Reviewer Profiles

Profiles are specialist reviewers — each with its own focus areas, system prompt, and severity guidelines. Run one or several in parallel to get targeted feedback from different perspectives.

By default nitpik runs --profile auto, which picks the right specialists for each diff (see Auto-Selection below). Pass --profile <name> to override.

Focuses on server-side code quality: correctness, error handling, performance, concurrency, API design, and data integrity. Catches bugs like logic errors, N+1 queries, missing error propagation, and race conditions.

Skips pure formatting issues and deep security analysis — it flags obvious problems (like unsanitized SQL) but leaves thorough security review to the security profile.

nitpik review --diff-base main --profile backend

`frontend`

Focuses on user-facing code: accessibility, rendering performance, state management, UX, and responsive design. Catches missing ARIA labels, improper heading hierarchy, unnecessary re-renders, and memory leaks from unsubscribed listeners.

Skips backend logic and deep security analysis.

nitpik review --diff-base main --profile frontend

`architect`

Focuses on system design: coupling, abstractions, module boundaries, API surface changes, and backward compatibility. Catches god objects, leaky abstractions, breaking changes, and dependency direction violations.

Skips localized implementation bugs — if a change is purely internal to one function, the architect won't nitpick it unless it reveals a systemic pattern.

nitpik review --diff-base main --profile architect

`security`

Focuses on vulnerabilities: injection risks (SQL, XSS, command injection), authentication and authorization flaws, cryptographic misuse, data exposure, and insecure configuration. Traces data flow from untrusted input to sensitive sinks.

Reports findings only when the vulnerability path can be verified — no speculative alerts.

nitpik review --diff-base main --profile security

`general`

A broad, language-agnostic catch-all reviewer. Focuses on universals: correctness, clarity, error handling, configuration mistakes, broken references, and obvious anti-patterns across any file type — documentation, shell scripts, Markdown, configuration formats, and languages without a dedicated specialist.

Used by auto mode as the catch-all when no language specialist (backend or frontend) is selected for the diff. Defers deep domain analysis to specialist reviewers when they run alongside it.

nitpik review --diff-base main --profile general

Combining Profiles

Run multiple profiles in parallel:

nitpik review --diff-base main --profile backend,security

When multiple profiles run together, each one is informed about the others and their focus areas. This coordination prevents duplicate findings and keeps each reviewer in its lane. See How Reviews Work for details.

Auto-Selection

When --profile is not specified, nitpik runs auto. You can also request it explicitly:

nitpik review --diff-base main --profile auto

Auto-selection examines three layers of signals to choose profiles:

File extensions and paths — unambiguous extensions (.vue, .css → frontend; .rs, .go, .py → backend) are classified directly. JS/TS files are disambiguated using directory names (e.g. controllers/ → backend, components/ → frontend) and filename patterns (e.g. *.controller.ts → backend).
Project root markers — when JS/TS path signals are absent or one-sided, nitpik checks the repo root for package.json dependencies (Express, React, etc.) and config files (nest-cli.json, wrangler.toml, etc.) to fill in the gaps.
Architect triggers — the architect profile is added when the diff touches cross-cutting files (CI configs, Dockerfiles, IaC, dependency manifests, API definitions, database migrations) or when the diff is large (many files or many distinct directories).

The security profile is always included because its frontmatter sets always_include: true. Any custom profile in your --profile-dir with the same flag is appended too — see Always-On Profiles.

When neither frontend nor backend signals fire (for example a docs-only, infra-only, or shell-only diff), the general profile is used as the catch-all. general and the language specialists are mutually exclusive: a strictly-backend or strictly-frontend diff never pulls general in alongside the specialist. JS/TS files with no clear signal still default to frontend rather than general.

Auto-Mode Strategy

Tune how --profile auto decides with --auto-mode:

Mode	Behavior
`heuristic`	File/path/dependency rules only — no LLM call. Fastest, fully offline.
`llm`	Always ask the model to pick profiles using a built-in `triage` system prompt. Most flexible for unusual diffs.
`hybrid` (default)	Heuristics first; consult the LLM only when the heuristic confidence is low (e.g. it would otherwise pick only `general`).

Hybrid mode is the recommended default — you get heuristic precision on common cases without losing flexibility on edge cases. Falls open: if the triage call fails, the heuristic result is used.

Tag-Based Selection

Select profiles by tag instead of name:

nitpik review --diff-base main --tag security
nitpik review --diff-base main --tag css,accessibility

All profiles (built-in and custom) whose tags contain any of the given values are included. Tag matching is case-insensitive.

Combine --tag with --profile to add tag-matched profiles on top of explicit ones:

nitpik review --diff-base main --profile backend --tag css

Built-in Profile Tags

Profile	Tags
`backend`	`backend`, `api`, `database`, `logic`, `performance`
`frontend`	`frontend`, `ui`, `ux`, `accessibility`, `css`, `javascript`, `typescript`
`architect`	`architecture`, `design`, `patterns`, `maintainability`, `coupling`
`security`	`security`, `auth`, `injection`, `xss`, `csrf`, `cryptography`
`general`	`docs`, `config`, `shell`, `scripts`, `prose`, `cross-cutting`

Listing Profiles

See all available profiles, including custom ones:

nitpik profiles
nitpik profiles --profile-dir ./agents

This shows each profile's name, description, and tags.

Custom Profiles — create your own reviewers
How Reviews Work — multi-agent coordination
Agentic Mode — give profiles access to tools
CLI Reference — all profile-related flags

Edit this page on GitHub →