The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text to construct a prompt that is subsequently passed to another AI agent via a deep link. This creates a potential surface where malicious user input could influence the target agent's behavior. However, this risk is mitigated by the fact that the output is delivered as a link for the user's own consumption on the official app.amplitude.com domain, and the skill does not possess sensitive capabilities like file-writing or arbitrary command execution.
Ingestion points: User input captured during the 'intake' phase defined in SKILL.md (Step 4).
Boundary markers: The skill uses a bulleted list template for the prompt but does not include explicit delimiters or instructions to ignore embedded commands in the user-supplied fields.
Capability inventory: None. The skill generates text and URLs for the user; it does not execute system commands or perform unauthorized network requests.
Sanitization: User input is URL-encoded before being embedded into the final deep link to ensure the URL remains valid.
[SAFE]: The skill utilizes platform-specific Model Context Protocol (MCP) tools to retrieve project and event metadata. This usage is restricted to the local environment and is intended to ground the agent's suggestions in the user's actual data context.
[SAFE]: All external links and resources point to the official amplitude.com domain, which is the infrastructure owned by the skill's author. No unauthorized data exfiltration or third-party connections were identified.

create-custom-agent