Skip to content

Core Concepts8 min read

Specialist Modes

21 ready-made personas for focused jobs

Beyond the ten core modes, Nexgile Code ships 21 specialist modes — focused personas pre-tuned for one job each, from generating tests to resolving merge conflicts to designing database schemas. They require zero setup: every specialist appears in the mode selector out of the box, right below the core modes, and you switch to one exactly the same way (click the mode chip, search, select).

Each specialist carries its own detailed working rules — the Test Gen mode knows coverage targets and mocking strategy per language, the Security mode knows OWASP categories and severity classification — so you get expert-shaped behavior without writing a custom mode yourself.

Slug Mode What it’s for
refactor 🔧 Refactor Modernize code within the same language — patterns, naming, dead code — without changing behavior.
testgen 🧪 Test Gen Generate unit and integration tests with the project’s own framework; targets >80% coverage, then runs and fixes them.
docgen 📝 Doc Gen Generate project documentation — README, ADRs, inline comments, API docs.
security 🛡 Security Scan for vulnerabilities (injection, XSS, secrets, weak crypto, CVEs), classify severity, and apply fixes.
apimod 🌐 API Mod Modernize APIs — SOAP to REST, OpenAPI spec generation.
depupgrade ⬆ Dep Upgrade Upgrade dependencies safely — fix CVEs, resolve breaking changes.
translate 🌐 Translate Translate code between 25+ programming languages.
issue-fixer 🔧 Issue Fixer Take a GitHub issue (bug or feature request) and implement the fix.
pr-fixer 🛠 PR Fixer Fix pull requests — address review feedback, failing tests, merge conflicts.
merge-resolver 🔀 Merge Resolver Resolve merge conflicts intelligently using git history for intent.
docs-extractor 📚 Docs Extractor Extract feature details or verify documentation accuracy against the actual implementation.
issue-investigator 🕵 Issue Investigator Investigate a GitHub issue to find root cause and propose a solution.
issue-writer 📝 Issue Writer Turn a rough bug or enhancement description into a well-structured GitHub issue.
db-migrator 💾 DB Migrator Database schema design, ORM mappings, migrating legacy data-access code to modern frameworks.
db-architect 🗄 DB Architect Database architecture planning, ERD diagrams, schema analysis.
code-architect 📐 Code Architect Analyze a codebase and generate architecture, sequence, component, class, ERD, and deployment diagrams.
devops ⚙ DevOps Generate CI/CD pipelines, Dockerfiles, Kubernetes manifests, infrastructure-as-code.
api-tester 🧪 API Tester Generate and run REST API tests with endpoint discovery and environment support.
code-reviewer 🔍 Code Reviewer Review code changes for quality, security, and best practices — structured findings with a quality score.
ui-designer 🎨 UI Designer Draft UI screens and components as throwaway React + Tailwind previews before implementing them for real.
dev 💻 Dev General full-stack development across any language or framework.

A mode’s tool groups are a hard boundary, not a style guide — a specialist without the edit group cannot write to your repository even if you ask it to. Before you hand one a production checkout, this is the table to read:

Mode read edit command mcp
🔧 Refactor, 🧪 Test Gen, 🛡 Security, 🌐 API Mod, ⬆ Dep Upgrade, 🌐 Translate
💾 DB Migrator, ⚙ DevOps, 🧪 API Tester, 💻 Dev, 🔧 Issue Fixer
📝 Doc Gen
🛠 PR Fixer, 🔀 Merge Resolver
📚 Docs Extractor .nexgile/extraction/ only
🗄 DB Architect, 📐 Code Architect agent_generated/ only
🕵 Issue Investigator, 📝 Issue Writer
🔍 Code Reviewer, 🎨 UI Designer

Every specialist also has the always-available tools, so any of them can ask you a question, keep a todo list, or delegate to another mode.

Seven of the specialists — Translate, Refactor, Test Gen, Doc Gen, Security, API Mod, and Dep Upgrade — form a family aimed at codebase modernization: moving legacy code forward language by language, API by API, dependency by dependency. They share a common shape (analyze what exists, transform it, verify nothing broke) and are covered in depth, with worked examples, in The Transformation Modes.

A few specialists behave differently from a plain read-edit-run mode:

Security produces a machine-readable report

Section titled “Security produces a machine-readable report”

A Security scan ends with two outputs. The first is a Markdown summary you read in chat: a severity table with Critical / High / Medium / Low counts and the files and lines scanned, then the Critical and High findings listed by id, file, and line. The second is a json:security-report fenced block carrying every finding in full — id, severity, category, CWE and OWASP references, file and line, the vulnerable snippet, and the remediation with its corrected snippet.

The JSON block renders as an ordinary code block, so its value is what you do with it next: paste it into an issue tracker, diff two scans of the same repo to see what a sprint fixed, or feed it to your own reporting script. Ask for Report only; don't fix yet when you want the inventory before any code changes.

Code Reviewer and UI Designer have no edit or command tools at all. Code Reviewer can only read and report — which is exactly what makes its review independent. UI Designer is explicitly not a coder: it answers a screen description with a self-contained React + Tailwind component in a code block (a single App function, inline sample data, no imports), designed to be pasted into a sandbox and looked at rather than committed. Iterate on the design in chat, then take the approved result to Code mode to implement for real.

Three specialists write only to output directories

Section titled “Three specialists write only to output directories”
  • Docs Extractor writes only under .nexgile/extraction/ (YAML, JSON, and Markdown output files).
  • DB Architect and Code Architect write only under agent_generated/.

These restrictions are enforced by the mode’s file regex, not convention: an attempt to edit outside the sandbox fails as a tool error. The payoff is that you can run an architecture analysis on a production repo knowing the source tree cannot change. Both architect specialists express their designs as Mermaid diagrams you can commit and render.

Issue Investigator, Issue Writer, Issue Fixer, PR Fixer, and Merge Resolver are built around issue-and-PR workflows, and their working instructions drive GitHub through the gh command-line tool rather than a web API of their own. Each one shells out to commands like these:

Mode Commands it relies on
🕵 Issue Investigator gh issue view <url> --json title,body,labels,comments
📝 Issue Writer gh issue creation
🔧 Issue Fixer gh to retrieve the issue and open the resulting PR
🛠 PR Fixer gh pr view --json …, gh pr checks, gh pr view --json mergeable,mergeStateStatus
🔀 Merge Resolver gh pr view <n> --json title,body,headRefName,baseRefName, plus git blame and log history

This is a prerequisite, not a nicety. Install gh and authenticate it (gh auth login) before using these modes, or they fail at their first step with a command-not-found or an authentication error — which reads like a mode problem and isn’t one. Verify with gh auth status.

Four of the five (all but Issue Fixer) also carry the mcp group, so they can use MCP tools where a GitHub MCP server is configured. That is an addition to the gh path, not a replacement for it — the workflows are written around the CLI. Issue Fixer has read, edit, and command tools only, so for that mode gh is the only route.

Because these modes work through the command group, gh invocations appear as ordinary command approvals. If you auto-approve commands, note that gh pr create and gh issue create are outward-facing actions — they publish to your repository. Many teams put them on the deny list and run them by hand.

Specialists compose naturally across tasks, because each one leaves an artifact the next can consume:

  • Audit → track → fix: run 🛡 Security for the findings report, hand it to 📝 Issue Writer to file well-structured issues, then work through them with 🔧 Issue Fixer.
  • Safety net → modernize: run 🧪 Test Gen to build coverage before letting 🔧 Refactor or 🌐 Translate loose — the generated tests become the behavioral contract the transformation must preserve.
  • Design → implement: let 🎨 UI Designer or 📐 Code Architect produce the previews and diagrams, get your team’s sign-off, then switch to 💻 Code (or 💻 Dev) to build against them.

For multi-specialist work you’d rather not conduct by hand, 🪃 Orchestrator can delegate each leg as a subtask in the right mode.

Use a specialist when its job description matches your task exactly — the pre-loaded rules (coverage targets, severity taxonomies, framework detection) genuinely change the output quality. Use plain 💻 Code when the work is mixed or doesn’t fit one box; a specialist prompt-tuned for test generation will keep steering toward tests even when you need something else. A good rule of thumb: if you can name the deliverable with one of the mode names above, switch; otherwise stay in Code.

Want a persona these 21 don’t cover — or a stricter variant of one of them? Define your own in minutes with a custom mode.