The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's preamble section executes a shell script located at a computed path relative to the skill directory: bash "$(dirname "${BASH_SOURCE[0]}")/../../check-update.sh".
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user feedback data, which is passed directly to multiple sub-agents without proper isolation or sanitization.
Ingestion points: User feedback is collected via file reads, direct text input, or web scraping using tools like mcp__exa__web_search_exa and mcp__exa__web_fetch_exa in Step 1.
Boundary markers: The Subagent Prompt Templates (e.g., for Feedback Categorization, Sentiment Analysis) do not implement delimiters or 'ignore embedded instructions' warnings for the data labeled as '[反馈数据]' (feedback data).
Capability inventory: The primary agent and its sub-agents have access to sensitive tools including Bash, Write, and Agent (sub-agent delegation).
Sanitization: There is no logic present to sanitize, escape, or validate the content of the external feedback before it is interpolated into agent prompts.
[EXTERNAL_DOWNLOADS]: The skill incorporates external data retrieval capabilities via WebSearch and Exa tools to crawl application stores and social media for user reviews.

pm-feedback