The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage the file system and Git repositories. Evidence includes:
Usage of mkdir -p, cp -r, and rm -rf .git to manipulate local directories.
Execution of gh repo create and git push for repository management.
Execution of local scripts such as scripts/clean_sample_data.py and gh-clone.sh using uv run and direct shell calls.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to command injection through the interpolation of untrusted user data into shell execution strings. Evidence includes:
Ingestion points: User inputs collected in Step 2 (專案路徑, 專案代號) and the source PDF path provided in the initial command.
Boundary markers: None are defined to isolate user-supplied strings from the shell command structure.
Capability inventory: The skill has the capability to write files, delete directories (rm -rf), and execute arbitrary shell commands/scripts.
Sanitization: There is no logic provided to sanitize or escape characters like semicolons, backticks, or pipes in user-provided strings before they are used in commands like "$CLONE_SCRIPT" <project_name> or mkdir -p data/pdfs; cp "<pdf_path>" data/pdfs/.

new-project