The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts to perform state management and auditing tasks.
scripts/filing_record.py initializes and maintains a markdown-based ledger (website-filing-record.md) to track the filing process.
scripts/profile_store.py reads and persists non-sensitive user and brand context in a local JSON profile.
[DATA_EXPOSURE]: The skill interacts with local configuration and filing records but implements significant privacy safeguards.
scripts/profile_store.py contains a SENSITIVE_PARTS filter (token, secret, password, credential, cookie, api_key, apikey) to programmatically prevent the storage of credentials.
The documentation and instructions (SKILL.md and references/authority-gates.md) mandate that full identity documents and session tokens must never be persisted.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external authoritative web pages and its own audit ledger.
Ingestion points: Data is read from the local website-filing-record.md and external government/provider filing platforms.
Boundary markers: Instructions require that only desensitized evidence summaries are recorded and that agent explanations be kept distinct from authoritative text.
Capability inventory: The skill can execute internal scripts via subprocess and perform file system writes for record-keeping.
Sanitization: scripts/filing_record.py includes a clean_cell function that sanitizes input by normalizing whitespace and escaping markdown delimiters.
[SAFE]: The skill is a legitimate utility for regulatory compliance. It follows security best practices such as atomic file writes, human-in-the-loop requirements for sensitive submissions, and explicit exclusion of private data from its operational logs.

lov-china-website-filing