git for agent brains — debug, time-travel, and migrate AI agent state across 10 frameworks
StateWeave's three tools have strong, domain-specific descriptions that explain WHEN and WHY to use them (addressing pattern:tool-description well). All three tools have complete input schemas with type definitions and parameter descriptions. However, output schemas are NOT documented in the source code, responses are returned as dicts with fields like 'success', 'payload', 'warnings', 'error', but the structure is never formally specified. Tool naming follows verb_noun convention and is clear. Parameter descriptions are detailed and explain constraints (e.g. framework enum with 10 values). A critical gap: NO tool provides output schema documentation, forcing LLMs to infer result structure. Additionally, error handling is basic (success/error fields) but lacks recovery guidance per pattern:recovery-guide. The server exports/imports agent state across frameworks, a specialized use case with excellent domain documentation but weak result documentation.
Compare two agent states and return a detailed diff report. Use this when you need to verify a migration succeeded, audit what changed between two checkpoints, or detect state drift after an operation. Shows exactly what was added, removed, or modified. Computes the structural differences between two StateWeavePayloads, showing what changed in conversation history, working memory, goals, and tool results.
Export an agent's cognitive state from a framework. Use this when an agent needs to save its state before shutting down, migrate to a different framework, create a backup before a risky operation, or share its learned context with another agent. Extracts the agent's full cognitive state (conversation history, working memory, goals, tool results) and packages it in the Universal Schema format.
Import an agent's cognitive state into a target framework. Use this when an agent needs to resume in a different framework, restore from a backup, or receive context from another agent. The payload is validated against the Universal Schema before import. Takes a StateWeavePayload (in Universal Schema format) and translates it into the target framework's native representation.
Output schemas are not documented. Tools return dicts with fields like 'success', 'payload', 'warnings', 'error', but the structure of 'payload' and detailed result fields are never formally specified in schemas or descriptions. LLMs cannot plan downstream operations or validate responses without knowing the result structure.
Error handling is minimal. Responses use generic success/error fields ('success': false, 'error': 'message') but provide no recovery guidance. Per pattern:recovery-guide, errors should tell the LLM what to do next (e.g. 'Framework not found. Available: langgraph, mcp, crewai...' is good, but doesn't guide remediation for validation errors).
No output pagination or result limiting documented. If export_agent_state returns a large StateWeavePayload with extensive conversation history, the response could exceed token limits. No mention of pagination, chunking, or size constraints in descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Parameter 'payload' in import_agent_state is type 'object' with only a brief description. The StateWeavePayload structure is not documented in the schema, LLMs have no guidance on what fields are valid or required within the payload dict.