day2-mcp-and-context-sync

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill defines a "STOP PROTOCOL" that strictly controls agent behavior during the tutorial, including specific instructions to refrain from calling certain tools (like AskUserQuestion) until specific conditions are met to ensure a structured lesson flow.
  • [PROMPT_INJECTION]: Category 8: Indirect Prompt Injection risk. The skill facilitates the creation of a "Context Sync" tool (templates/context-sync.md) that aggregates data from external, potentially untrusted sources.
  • Ingestion points: Data is pulled from Slack messages, Notion databases, Linear issues, and Gmail/Google Calendar via various MCP tools (e.g., slack_read_channel, notion__search).
  • Boundary markers: The provided template and instructions do not specify the use of delimiters or "ignore" instructions for the ingested content when presenting it to the agent.
  • Capability inventory: The skill utilizes subagents and MCP tools with the ability to read/write files and execute external service commands.
  • Sanitization: No explicit sanitization, filtering, or escaping of the retrieved data is implemented in the provided scripts or templates.
  • [REMOTE_CODE_EXECUTION]: The tutorial instructs the agent to run npx skills add ai-native-camp/camp-2 --agent claude-code --yes to install the curriculum. This command downloads and executes code from the vendor's repository to set up the learning environment.
  • [EXTERNAL_DOWNLOADS]: The scripts/mcp_servers.py script makes outbound HTTP requests to raw.githubusercontent.com to retrieve a list of MCP servers. This functionality is intended to help users discover and configure tools from the official MCP server registry.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 10:55 AM