Claude Skills by edmundmiller

Diagnoses performative process that displaces delivery. Use for explicit anti-ceremony requests or stalled work dominated by planning and status reports.

Design and implement text-forward technical diagrams embedded in MDX using semantic HTML, CSS grids, ASCII-inspired typography, and SVG paths or motion. Use for article diagrams, process loops, comparison tables, formula illustrations, or terminal-style explanatory graphics; not for ordinary data charts, Obsidian knowledge graphs, or Mermaid-first architecture documentation.

A subtle creativity mode for escaping obvious solutions without losing coherence. Use when the work feels technically correct but predictable, flat, or overly constrained.

Assigns overlapping tools or runtimes to the workloads they fit. Use when evaluating replacement, consolidation, or staged adoption across systems with different isolation, durability, or operational requirements.

Prepares intended GitHub pull requests from today's Codex tasks for a concise human review handoff. Use when asked to find today's PRs, make PRs or a PR stack review-ready, audit current-head checks and reviews, improve PR titles/bodies, or draft or send a Slack review handoff.

Use when writing standalone Python scripts (one-off scripts, utilities, not part of a larger Python project with pyproject.toml or setup.py). Provides UV shebang template and best practices.

Review changed code for reuse, quality, and efficiency, then fix worthwhile issues.

Validates skill structure and invocation boundaries. Use when auditing a skill, debugging its trigger, or checking portability across agent runtimes.

Preferences and procedures learned from past local agent sessions.

Uses Spock metadata to connect executable specs to design decisions. Use when authoring or refactoring Spock specifications as design documentation.

Maintains dependent GitHub PR stacks. Use for restacking, repairing PR bases, or coordinating stack merges and merge queues.

Removes an offending or unauthorized delta while preserving valid and pre-existing work. Use when asked to undo an accidental implementation, revert only part of a change, or retain desired local work while reversing a commit.

Write, change, review, or remove automated tests with independent expectations and observable public behavior; reject tautological, vacuous, implementation-coupled, and mock-only coverage.

Review charts with a simple Tufte-style rubric. Use when judging whether a chart is clean, honest, readable, and insight-forward, or when iterating on a visualization until it passes a concise quality bar.

Applies Tufte principles to chart design and critique. Use for graphical integrity, chartjunk reduction, or high-density comparison layouts.

Applies an opinionated React/Tailwind UI checklist. Use when explicitly requested or when adopting that stack's component and interaction conventions.

Simplifies bloated products, offerings, documents, plans, and flows through purpose-locked subtraction. Use when asked to cut scope, remove bloat, create a kill list, decide what to keep or remove, or improve something by subtraction. Do not use for code-only refactors, optimizing individual flow steps, release prioritization, or evaluating one proposed addition.

Entity-level merge with weave. Use when setting up weave, previewing merges, reducing false Git conflicts, or validating merge-driver behavior.

Write, design, or review agent-friendly CLIs and agentic interfaces for humans, scripts, and coding agents, with progressive disclosure, structured output, actionable errors, stdout/stderr separation, and explicit non-interactive modes.

Split staged or uncommitted changes into conventional commits. Use when asked to commit, generate commit messages, split commits, squash commits, or reorder history.

Low-interruption coordination for realtime voice sessions. Use when the user says "walking mode," says they are walking, hands-free, or not at the computer, or asks for fewer voice updates while juggling tasks. Do not use for ordinary requests to write concisely.

Control live Herdr workspaces, tabs, panes, agents, worktrees, plugins, and waits. Use when running inside Herdr (`HERDR_ENV=1`) to inspect sibling agents, delegate work—including dedicated Pi workspaces—run services, debug agent detection, or coordinate terminal state.

Review and steer live Hunk diff sessions, inspect focused files/hunks, navigate, reload sources, resume worktree reviews, and add validated inline comments. Use when the user has Hunk open, asks for an interactive diff review, mentions Hunk comments, or wants to resume a paired agent/Hunk worktree.