Loading ToolBench
Dual-Layer Context Management MCP Server providing session lifecycle management, OCTAVE compression, and AI agent binding with system governance injection
This server has moderate definition quality with significant gaps. All 5 tools have descriptions and basic input schemas, but parameter descriptions are sparse, output schemas are undocumented, and error handling lacks actionable recovery guidance. The tool names follow verb-noun conventions (clock_in, clock_out, bind, submit_review, submit_rccafp_record) which is good, but several parameter descriptions are too terse (<20 chars). Parameter constraints are often missing (no enums for role/record_type, no format specs). The descriptions explain WHAT the tools do but rarely explain WHEN to use them or what they return. This is typical of C-grade community servers.
Lightweight agent binding bootstrap for establishing agent identity and context
Register session start and return context paths for session initialization
Archive session transcript with OCTAVE compression and learnings indexing
Submit RCCAFP (Requirements, Contracts, Capabilities, Affordances, Fallback, Patterns) governance records
Submit architecture/code reviews to .hestai/ project documentation
Output schemas are not documented. No tool description states what fields are returned or how downstream tools should chain results.
Parameter 'role' in clock_in and submit_review lacks an enum constraint. Examples given ('implementation-lead', 'architect') suggest a closed set, but LLMs cannot infer valid values from freeform strings.
Parameter 'record_type' in submit_rccafp_record is described as enum values but not formalized in the schema, no enum array is present. The description lists (REQUIREMENT, CONTRACT, CAPABILITY, AFFORDANCE, FALLBACK, PATTERN) but these must be enforced in the schema.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Parameter descriptions are too short and lack actionable detail. E.g., 'Session role (implementation-lead, architect, etc.)' is 52 chars but omits dependency hints (e.g., 'Role determines which review permissions apply' or 'Must match VALID_ROLES; see submit_review').
No error handling guidance in tool descriptions. LLMs do not know: what to do if session_id is invalid? If transcript encoding fails? If governance injection is denied? Error responses should guide recovery (e.g., 'Invalid session_id. Call clock_in() first.').
Tool descriptions do not mention idempotency or side effects. Agents need to know: is clock_in safe to retry? Does clock_out have irreversible consequences? Does submit_review create a new record or update an existing one?
bind tool has an optional 'capabilities' array parameter with minimal documentation ('List of agent capabilities'). Lacks format/schema for what a capability looks like, maximum array length, valid examples, or dependency on agent_id.