist.
Capability inventory: The skill can execute shell commands via bash (specifically curl) and modify local files using the Edit tool.
Sanitization: Absent; the skill is designed to persist user-provided workflow patterns and URLs directly into its long-term configuration files.

zapier-workflows