v8-isolate hypervisor on workerd — declarative route table, cryptographically-verifiable identity substrate (Interlace), capability-shaped credential distribution. MCP is one tenant.
Cloister exposes 9 tools with complete JSON schemas and basic descriptions. However, descriptions are uniformly terse (10-30 chars), parameter descriptions are entirely absent, and output schemas are not documented. Tool naming follows verb_noun convention (bead_create, bead_list, bead_search) but lacks clarity on domain-specific terminology ('bead' is unexplained). No error handling guidance, no parameter constraints beyond enums on state/priority, and no indication of idempotency or side effects. Schemas are well-formed but descriptions fall far short of the 50-200 char LLM-optimized baseline. This is a typical community server: functional but not production-ready for agent reasoning.
Close a bead (issue/task)
Add a comment to a bead
Create a new bead (issue/task)
List beads (issues/tasks) in a repository
Search for beads in a repository
Update a bead (issue/task)
Enrich code analysis with additional passes
Reparse source code in a specified language
Get the status of the cloister server
All tool descriptions are under 20 characters (e.g., 'Close a bead (issue/task)', 'Add a comment to a bead'). LLMs cannot determine when/why to select tools from these terse labels.
No parameter descriptions anywhere. Schema declares types (string, integer, array) but LLMs cannot infer meaning of 'repo', 'pass', 'lang', 'author' without descriptions.
Domain terminology 'bead' is never explained. LLMs will not understand that 'bead' means 'issue/task' without explicit definition in tool descriptions or server documentation.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
No output schemas documented. Tools return data but LLMs have no specification of response structure, field names, or types. Cannot plan downstream tool calls or extract required IDs.
No error handling guidance. Tools declare no recovery paths, retryability, or actionable error messages. LLMs cannot self-correct on failure.