The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies heavily on shell command execution for its primary functionality.
pipeline_orchestrator.py, mswnlz_publish.py, and url_router.py use subprocess.run to orchestrate local Python scripts and execute Git commands (git clone, git pull, git add, git commit, git push).
trigger_site_rebuild.sh executes npm run build and npm run validate to verify the state of the target GitHub Pages site before triggering a rebuild.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by processing external data.
Ingestion points: Resource titles and URLs are ingested from items.json (user/external input) in pipeline_orchestrator.py and quark_batch_run.py.
Boundary markers: The skill does not use explicit boundary markers for the classification logic, though it uses a regex-based classification system.
Capability inventory: The skill has the ability to write to local Git repositories and push changes to remote GitHub repositories (mswnlz_publish.py).
Sanitization: mswnlz_publish.py performs basic sanitization by replacing Markdown brackets [ and ] with full-width versions 【 and 】 to prevent injection into the generated Markdown links.
[DYNAMIC_EXECUTION]: The script copy_promo_to_folders.py uses ast.literal_eval in the parse_cookies function to process cookie strings. While safer than eval(), this is a form of dynamic data parsing.
[EXTERNAL_DOWNLOADS]: The documentation (README.md) instructs the user to clone third-party tools such as QuarkPanTool, BaiduPCS-Go, and aligo. These are legitimate dependencies for the skill's stated purpose of cross-drive resource management.

quark-mswnlz-publisher