The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README.md provides instructions for users to install the full suite of tools from the author's GitHub repository using npx and clawhub commands.
[REMOTE_CODE_EXECUTION]: The installation instructions include commands that fetch and execute remote code (npx -y skills add maojiebc/majia-siyu-team). These are standard installation procedures for the vendor's environment.
[COMMAND_EXECUTION]: The SKILL.md defines a workflow utilizing SiyuRuntime.plan() to dynamically select and execute sub-skills based on user intent and conversation context.
[DATA_EXFILTRATION]: The skill implements local data management features such as siyu-save and siyu-report, which are designed to record and aggregate customer findings within the agent's local environment.
[PROMPT_INJECTION]: The skill contains specific behavioral guidelines to ensure the agent remains focused on its primary function, such as explicitly refusing to engage in off-topic 'chitchat'.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a high-level intent router, ingesting untrusted user input to drive the selection and execution of specific sub-skills.
Ingestion points: User input retrieved through triggers defined in SKILL.md (e.g., /siyu, '/私域').
Boundary markers: The instructions lack explicit delimiters or specific guardrails directing the model to ignore potentially malicious instructions embedded in the user data.
Capability inventory: The skill routes to specialized tools capable of performing file writes (siyu-save) and other operational tasks.
Sanitization: There are no explicit data validation or sanitization steps defined for user input before it is processed by the routing logic.

siyu