The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/fix_transcript_enhanced.py contains a function find_glm_api_key that programmatically iterates through and reads sensitive shell configuration files in the user's home directory, including ~/.zshrc, ~/.bashrc, ~/.bash_profile, and ~/.profile. It searches these files for patterns matching API keys and tokens to extract them.
[REMOTE_CODE_EXECUTION]: Both SKILL.md and scripts/ensure_deps.py recommend or implement the download and direct execution of installer scripts for the uv tool from the Astral domain using piped shell commands (curl | sh and powershell | iex).
[COMMAND_EXECUTION]: The skill uses the subprocess module in several scripts to execute external system commands:
scripts/ensure_deps.py executes uv for virtual environment and dependency management.
scripts/fix_transcript_enhanced.py executes OS-specific commands (open, xdg-open) to launch web browsers and uv for script execution.
[EXTERNAL_DOWNLOADS]: The core logic in scripts/core/ai_processor.py and scripts/core/ai_processor_async.py makes outbound HTTP requests to external API endpoints at open.bigmodel.cn to perform text corrections.
[PRIVILEGE_ESCALATION]: The installation instructions in SKILL.md suggest bypassing system security policies using the Powershell -ExecutionPolicy ByPass flag to execute unverified remote scripts.
[DYNAMIC_EXECUTION]: The utility script scripts/utils/health_check.py uses the __import__ function to dynamically load modules from strings during dependency verification checks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data from external files and interpolates it into prompts sent to an LLM without adequate boundary markers or sanitization:
Ingestion points: scripts/cli/commands.py reads user-supplied Markdown and text files.
Boundary markers: Absent in the prompt templates used in scripts/core/ai_processor.py.
Capability inventory: The skill possesses the ability to execute shell commands, perform network operations, and write to the file system.
Sanitization: No filtering or validation is performed on the ingested file content before it is processed by the AI.

transcript-fixer