The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md (Step 3) explicitly direct the agent to verify the validity of project commands by executing them. The instruction states that commands such as build, test, and lint should be 'actually run and confirmed' (实跑确认). This allows for the execution of arbitrary code defined within a project's configuration files (e.g., package.json, Makefile).
[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data to generate project rules and performs sensitive operations based on that data.
Ingestion points: In SKILL.md, the agent reads project files like AGENTS.md, README.md, and configuration files (Step 3), as well as processing the conversation history for user-defined 'lessons' (Step 1).
Boundary markers: The skill emphasizes that all file modifications and movements must be presented as a proposal and receive explicit user approval before execution (Step 5, Step 7).
Capability inventory: The agent is empowered to execute shell commands (Step 3), write/edit project files (Step 5), update agent memory and progress logs (Step 6), and move or rename files (Step 7).
Sanitization: There is no explicit content filtering; the skill relies on the agent's distillation process and final human approval of proposed changes.
[EXTERNAL_DOWNLOADS]: The README.md file provides instructions for installing the skill from the vendor's GitHub repository using standard tools like npx and git clone. These references are consistent with the identified author and represent the intended installation method for the skill.

agentsmd