Thought threads for AI agents — create, track, and resolve ongoing lines of thinking. Plugin for cortex-engine.
Four tools with complete input schemas and descriptions. Naming follows verb_noun convention (thread_create, thread_update, thread_resolve, threads_list). Descriptions are adequate (50-100 chars) but lack LLM-optimized guidance on WHEN to use each tool and what happens on retry. Parameters are typed and mostly described, but lack enums for constrained values (kind, status). Output schemas are not documented, callers cannot predict response structure. Error handling returns basic error objects but lacks recovery guidance or actionable next steps. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear write/read semantics.
Create a new thought thread for tracking ongoing questions, explorations, or topics.
Resolve a thread with a resolution note.
Update a thread — change status, edit title/body, add session refs, link memories.
List thought threads. Filter by status (default 'open'), project, kind, or tag.
Output schemas not documented. Callers cannot predict response structure (e.g., does thread_create return full thread object or just {id, title}?). LLMs cannot plan downstream tool calls without knowing what fields are available.
Constrained parameters lack enums. 'kind' accepts 'work|exploration|creative|revenue|meta' and 'status' accepts 'open|active|blocked|parked|resolved', but these are documented only in descriptions, not as JSON Schema enums. LLMs may hallucinate invalid values.
Error responses lack recovery guidance. Errors return {error: 'message'} but do not tell the LLM what to do next. E.g., 'Thread not found: xyz' should suggest 'Call threads_list() to find valid thread IDs.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
No tool annotations. thread_create and thread_update are destructive (write state), thread_resolve is irreversible, but no destructiveHint or idempotentHint annotations present. Agents cannot reason about retry safety.
Descriptions lack WHEN-to-use guidance. 'Create a new thought thread' is functional but does not explain when to call this vs thread_update, or what happens if called twice with the same title. LLMs need explicit selection criteria.