The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, specifically user tasks in IM platforms and code diffs via GitHub webhooks as defined in configs/feishu/agents/duchayuan.md and configs/feishu/agents/silijian.md.
Ingestion points: User-provided task descriptions and automated code diffs from version control systems.
Boundary markers: The agent instructions lack explicit delimiters or safety instructions to treat external data as untrusted, increasing the risk of the agent following instructions embedded within the data.
Capability inventory: The 'silijian' agent has the capability to execute local node scripts (node scripts/task-store.js) and send messages to external channels via the message tool.
Sanitization: No sanitization or escaping mechanisms are specified for processing external inputs before they are used as arguments for script execution or communication tools.
[COMMAND_EXECUTION]: The skill utilizes local shell and node.js scripts to perform its core functions.
Evidence: The scripts/doctor.sh script is used to validate the installation by checking local file existence and parsing JSON content. The silijian.md prompt instructs the agent to use node scripts/task-store.js to manage task states. These are intended behaviors but represent a capability surface that relies on the agent's internal logic for safe parameter handling.

ai-court