codegen_orchestrator

Scan codebase for dead code, code smells, security issues, contract violations, missing DTOs, and architectural deviations. Creates/updates docs/audit.md and adds actionable items to backlog. Use when user says "audit", "scan", "check code quality", or wants to find hardcoded strings, bypassed abstractions, missing schemas, untyped API boundaries, or drift from shared contracts.

Structured thinking session on a topic. Creates docs/brainstorms/<topic>.md with Status tracking and Action Items. User routes results manually (new sprint, backlog, VISION update, hotfix).

Close the current sprint phase — verify all tasks done, run/write integration tests, advance to next phase.

Close the current sprint — final gate after endgame. Push all commits, update STATUS.md history, update CHANGELOG.

Sprint dispatcher — reads STATUS.md and invokes the right skill based on current sprint state. Use when user says "go", "next", "continue", or wants to proceed with sprint work.

Implement the current sprint task using TDD. Reads task from sprint directory, creates git branch, updates task status on completion. Main development skill.

Create a new sprint — checks tech sprint cadence, proposes scope from VISION/ROADMAP/backlog, creates sprint directory and files.

Generate task files for the current sprint phase. Reads sprint.md, creates 2-4 task files with descriptions, tests, and acceptance criteria.

Update living documentation to match current codebase state. Three modes: `/update-docs` — incremental, all groups, checks CHANGELOG + git log since last run; `/update-docs <group>` — incremental but scoped to one group; `/update-docs full <group>` — deep-reads actual code and compares with docs in a group. Groups: architecture, contracts, agents, ops, workflow. Use when user says "update docs", "docs are stale", "sync documentation", or after a batch of changes lands. Also useful from /checkpoint.

Decompose a Story into Tasks. Reads story from API, analyzes scope, creates concrete tasks with descriptions and acceptance criteria.

Run Line 2 E2E test — submit engineering task, wait for completion, verify, write report. Use when user wants to test the engineering pipeline end-to-end.

Accompany a real user through the full pipeline — from project/feature request to deploy. Unlike e2e tests, this skill does NOT fail fast. The goal is to ensure the user gets their result (project or feature), no matter what breaks along the way. Produces a detailed report with all issues, warnings, and recommendations. Use when the user says "escort", "accompany", "babysit", "monitor user", "watch the pipeline", "сопровождение", or wants to observe a real user's task going through the system end-to-end and intervene if needed.

Run integration and service tests locally, fix stale/broken tests, propose new tests for uncovered logic. Use when user says "run integration tests", "test maintenance", "update tests", "check test health", "прогони интеграционные", "актуализируй тесты", or wants to ensure test suites reflect current codebase. Also use proactively from /checkpoint when integration tests haven't been run in >1 week.

Process skill feedback entries — auto-fix obvious prompt issues, bring non-obvious ones to user for discussion.