The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the sf CLI (Salesforce CLI) to interact with the Tooling API. Commands are restricted to sf data query, sf api request rest, and sf org display. These are standard tools for Salesforce metadata management and are executed through a Bash environment that respects the user's existing authentication session.
[DATA_EXPOSURE]: The skill explicitly forbids printing raw access tokens or full sf org display output. It also includes logic to mask credentials in summary reports (e.g., https://user:***@host/...) even if they are present in the underlying configuration data, preventing accidental exposure in logs or UI outputs.
[REMOTE_CODE_EXECUTION]: While the skill executes local Node.js scripts (plan-phases.mjs, map-metadata-key.mjs, etc.), these are packaged as internal utility scripts for deterministic logic (phasing, field mapping, result classification). There is no pattern of downloading and executing remote code or scripts from unverified sources.
[PROMPT_INJECTION]: The instructions contain clear boundary markers and specific procedural steps that the agent must follow. It includes a "STOP" section requiring the agent to read and validate the configuration file end-to-end and confirm the target org alias before proceeding, reducing the risk of erratic behavior due to input data.
[PRIVILEGE_ESCALATION]: The skill operates within the permissions granted to the Salesforce CLI user. It does not attempt to gain elevated system privileges (like sudo) or bypass Salesforce's internal CRUD/FLS/Sharing models, as all SOQL and REST calls are routed through standard API channels.

automation-sandbox-post-copy-configure