The Agent Skills Directory

[SAFE]: The skill follows security best practices for managing access control. It uses a standalone Python script to generate a deterministic change plan, enforces a mandatory approval step by the user before executing any mutations, and explicitly prohibits the storage or logging of sensitive credentials. It also implements a 'least privilege' approach by preserving existing user roles and limiting data exposure to source-relevant accounts.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from Feishu spreadsheets, which represents an indirect prompt injection surface. This risk is effectively managed through several defense layers:\n
Ingestion points: Data is fetched from Feishu spreadsheets using the lark-sheets tool, as described in Step 2 of the mandatory execution sequence.\n
Boundary markers: The skill explicitly treats spreadsheet content as 'requested state' rather than 'execution authorization' and requires the user to review and approve a plan hash in a separate turn (Step 6) before any changes are applied.\n
Capability inventory: The skill employs targeted ae-cli commands for permission and member updates, restricted to a single verified project scope.\n
Sanitization: The included script scripts/build_sync_plan.py performs rigorous validation of input fields, deduplication, and conflict detection. It translates source data into structured, narrow equality predicates (e.g., platform = <value>) rather than complex filters, preventing the injection of unintended logic into the target platform.

ae-feishu-data-permission-sync