Connect AI agents to 200+ enterprise SaaS systems via StackOne's AI Agent Integration Gateway. Provides tools for searching and executing StackOne actions across HRIS, ATS, CRM, LMS, and other systems.
Two tools with basic descriptions but critical schema and parameter documentation gaps. stackone_search has a simple string parameter with minimal guidance; stackone_execute accepts a JSON string parameter (parameters) with no schema validation hints, type constraints, or format documentation. Neither tool documents output structure. Error handling is minimal, no recovery guidance, no categorization of retryable vs fatal errors. Tool names follow verb_noun convention (search, execute) but descriptions lack WHEN/WHY context and dependency hints. No parameter descriptions beyond the tool docstring. Output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured data reliably.
Execute a StackOne action by name with the given parameters. Use this after stackone_search to run a discovered action. Pass the action_name exactly as returned by stackone_search, and provide the required parameters as a JSON string.
Search for available StackOne actions by natural language description. Use this tool to discover what actions are available before executing them. For example, search for "list employees" or "create job posting" to find relevant actions across connected HRIS, ATS, CRM, LMS, and other systems.
No output schema documentation. LLMs cannot predict return structure (fields, types, nesting). stackone_search returns {status, query, results_count, actions} but this is invisible to the agent, forces guessing on downstream field extraction.
stackone_execute 'parameters' input is a free-form JSON string with no format guidance, constraints, or examples. Description says 'JSON string' but does not explain what keys/structure are valid, required fields, or how to discover them. Forces LLM to guess parameter shape.
Error handling returns bare error messages ('Invalid JSON parameters: ...') with no recovery guidance. No distinction between retryable (malformed JSON) vs fatal (action not found) errors. LLM cannot decide whether to retry, ask user, or abort.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
stackone_search description lacks WHEN/WHY context and dependency hints. Does not explain that this must be called before stackone_execute, or what 'natural language description' means in the context of StackOne's 200+ actions. No guidance on query phrasing.
stackone_search 'top_k' parameter (default 5) is hardcoded in factory function, not exposed to agent. If agent needs more results, it cannot request them. Limits discoverability of available actions.