The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network requests to official Chinese government websites (including gov.cn and ccopyright.com.cn) via scripts/check_rules.py to verify software copyright registration rules. It also contains logic in SKILL.md to perform a daily check for upstream skill updates.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes content from the user's local project files to build evidence maps and documentation. This creates a surface for indirect prompt injection if the project contains malicious instructions designed to mislead the agent during drafting.
Ingestion points: Project source code and documentation are processed by scripts/build_evidence_map.py and scripts/scan_project.py.
Boundary markers: The skill uses explicit guardrails (prompts/shared/guardrails.md) that forbid the agent from guessing legal or human facts (like ownership or completion dates), requiring human confirmation via user-confirmations.json instead.
Capability inventory: The skill has the ability to read project files, write generated materials to a task-specific output directory, and perform network fetches for rule registries.
Sanitization: The skill implements a defensive security measure in scripts/detect_secrets.py, which heuristically scans generated documents for API keys, private keys, and passwords, redacting them before finalization.
[DYNAMIC_EXECUTION]: The skill uses scripts/render_source_docx.py and scripts/render_source_pdf.py to assemble and render formatted documents. The PDF generation relies on the standard reportlab Python library. These operations are restricted to generating documentation from project source copies and do not involve executing untrusted project code.

copyright-forge