The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches agent instructions from a remote host via ae-cli agent +get-agent-context and incorporates them into the local task guidance. This creates a surface for indirect prompt injection where malicious external instructions could influence the agent's behavior.
Ingestion points: Instructions are retrieved from a user-supplied host using ae-cli command output (found in SKILL.md and references/local-agent.md).
Boundary markers: The instructions are integrated into the task guidance hierarchy; however, no specific delimiters or boundary warnings are mandated for the remote content.
Capability inventory: The skill uses ae-cli for shell execution, file writing (via --output), and accessing sensitive user credentials.
Sanitization: The skill explicitly instructs the agent to inspect scripts before any execution and to ensure that sensitive credential output remains out of the chat conversation and logs.
[COMMAND_EXECUTION]: The skill relies on multiple ae-cli shell commands to manage agent discovery, authentication status, and retrieval of remote content (references, scripts, and assets).
[EXTERNAL_DOWNLOADS]: The documentation guides the installation of the ae-cli tool from both the public NPM registry and a private vendor registry at https://npm.thinkingdata.cn:3443. These resources originate from the skill's infrastructure and are intended for its primary function.
[DATA_EXFILTRATION]: The skill provides a flow for exporting MCP credentials using the --include-mcp-credentials flag. Although the skill includes instructions to handle this data safely, the presence of functionality to retrieve and process sensitive connection secrets constitutes a high-value data access capability.

ae-use-agent