agent-skills

End-of-session command that persists context to Obsidian. Resolves project from local registry, loads any session handoff, then writes an Obsidian session note (when configured) and an Obsidian daily note. Use when finishing a work session, saving progress before clearing context, or documenting what was accomplished. Not for mid-session note-taking (Obsidian writes happen at the end only).

Manages the delivery lifecycle from roadmap and epic planning through story tracking, across 4 artifact types (Epic, Story, Bug, Task) plus a roadmap. Use when creating or editing an epic, story, task, bug, or roadmap; decomposing a roadmap into epics or an epic into stories; updating or listing delivery status; moving an artifact to another epic; or pushing artifacts to Linear or GitHub. Not for implementing a named story with an existing spec, project-wide overview, feature status within a spec, or quick fixes.

Authors and judges copy.yaml, the content payload a design consumes: write fresh brand or product copy from intent, or extract and structure existing content from a URL, brief, codebase, or screenshot — content only, composable under any design — then critique or audit it before it ships. Use when writing headlines, value propositions, landing-page or CTA copy; extracting, structuring, tightening, or revoicing content; capturing copy from a screenshot; syncing copy.yaml from a drifted implementation; critiquing copy or checking whether it reads as AI slop; scoring copy on voice and proof; or auditing copy for readability, claims, and microcopy before handoff. Not for visual identity or design tokens, page layout or screen flow, or standalone social bios.

Structured idea exploration from vague to direction, with a grill phase that stress-tests every chosen direction before capture. Two entries: greenfield maps a problem space from scratch; grill entry starts from an existing idea or plan to expose weakness, validate, refine, or surface a pivot. Diverge generates alternatives, converge evaluates trade-offs, capture produces the artifact; the deep argument widens the grill to every assumption. Use when brainstorming, exploring options, comparing approaches, rethinking a direction, considering a pivot, asking for a second opinion, finding holes in a plan, or grilling assumptions before building. Not for documenting a finalized direction.

Greenfield design pipeline for any digital product: explore a visual direction when no reference exists, then author and refine the visual identity in DESIGN.md. Use when designing a landing page, marketing site, web or mobile app, dashboard, or e-commerce store; exploring or deciding a mood or visual direction from scratch with no reference or moodboard; authoring a design system, extracting design tokens, building or refreshing DESIGN.md from references, images, briefs, or a codebase; applying named tones to the identity; redesigning, modernizing, or refreshing the brand identity; previewing and tuning the design tokens as a visual styleguide then committing tuned values back; reconciling design drift or syncing from implementation; evolving the identity against a product's stated intent from a PRD or PRODUCT.md. Not for rendering page variants, feature implementation spec, technical design docs, system architecture, or PR/code review.

Creates and manages Obsidian notes for projects, technical challenges, brag documents, meeting transcriptions, and company tracking. Use when documenting projects, recording technical challenges, maintaining a brag document of achievements, preserving meeting, 1:1, standup, or course transcriptions, or tracking a company or job application. Not for end-of-session persistence across memory systems or for initializing repository-wide project context.

Iterative debugging workflow with confidence scoring, pattern comparison, and strategic log injection. Investigate, fix, verify loop with auto-cleanup of debug logs and escalation to architectural review after 3 failed attempts. Use when diagnosing unexpected behavior, silent errors, or intermittent failures, when tests pass but the app fails or it works in dev but not in prod, when tracing an issue, or when adding or cleaning up debug logs. Not for known one-line fixes where the user names file and line, runtime error review on deployed services, or PM bug-report triage.

Creates and manages Claude Code rules in .claude/rules/ with the Incorrect/Correct template, classifies the input before writing, and decides scope, location, and frontmatter from signals in the request. Supports create, list, edit, extract from oversized CLAUDE.md, delete, and refusal with redirect when the input belongs to a skill or hook. Use this skill whenever the user defines a coding convention, team standard, or constraint Claude should enforce — even when the user does not explicitly say "rule". Also use for creating, listing, editing, or deleting a rule, scoping a rule to specific paths, splitting a growing CLAUDE.md into rule files, or auditing existing rules. Not for procedural workflows, lifecycle hooks, or one-off task instructions.

Save current conversation state to disk so a later session resumes with full context. Captures focus and next step, plus optional decisions, findings, open threads, blockers, and references; appends snapshots newest-at-top. Use when ending a session before context loss, checkpointing mid-session, saving, loading, or clearing a handoff, or starting a session that should continue prior work. Not for end-of-session persistence across memory systems or for repository-wide project context.

Explores and judges the visual direction of a UI. Use when exploring, comparing, or deciding a visual direction for a UI — generating design variants, previewing a page or screen, tuning a look — or when judging design quality: critiquing a design, checking whether it reads as AI slop, scoring usability, or auditing a built UI for accessibility, performance, responsiveness, and theming before release. Covers landing pages, marketing sites, dashboards, product UI, app screens, and components, across color, typography, layout, motion, interaction, responsive behavior, anti-patterns, and visual hierarchy. Non-mutating — produces design directions and reviews, never edits production code, tokens, layout, or copy. Not for authoring the visual identity, planning layout structure, writing copy, or source-code / PR review.

Spec-driven feature development with auto-sized depth. Produces spec.md, design.md, tasks.md, and validation.md with requirements traceability, and closes with an independent audit tied to Goals and acceptance criteria. Depth scales to scope — Small runs inline, Medium and up run the full pipeline. Use when planning or specing a feature, turning a PRD into a spec, breaking a change into tasks or user stories, designing a feature, implementing a named task or user story, auditing goals at a commit boundary or before a PR, running UAT on a user-facing change, or discussing how to build a feature. Not for diagnosing unknown bugs, authoring standalone PRD/RFC/ADR/Design Doc documents, PR/commit mechanics, or PM backlog tracking.

Git workflow helper for conventional commits, pull request creation, and branch lifecycle. Use when committing staged or unstaged changes, opening or pushing pull requests, merging branches, or cleaning up after a merge. Not for code review, acceptance-criteria verification, visual design review, or session wrap-up.

Generates structured product and technical documents through guided discovery: PRDs, product positioning docs, Design Docs, and architecture decision records. Use when defining product requirements, capturing product strategy or positioning, weighing architecture and trade-offs, or recording an architecture decision. Not for visual or UI design, feature specs tied to implementation, or meeting and session notes.

Plans the layout structure a design consumes — a region tree plus screen flow that arranges surfaces before any visual identity — then draws it as a low-fi wireframe, greyscale and content-blind, like a Figma sketch. Structure only, composable under any design. Use when planning information architecture, page composition, screen inventory, or screen flow; arranging surfaces, blocks, or navigation; wireframing, rendering, or drawing a low-fi wireframe, ASCII sketch, or screen flow; validating a wireframe or reconciling its plan; drafting a layout plan from a brief or conversation. Not for visual identity or design tokens, copy or content.

Confidence-scored code review in two modes — quick (a fast pass pairing a change walkthrough with inline findings, the default) and deep (a multi-material fan-out across the diff, guideline files, git history, and prior pull requests, with an independent confidence judge). Runs locally before a pull request. Use when reviewing code changes, checking a diff before commit or PR, running a deep or full review, auditing guideline compliance, suggesting or applying fixes, or re-reviewing fixes. Not for acceptance-criteria verification, visual design review, or commit/PR/branch mechanics.

Analyzes a skill's end-to-end reliability — trigger precision plus compound workflow risk (step count, step nature, variance hotspots) — and proposes verifiable fixes it can apply on confirmation. Use when evaluating a skill's hit rate, checking whether a description fires on the right requests, finding prompt-driven or generation steps that drag reliability, or deciding where scripts, idempotence, or explicit inputs/outputs help most. Invoke as /skill-reliability with a skill name, or when the user asks about steps vs reliability, hit rate, trigger precision, or confiabilidade of a specific skill.