Coding standards, frameworks, refactoring, and code review (18.3K skills)
Coordinate implementation and review work across OpenAI Codex and Google Gemini. It can route tasks between models, arrange independent review, and define escalation paths that may include Claude and a human reviewer.
Review a feature or architecture proposal through six specialist perspectives: product, architecture, design, security, UX, and CTO. The process requires unanimous approval, consolidates objections, and allows up to three review rounds before escalating to a person.
Break a technical specification into small stories that one coder can complete in a batch. Map the existing modules and interfaces involved, order stories by dependency, estimate effort, and prepare a prioritized queue for implementation, including parallel work where appropriate.
Review a software project for maintenance needs by checking branches, documentation, TODO markers, test coverage, and dependencies. Produce a project health score and remediation recommendations to help teams track and prioritize technical debt.
Maintain a DOT knowledge graph of architectural patterns and design decisions. Capture new knowledge, check graph consistency, find relevant patterns in the codebase, sync the graph with current code, and export it for visualization.
Review code changes against four principles: avoid duplication, skip speculative features, use appropriate abstractions, and provide adequate test coverage. Get numeric quality scores and specific feedback tied to file names and line numbers.
Check GSD plans and phase artifacts for structural errors, missing files, broken references, and unmapped requirements. It returns a structured report showing passed checks, warnings, failures, and recommended fixes.
Manage Git work for GSD plans with one structured commit per task, separate commits for planning files, branch status checks, history and diff summaries, and annotated milestone tags. Keep code and planning changes organized with consistent references to phases and tasks.
Run language-specific linting, formatting, and type checks for Python, TypeScript or JavaScript, and Go. The workflow applies available fixes, repeats checks up to three times, and reports issues that still need attention.
Review code changes with a depth matched to their size, from detailed checks to architectural review. It organizes findings with file locations, reasoning, positive feedback, and a clear verdict that leaves the final decision to the user.
Preserve an agent’s work before its context window fills by saving task progress, quality check results, and TDD status. Restore that state when a session resumes so development can continue from the last recorded point.
Implement only the code needed to meet the stated requirements, following the project’s existing structure, naming, imports, and error-handling patterns. It supports focused feature work and fixes without adding unrequested features or premature optimization.
Track beads as issues stored with Git, following their assignment, progress, review, and completion. Record which agent handled each bead along with time and quality data, and monitor dependencies and convoy progress within Gas Town workflows.
Coordinates agent sessions from initialization through handoffs and resumption. It can restore a dead agent's session state and manage persistent identity for Polecat and Crew agents, with commands for attaching to coordination and passing context.
Monitor active agents and long-running workflows through health checks and heartbeats. Detect unresponsive work, choose a recovery action such as restarting or reassigning an agent, and report findings with trend analysis.
Coordinate pending changes from multiple agents by checking for branch conflicts, resolving them with a chosen strategy, and merging changes in dependency order. Then verify the integrated work with tests, lint, and a build before it lands.
Define multi-step workflows and variables in TOML, validate their structure, and turn templates into durable Gas Town workflows that can survive restarts. Try an instance before registering the reusable formula in the library.
Coordinates long-lived Crew agents and transient Polecat workers by assigning work through hooks, handing tasks between agents, and nudging stalled sessions. It also explains a three-level hook hierarchy and provides commands for listing, feeding, and reviving agents.
Research a development problem before implementation by checking existing projects, packages, documentation, and design patterns. Compare at least three approaches across effort, time, risk, and scalability, then provide a ranked recommendation with rationale and go/no-go criteria.
Carry a specification through implementation, testing, review, and up to three improvement cycles. The skill checks quality against a threshold, organizes changes into atomic commits, and reports requirement completion, test coverage, quality, and remaining work.
Research a codebase's patterns, dependencies, tests, and constraints to create a feature specification. The document can include scoped requirements, testable acceptance criteria, architecture decisions, phased implementation steps, API and data models, test strategy, and risk mitigations.
Map a codebase by summarizing its folders, languages, frameworks, entry points, dependencies, and test layout. The resulting overview gives agents project context for research, specifications, and code generation, including awareness of conventions and likely imports.
Coordinate a parallel code review across architecture, security, performance, testing, code quality, and documentation. The review combines weighted scores and critical findings into an approval, change request, or rejection recommendation.
Create Git checkpoints before risky work or after quality checks, recording changed files and line counts. List checkpoints, preview the changes a restore would undo, and return the working state to a selected checkpoint.