Loading ToolBench
An intel layer for AI coding agents that assumes its own rules might be wrong — and keeps the evidence. Authoritative north-stars the agent must cite, a task-core written before compaction eats it, review by an expert in YOUR domain, and an installer that verifies its own claims instead of printing success.
Bearing exposes 6 GitNexus code-analysis tools with well-structured schemas and detailed descriptions. All tools have clear names (verb_noun pattern: gitnexus_query, gitnexus_context, etc.), comprehensive parameter documentation, and explicit enum constraints (e.g., direction, mode). Descriptions are 150 - 250 chars, exceeding the 10 - 1024 baseline and providing context on WHEN to use each tool. However, output schemas are not documented in the visible source, the Python tool definitions show input schemas but no return type specifications. Error handling guidance is absent; tools do not indicate what failures look like or how to recover. No tool annotations (readOnlyHint, destructiveHint) are present, though all 6 are READ_ONLY by design. Parameter descriptions are strong (e.g., 'Cypher query statement' with schema reference), but missing are examples of expected output structure and pagination guidance for tools that may return large result sets.
Get a 360-degree view of a code symbol: callers, callees, references, process participation, and file location. Use after query to drill into specific symbols.
Run a Cypher query on the code knowledge graph. Use for precise structural questions: field read/write (ACCESSES), N-hop call chains (CALLS), method overrides (METHOD_OVERRIDES), process steps (STEP_IN_PROCESS). READ the schema first: gitnexus://repo/{name}/schema
Explain taint findings — source→sink data flows for security review. Shows intra-procedural taint (statement-level hops) and cross-function flows. Categories: command-injection, path-traversal, sql-injection, xss. Requires the repo to be indexed with --pdg.
Analyze the blast radius of changing a code symbol. Returns affected symbols grouped by depth, risk assessment, and affected execution flows. ALWAYS run this before editing shared/runtime code. Use mode='pdg' for statement-level impact with control/data dependence.
Query the Program Dependence Graph for statement-level control and data dependence. mode='controls': what condition gates a statement? What runs under true/false branches? mode='flows': where does a variable's value flow (REACHING_DEF edges)? Requires the repo to be indexed with --pdg.
Output schemas not documented. Tools define input parameters with full JSON Schema, but return types are absent. LLMs cannot plan downstream tool calls or extract required fields (e.g., does gitnexus_query return symbol locations, file paths, or both?).
No error handling guidance. Tools do not document failure modes (e.g., 'repo not found', 'symbol ambiguous', 'PDG index missing'). LLMs receive raw errors with no recovery path.
No tool annotations. All 6 tools are READ_ONLY, but none declare readOnlyHint in their schema. This prevents clients from optimizing caching or parallelization.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Search the code knowledge graph using hybrid BM25 + embedding vectors. Use this FIRST to orient yourself on any task — find relevant code, processes, or concepts. Returns ranked execution flows with symbol locations.
Pagination not addressed. gitnexus_query accepts 'limit' (default 5) and 'max_symbols' (default 10), but no cursor, offset, or total_count mechanism is documented. Large result sets may be truncated silently.
Parameter 'repo' is optional with auto-detection, but no guidance on when auto-detection fails or how to specify repo explicitly. Ambiguous for multi-repo scenarios.