The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of data from untrusted or external sources.
Ingestion points: The agent reads the 99_Meta/structure-doctrine.md file from the local vault to determine filing and naming conventions. It also fetches and processes data from the Meta Insights API (Ads Manager data).
Boundary markers: The instructions do not define any explicit boundary markers or security delimiters to distinguish between system instructions and data retrieved from external sources.
Capability inventory: The skill uses the lark-cli tool to modify database records and workflows and writes markdown files to the local filesystem using standard file-write operations.
Sanitization: There is no evidence of sanitization, escaping, or explicit instructions to disregard embedded commands within the ingested content.
[COMMAND_EXECUTION]: The skill makes extensive use of the lark-cli command-line tool to perform administrative actions on the Lark Base. This includes fetching resource tokens, listing records, updating field properties, and managing automated workflows.
[CREDENTIALS_UNSAFE]: Reference files within the skill contain a hardcoded Lark Base identifier (Yt2WbfTeQa0tjQsjMUwlfaEvgXb), which is explicitly labeled as the "Master base token." Although the skill uses this identifier as a safety mechanism to prevent the agent from accidentally modifying the vendor's template, hardcoding identifiers of live production resources is a suboptimal practice. The skill otherwise demonstrates good credential hygiene by ensuring that Meta System User tokens are never persisted in the vault or shell history.

performance-db-setup