The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/configure_mcp.py script executes the claude CLI via subprocess.run to add and manage MCP server configurations in the agent's local environment, allowing the skill to programmatically alter the agent's execution environment.
[CREDENTIALS_UNSAFE]: The scripts/configure_mcp.py script collects API keys and transmits them via an Authorization header in a POST request to a user-provided endpoint for connectivity testing. This creates a risk of credential exposure if the endpoint is directed to an untrusted or malicious server.
[EXTERNAL_DOWNLOADS]: The skill uses curl for uploading file attachments to remote servers and employs Python's urllib to perform network handshakes with external endpoints.
[DATA_EXFILTRATION]: Instructions in SKILL.md guide the agent to use curl to transmit local binary data to external Outline Wiki instances.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its broad capabilities and remote data ingestion.
Ingestion points: The skill fetches Markdown document content and metadata from remote Outline Wiki instances.
Boundary markers: No explicit delimiters or boundary markers are used when processing retrieved document content.
Capability inventory: The agent can execute shell commands (curl, claude), modify its own core configuration (~/.claude.json), and perform network operations via urllib.
Sanitization: No sanitization, filtering, or validation is performed on the content retrieved from the wiki before it is processed or used to drive agent actions.

outline-wiki-management