ollama_code_review

Code review. REVIEW. Given a unified diff (and optional source_paths for context), returns STRUCTURED FINDINGS: {findings:[{severity, category, file, line?, symbol?, description, recommendation}], summary, diff_size_bytes}. Severity enum critical|high|medium|low; category enum bug|security|performance|style|maintainability. Distinct from ollama_multi_file_refactor_propose (proposes refactors) — code_review flags issues to fix on the diff as-is. Optional severity_floor filters out below-floor findings; max_findings caps result. Diff capped at 2MB; source_paths max 50. Tier defaults to workhorse; pass tier:'deep' for high-stakes review. coerceReview drops malformed entries instead of throwing.

Parameters

Parameter	Type	Required	Default	Description
diff_text	string	yes	—	Unified diff text (e.g. git diff output). Required. Max 2 MB to keep prompts within sensible context budgets — for larger reviews chunk the diff per logical change. Field name matches changePack / changeBrief convention.
source_paths	string[]	no	—	Optional full source files for the changed locations. Provides the model with full context for cross-line/cross-symbol concerns (e.g. ‘this rename breaks line 200 in the same file the diff only touches line 50’). Max 50 paths.
severity_floor	"critical" \	"high" \	"medium" \	"low"
max_findings	integer	no	50	Cap on returned findings after filtering. Default 50, max 200. The cap protects against a chatty model on a small diff burying real signal.
tier	"instant" \	"workhorse" \	"deep"	no

Full input schema

The JSON Schema below is generated from the same zod schema the server validates against at the wire — it cannot drift from runtime behavior.

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "diff_text": {
      "type": "string",
      "minLength": 1,
      "maxLength": 2097152,
      "description": "Unified diff text (e.g. `git diff` output). Required. Max 2 MB to keep prompts within sensible context budgets — for larger reviews chunk the diff per logical change. Field name matches changePack / changeBrief convention."
    },
    "source_paths": {
      "description": "Optional full source files for the changed locations. Provides the model with full context for cross-line/cross-symbol concerns (e.g. 'this rename breaks line 200 in the same file the diff only touches line 50'). Max 50 paths.",
      "maxItems": 50,
      "type": "array",
      "items": {
        "type": "string",
        "minLength": 1
      }
    },
    "severity_floor": {
      "description": "Drop findings below this severity from the result. Default 'low' (returns everything).",
      "default": "low",
      "type": "string",
      "enum": [
        "critical",
        "high",
        "medium",
        "low"
      ]
    },
    "max_findings": {
      "description": "Cap on returned findings after filtering. Default 50, max 200. The cap protects against a chatty model on a small diff burying real signal.",
      "default": 50,
      "type": "integer",
      "minimum": 1,
      "maximum": 200
    },
    "tier": {
      "description": "Which tier to run on. 'workhorse' is the default; 'deep' is recommended for security-critical or high-stakes reviews; 'instant' for fast smoke passes on tiny diffs.",
      "default": "workhorse",
      "type": "string",
      "enum": [
        "instant",
        "workhorse",
        "deep"
      ]
    }
  },
  "required": [
    "diff_text"
  ]
}

Envelope

Every tool returns the standard envelope — tier_used, model, tokens_in/tokens_out, elapsed_ms, backend provenance, warnings. See Envelope & tiers.

Source

Schema + handler: src/tools/codeReview.ts · registration: src/index.ts

---

Back to the Tool Reference.