The Agent Skills Directory

[SAFE]: The skill is authored by forcedotcom and facilitates legitimate Salesforce metadata analysis using established tools.
[COMMAND_EXECUTION]: The skill executes sf CLI subcommands using subprocess.run without a shell. Input validation is enforced at multiple layers using fs_guard.py, which utilizes regular expressions to restrict identifiers to alphanumeric characters and underscores, effectively preventing command and SOQL injection.
[CREDENTIALS_UNSAFE]: The skill manages Salesforce access tokens securely. It includes specific logic in rest_client.py and write_emit_ctx.py to redact tokens from error messages and logs. It also features a custom HTTP redirect handler that strips the Authorization header when a request is redirected to a different hostname, preventing credential leakage.
[EXTERNAL_DOWNLOADS]: The skill communicates with official Salesforce API domains (salesforce.com) to retrieve metadata related to the agent architecture.
[PROMPT_INJECTION]: Instructions within SKILL.md are focused on operational triggers and do not attempt to override agent safety protocols or extract system prompts.

agentforce-architecture-analyze