hsl

Organize agent-facing documentation by purpose for progressive discovery. Use to initialize greenfield documentation, migrate an existing documentation tree, install or update this project-local skill, or consult and maintain its current/design/reference conventions. Audit documentation only on an explicit user request, presenting evidence before edits and refining incrementally through user feedback.

Use when the user asks to do coding work in a Git repository. Enable directly when the user invokes or explicitly requests this skill; otherwise ask first.

Audit code for maintainability, excessive complexity or nesting, unclear integration boundaries, exception-handling mistakes, and duplication caused by poor structure. Use when the task calls for a code-quality audit or a critical review of produced code.

Use when implementing, reviewing, or refactoring code whose readability is a first-class concern: naming, abstraction boundaries, indirection, module or file structure, and choices between similarly correct designs. Trigger when the user asks for clear, narrative, concise-but-readable, low-cognitive-load code, or wants code that is easy to evaluate and give feedback on. Optimize for the reader's limited cognitive budget while preserving correctness and local conventions.

Create or refine frontend interfaces with cohesive reusable components, clear information architecture, restrained copy, and verified interaction. Use for new screens, application shells, UI previews, and substantial frontend changes.

Decruft code by finding implementation remnants, convention drift, needless abstractions, stale data or dependencies, and legacy runtime paths. Use when the user asks to decruft, remove stale or legacy code, investigate suspiciously useless code, or reconcile competing old and new patterns.

Run the feature workflow. Plan, implement, and review a code change.

Use when implementation is likely to reveal consequential choices that could not be settled from requirements alone, especially at unfamiliar integration seams, in projects without a governing convention, or whenever a solution starts to feel patched together or hacky. Ask a technical user for focused, concrete feedback when it is cheaper than silently choosing and later reworking the result. Use with cognitive-budget-coding so the implementation and the choices presented remain easy to evaluate.

Use when the user wants a task to unfold gradually and interactively, with meaningful visibility into details, findings, or choices while they can still influence what happens next. Applies across explanation, research, maintenance, analysis, planning, writing, implementation, and other multi-step work. Complete one useful bounded unit at a time, invite feedback suited to the situation, and adapt the next unit to the user's response instead of withholding the work until one large final result.

Turn a prior discussion, source study, or agreed design into an implementation plan the user reviews and approves before any implementation begins.

Create a skill from a user brief or systematize a process from the current conversation into a reusable skill. Use only when the user requests skill creation or invokes issueskill.

Load this skill before writing any message that presents a plan, proposal, findings, report, or a set of decisions for the user to review, approve, or answer. Shapes the message: orient first, main path first, representative abridged excerpts labeled proposed vs current, visible connections, a map of affected files, and an explicit approval question. Presentation only; it does not decide what to do.

Use when the user wants a plan and approval checkpoint before implementation.

workflowz. Survey a repository with multiple Git worktrees after time away, reporting each worktree's divergence, local changes, purpose, shared topic ancestry, and clean duplicates before deciding whether to remove anything.