Loading ToolBench
Task-complete context packs for AI coding agents. CLI-first context broker with optional MCP.
ScopeKit provides 12 well-named, verb-prefixed tools with clear descriptions and documented input schemas. Tool names follow verb_noun convention (context_pack, graph_query, search_code). Descriptions are detailed and include usage guidance (e.g., 'PRIMARY TOOL', 'FALLBACK ONLY'). However, output schemas are not documented in the source, no return type specifications visible. Parameter descriptions are present but some lack format/constraint details. Error handling guidance is minimal. The server is STDIO-only, which caps protocol readiness but does not directly penalize definition quality.
PRIMARY TOOL. Use first for discovery, debugging, edit planning, and test planning. Prefer over repo_map/search_code/graph_query. Returns the smallest useful context package within budgetTokens.
Check graph/context cache status. Use before context_pack when unsure indexing is ready. Suggests npm run graph:build / context:build if missing.
Likely test/lint/dev/install commands. Fallback when context_pack omits commands.
Exact function/class snippets (compact). Use only after context_pack identifies a symbol and exact implementation verification is required.
Lower-priority: expand graph neighbors for a node, path, or symbol. Keep compact.
Lower-priority: find short paths between files/symbols in the graph.
Output schemas not documented. No return type specifications visible in source for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
budgetTokens parameter lacks format guidance. Accepts 300-2500 but no description of what happens at boundaries, how tokens are counted, or what 'insufficient budget' returns.
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or actionable error messages. E.g., what if graph index is missing? What if symbol not found?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
FALLBACK ONLY. Use when context_pack is insufficient or when graph-level detail is explicitly requested.
Read-only graph index status. Run npm run graph:build if missing.
FALLBACK ONLY. Symbol lookup in graph (path/line/neighbors) when context_pack is insufficient or graph details are explicitly requested.
Use for changed-files or diff tasks: likely dependents, related tests/commands, and risk. Call after context_pack when change impact is needed.
LAST-RESORT FALLBACK. Compact repo tree/scripts for missing cache or insufficient context/graph results. Do not use first when context_pack is available.
LAST-RESORT FALLBACK. Use only when context/graph tools are missing or insufficient. Do not use first when context_pack is available.
maxResults parameter descriptions lack clarity on sorting/ranking. When graph_query returns 20 results, are they ranked by relevance, recency, or arbitrary order? LLMs cannot reason about result quality.
No pagination guidance. Tools accept maxResults but no documentation of how to iterate through large result sets or whether cursors/offsets are supported.