skills

Transform routed work into an executable dependency graph with critical path, checkpoints, and safe concurrency. Use when a request has prerequisites, phases, or cross-artifact dependencies.

Recover from blocked, failed, low-confidence, or contradictory work without losing completed value. Use when an agent cannot access a dependency or tool.

Run independent work streams concurrently while preventing context drift, duplicated effort, and write conflicts. Use when several fragments have no blocking dependencies.

Merge specialist outputs into one coherent deliverable while preserving correctness, traceability, and user intent. Use when several agents return files, findings, or recommendations.

Assign each routed fragment to the most capable agent or skill with a bounded brief and explicit return contract. Use when a routing plan is ready for execution.

Run a final cross-format quality pass that catches mechanical, visual, textual, structural, and packaging residue left by generated work. Use when a deliverable is ready to ship.

Remove AI-generated code artifacts that obscure intent, hide failure, duplicate logic, or pretend incomplete work is complete. Use when code contains narrative comments, catch-all fallbacks, unsafe casts, hallucinated imports, duplicate helpers, stubs, or oversized generated functions.

Detect repeated AI-default visual, structural, interaction, and copy patterns before they harden into a design. Use when a mockup, screenshot, design system, or ui implementation needs an anti-slop audit.

Rewrite AI-shaped prose into precise, natural, context-aware human writing while preserving verified meaning. Use when text feels generic, inflated, promotional, repetitive, overly symmetrical, or mechanically polished.

Remove recognizable AI-default interface patterns and replace them with intentional structure, hierarchy, restraint, and product-specific character. Use when an interface resembles generic ai saas output.

Define and enforce a usable voice system grounded in audience, context, and real examples rather than generic brand adjectives. Use when a project needs consistent product, documentation, support, or marketing voice.

Produce accurate, example-rich API documentation that explains semantics, authentication, errors, lifecycle, and compatibility. Use when an http/rpc/event/module api needs consumer documentation.

Capture a consequential architecture decision, its context, options, trade-offs, and follow-up in a durable ADR. Use when a technical choice has meaningful long-term or cross-team impact.

Convert a system or feature design into an implementation-ready document with context, decisions, interfaces, risks, and validation. Use when a design needs review before implementation.

Create an elegant, task-oriented documentation set that lets contributors understand, run, use, and maintain a project. Use when a project needs readme, contributor, setup, architecture, or usage documentation.

Turn verified changes into audience-specific release notes that explain impact, action, compatibility, and risk. Use when a version, deployment, migration, or feature set is being released.

Write an executable operational runbook for recurring procedures, alerts, incidents, recovery, and verification. Use when operators need a repeatable procedure.

Audit and remediate an interface for perceivability, operability, understandability, robustness, and inclusive use while retaining deliberate design. Use when a ui is being reviewed before release.

Create or evolve a design system that encodes product decisions, accessibility, and composable behavior without turning every UI into the same generic template. Use when multiple products or teams need shared tokens/components.

Design understandable behaviors, state transitions, feedback, and recovery across a user journey, with mandatory deslop review. Use when a workflow, form, command, navigation pattern, or stateful component needs behavior design.

Design a visually intentional, responsive interface that makes hierarchy and action obvious while conforming to mandatory deslop gates. Use when a new page, application surface, dashboard, landing page, or component family needs design.

Design purposeful, interruptible, performant motion that communicates causality and spatial change without becoming decorative sludge. Use when transitions, gestures, loading, layout change, or microinteractions need motion.

Evaluate an end-to-end experience against user goals, comprehension, efficiency, accessibility, trust, and distinctiveness, then prioritize evidence-based improvements. Use when an existing product flow needs critique.

Build a least-privilege, reproducible CI/CD pipeline that validates changes and promotes immutable artifacts safely. Use when a project needs automated checks, builds, releases, or deployments.

Decide whether a change is safe to deploy by verifying artifact, configuration, migration, observability, security, capacity, and rollback evidence. Use when a release candidate is approaching deployment.

Make local, test, and shared environments reproducible, diagnosable, and safe to reset. Use when developers or agents cannot reproduce the runtime.

Coordinate incident detection, containment, diagnosis, recovery, communication, and learning with clear roles and evidence. Use when a production service or security control is degraded.

Create a coherent project skeleton with conventions, tooling, quality gates, documentation, and secure defaults matched to the chosen stack. Use when a new software project or package is starting.

Tie code, documentation, security, artifacts, approvals, deployment, and communication into one traceable release. Use when a multi-component or customer-facing release must be assembled.

Configure source control, collaboration, ownership, branch protection, templates, and repository automation for safe change. Use when a repository is new or inconsistently governed.

Turn request fragments and constraints into observable, testable completion criteria. Use when a deliverable is subjective or underspecified.

Extract, normalize, rank, and attach constraints to the exact work fragments they govern. Use when a request contains technical, legal, stylistic, budget, time, compatibility, safety, or format constraints.

Convert a compound prompt into atomic work fragments that can be assigned, verified, and recombined without losing intent. Use when a prompt contains multiple deliverables, domains, files, audiences, or phases.

Map each work fragment to the smallest sufficient set of existing skills and expose routing gaps before execution. Use when a decomposed request must be matched to skills.

Find and reconcile authoritative product, platform, API, standards, and implementation documentation. Use when a technical answer depends on current official documentation.

Crawl and understand a software repository’s structure, skills, patterns, licenses, and reusable concepts without misattributing copied work. Use when a user references one or more repositories.

Turn a question into a bounded evidence plan with subquestions, source strategy, freshness needs, and synthesis criteria. Use when a request requires external facts, current information, comparison, or deep investigation.

Verify provenance, authority, recency, scope, and claim support before evidence enters a deliverable. Use when a source is load-bearing, surprising, disputed, or current.

Collect, compare, and synthesize web evidence into a decision-ready answer with calibrated confidence. Use when a request needs market, policy, ecosystem, company, or current public information.

Assess direct and transitive dependency vulnerabilities for actual applicability, exploitability, and safe remediation. Use when a cve/advisory affects a dependency.