The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a procedural pattern in templates/prompt_template.md where it explicitly instructs the AI agent to write a shell script (.sh) to the debug_log directory and subsequently execute that script to perform data operations. Generating and executing code at runtime is a high-risk pattern as it can be exploited if the generation logic is influenced by malicious input.
[COMMAND_EXECUTION]: The skill relies on shell commands to manage its workflow. It instructs the agent to use sed to update task status in templates/prompt_template.md and uses the memocli binary for database interactions. The script scripts/generate_prompt.py also spawns subprocesses to capture help text from the memocli tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted chat logs which are interpolated directly into the sub-agent's prompt in templates/prompt_template.md using the {{current_content}} placeholder.
Ingestion points: Reads markdown chat logs from the user-provided source_dir via scripts/generate_prompt.py.
Boundary markers: Uses XML-style tags like <当前的讨论> to delimit content, but lacks explicit instructions for the agent to ignore commands within those tags.
Capability inventory: File system write access (creating .sh scripts), shell execution (running the generated scripts and sed), and repository management via memocli.
Sanitization: There is no evidence of escaping, filtering, or sanitizing the chat content before it is placed into the prompt template.

im-wiki-extractor