The Agent Skills Directory

[SAFE]: The skill implements a secure identity management model by delegating all authentication and API interactions to lark-cli, a local dependency. No credentials or access tokens are handled directly by the skill scripts or stored in metadata.
[SAFE]: A dedicated preflight script (scripts/preflight_publish.py) enhances security by scanning source documents for sensitive information like absolute system paths and calculating cryptographic hashes to ensure content integrity throughout the synchronization process.
[SAFE]: The profile persistence mechanism (scripts/profile_store.py) includes specific security controls, such as atomic write operations and a blocklist for sensitive keys (e.g., tokens, secrets, credentials), preventing the storage of secrets in local configuration files.
[SAFE]: High-risk operations, specifically changing document visibility to 'anyone_readable', are protected by a mandatory verification loop. The skill requires explicit user confirmation and performs an immediate readback of the remote permission state to confirm the desired security posture.
[INDIRECT_PROMPT_INJECTION]: The skill ingests local Markdown files which represent an attack surface for indirect instructions. However, the skill's operational scope is limited to document management through the user's local authorized environment, and it treats content as verbatim data for synchronization purposes, which is a standard and acceptable risk for this use case.
[COMMAND_EXECUTION]: The skill instructions involve executing local scripts and lark-cli commands. These commands are well-structured, purpose-built for document synchronization, and operate under the user's local context without utilizing elevated privileges.

lov-publish-feishu-doc