Persistent Markdown+git memory for AI agents (MCP) with hybrid search, stealth web fetch/search (obscura + on-demand SearXNG), cross-domain execution-discipline skill, token-saver hooks, local usage doctor and spec-builder
6 tools with mixed quality. Naming is verb-based and clear (vault_*, obscura_*, vkm_*). Descriptions exist but are inconsistent: vault_pg_status has a complex, jargon-heavy description (unclear when to call it); vault_graph_hops and vault_timeline have good descriptions but lack actionable error guidance; vault_hybrid_search and obscura_fetch lack parameter constraints (no limit bounds, no enum for readability); vkm_downloads has minimal description. Input schemas are present but incomplete: most parameters lack type constraints (e.g., 'direction' in vault_graph_hops should be enum, 'limit' parameters lack min/max bounds). No output schemas documented. Error handling is mentioned in descriptions but not formalized. Overall: definitions are functional but fall short of production-grade clarity.
Stealth web fetch with readability extraction and optional SearXNG integration
GET /api/graph. Never starts the service (read tools stay side-effect-free); disabled or not-running → uniform actionable error
Hybrid FTS5 + vector (semantic) search across Markdown vault with optional reranking
Disabled → the how-to-enable message (a normal result, not an error — status is the tool you ask when unsure); enabled → best-effort ensureServiceRunning, then GET /api/health shaped by formatStatus
GET /api/timeline. Never starts the service (read tools stay side-effect-free); disabled or not-running → uniform actionable error
Manage downloads with metadata extraction and storage
vault_pg_status description is jargon-heavy and unclear about when to call it. Phrase 'Disabled → the how-to-enable message' is cryptic; LLMs cannot infer intent.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does vault_graph_hops return? Does it include node IDs for follow-up queries?).
Numeric parameters lack bounds. 'limit' in vault_graph_hops (1-200 stated in description but not in schema), vault_timeline (no bounds), vault_hybrid_search (no bounds). 'depth' in vault_graph_hops (1-4 stated in description but not enforced in schema).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
'direction' parameter in vault_graph_hops should be an enum (out|in|both) but is a free-form string. LLMs may hallucinate invalid values like 'bidirectional' or 'forward'.
vkm_downloads description is minimal ('Manage downloads with metadata extraction and storage'). Does not explain what metadata is extracted, what 'storage' means, or when to use it vs obscura_fetch.