kitt

Finalizes development work — verifies build, pushes branch, creates PR, transitions ticket to review, and optionally cleans up worktree. Called by orchestrate when the user signals work is done.

Implements tasks from plan.md with TDD, task manager integration, and PR creation. Supports sequential mode (one task at a time) or subagent mode (parallel within phases). Reads commit format and platform config from kitt.json.

Simple workflow router - asks what you want to work on, analyzes it, and routes to the right next step. Handles epic → US workflow and flat feature workflow. Auto-syncs metadata on US completion.

Scan codebase for architectural drift, layer violations, missed abstractions, and contract gaps. Produces structured proposals with context, problem, and expected fix. Integrates with task manager and implement pipeline.

Validates refined feature against project architecture. Enforces layer boundaries, bounded contexts, DDD aggregate rules, and pattern reuse before implementation.

Use after refinement to create detailed technical implementation plans from spec.md - breaks down user stories into tasks, dependencies, technical decisions, and optional task manager tickets

Automated code review skill — runs between implement and finish-development. Reviews diff against spec, plan, code-standards, and agent docs. Produces structured review with actionable findings.

Feature refinement process. Validates that a feature is ready for architecture validation and implementation. Supports EPIC mode (creates spec with US breakdown) and US/FEATURE mode (creates individual spec). Does not mutate task manager without explicit confirmation.

Use when implementing any feature or bugfix. Write the test first, watch it fail, write minimal code to pass. No exceptions.

Use before claiming any work is complete, fixed, or passing. Run the verification command, read the output, then make the claim. Evidence before assertions.

Turn a raw idea into a validated design through collaborative dialogue. Use BEFORE orchestrate when you don't yet know what to build. Outputs a design.md in workspace/ then hands off to orchestrate.

Captures a coding correction or design fix as a permanent rule in the right context file. Invokable manually (/capture-rule) or auto-invoked from implement after a mid-task correction. Four user-facing destinations: feature-level spec, app-scope agent doc, repo-wide code-standards.md, or domain-wide product.md. Plus a programmatic skill-diff mode used exclusively by /revise for post-completion kitt-skill updates.

Onboards a new developer — reads project context, scans codebase, asks about their role and focus area, and generates a personalized onboarding guide.

Backend QA — loads spec context, gathers API credentials and payloads, runs HTTP scenarios, writes a local qa-run file updated live, then publishes to the configured report adapter.

Frontend QA — loads spec context, gathers test data, runs browser scenarios, writes a local qa-run file updated live, then publishes to the configured report adapter (Notion, local, etc.).

Interactive kitt configuration wizard. Scans the project repo, asks targeted questions, writes kitt.json, generates context file drafts (product.md + code-standards.md), and validates completeness.

Post-completion review of a work item's development session — analyzes session log events, skill usage patterns, feedback frequency, spec drift, and time distribution. Outputs a review.md with metrics, actionable improvements classified into 5 typed categories (kitt-skill, context-stale, spec-quality, skill-gap, process-waste), and skill effectiveness evaluation. Internally delegates narrative extraction to session-summarize and reads the cached {key}-summary.md alongside the raw session log.

Produces a structured narrative summary of a work item's development session from its session-log.jsonl. Callable standalone for quick personal notes, or invoked upstream by session-review and session-aggregate. Output is a cached {key}-summary.md file structured for both human reading and LLM consumption by downstream pattern-mining skills.

Handles post-completion revisions of a feature — QA defects, staging bugs, post-merge incidents, reviewer rejections, customer reports. Classifies the root cause into one of 8 categories, proposes in-place artifact updates and systemic lessons via capture-rule, tracks everything in workspace/{key}/revisions/{timestamp}/. Invoked only by /orchestrate option D. Never a direct CLI entry point.

Systematic bug investigation — reproduce, isolate, hypothesize, confirm root cause, fix, verify, regress. Works with any configured adapter (Jira, Linear, local).

Standalone task manager operations — create, read, update, transition, comment on tickets. Works with any configured adapter (Jira, Linear, GitHub Issues, Local).

Creates a git branch from a task manager ticket key using the configured VCS adapter. No interactive bash.

Creates a PR/MR via the configured VCS adapter, links it to the task manager ticket, and transitions ticket status.

DEPRECATED — moved to branch-creator. This file is not discoverable by the Skill tool.

DEPRECATED — moved to pr-creator. This file is not discoverable by the Skill tool.

Cleans up a git worktree after development is complete. Removes the worktree folder, unregisters it from git, and optionally deletes the local branch. Called by finish-development.

Creates an isolated git worktree for a branch outside the repository. Reads path and setup commands from kitt.json. Called by orchestrate when user wants isolation.