The Agent Skills Directory

[SAFE]: The skill is designed to manage Locus Workflows through an existing authorized connection, ensuring that all operations remain within the user's established security context.\n- [DATA_EXPOSURE]: Instructions and templates explicitly advise against including secrets, API keys, or personal credentials in the source code or configuration files. The workflow model leverages managed platform-level authentication.\n- [COMMAND_EXECUTION]: The skill uses dedicated platform tools (e.g., workflow_run, workflow_validate) to execute code in a remote, isolated sandbox environment with restricted network access, rather than local shell execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for processing untrusted data (Ingestion points: input objects in workflow.ts), which is mitigated by mandatory input and output validation boundaries (Boundary markers: defineWorkflow parsers). The assets demonstrate thorough validation and normalization (Sanitization: parseInput and parseIdentifiers in assets/gtm-enrichment/workflow.ts) before any Locus bindings are invoked (Capability inventory: row.call).\n- [REMOTE_CODE_EXECUTION]: While the skill involves remote code execution as its primary purpose, this is restricted to a fixed runtime environment (locus-typescript-1) using a specific vendor SDK (@withlocus/workflows), and the platform rejects arbitrary package installations or shell access.

locus-workflows