Claude Skills by ashugaev

Clean-room test of the Spur server install by planting two coding agents — cursor-agent and Claude Code — on a persistent cloud VM that gets reset to a pre-install state each run; the harness never creates or deletes a VM. Cron-safe — runs end to end from the skill name alone, opens a PR for any doc fix, reports in-session. Reset the box over plain SSH, plant both agents by copying the operator's credentials (itest-only shortcut), run each on the README Install block verbatim, capture the tra...

Make code, configs, docs, and prompts leaner in this repo. Use when the task is to simplify, reduce abstraction or duplication, delete dead paths, or pressure-test whether a change can be smaller.

Audits everything that enters an agent's context — skills, agent definitions, CLAUDE.md/AGENTS.md, rules files, MCP servers, tool surfaces — and reports what to cut, merge, or migrate off MCP. Measures the context budget and judges how files relate across repo and user-global scope. Use when reviewing a diff that touches agent-facing text or MCP config, when context feels bloated, when a stated rule is being ignored, or when asked to census or shrink agent configuration. Don't use for prose q...

Author and export a UI design via Claude Design (claude.ai/design) for the pre-implementation design gate. Load when producing/exporting a design or defining the export contract. Covers DesignSync usage, the export bundle, and the approval protocol.

Governs published docs (README.md, docs/, SETUP/TROUBLESHOOTING/CONTRIBUTING/SECURITY), including the reference-doc format for docs/commands.md, docs/configuration.md, docs/daemon-api.md. Load on any published-doc edit — create, change, move, trim, or link fix. Enforces open-source standards: Diataxis mode, granular single-topic files, caveman prose, one source per topic, link don't restate. Which facts land in the reference docs (config-doc rule) stays with AGENTS.md/CLAUDE.md and the spur s...

Visual codestyle and design system rules for packages/web. Use when reviewing or implementing frontend changes.

Work with GitHub via gh CLI — create PRs, manage issues, review diffs, check CI status.

Orchestrate every repo task by routing each todo to agents and skills based on its properties. Decompose, delegate, aggregate, close out. Mandatory for every task in this repo.

Fetch PR review comments, evaluate each critically, fix valid ones, reject wrong suggestions. Use when asked to check PR comments, fix review feedback, or handle PR review.

Validate manager close-out for repo work. Use when implementation is done and the final pass must confirm PR, validation, and required manager gates. Don't use for planning or implementation.

Route a task to a deliberation tier by ambiguity × blast radius. No tools, pure reasoning, < 5 seconds.

Write and review agent content — skills, agent definitions, prompts, workflow docs. Optimized for token economy, precision, and zero filler. Use when creating or editing SKILL.md, agent .md files, or orchestrator instructions.

Roll a Spur host onto a published npm version by hand and verify it, when the automatic path fails. Use when `spur update` rolled back, the UI version switch reports failed, the dashboard shows a stale or 0.0.0 version, or the daemon stays unreachable after a version change. Don't use for cutting a release, for publishing to npm, or for a Spur running from a repo checkout.

Spur orchestrates AI coding agents (claude/codex/cursor/opencode) via CLI, daemon, web UI, or Telegram. Use when spawning, messaging, automating, or configuring a session, tmux pane, or Spur config. Don't use for driving an agent's CLI, or plain git worktree work.

Live end-to-end test of the Telegram source on a dedicated test bot through an isolated daemon. Use after any change to the Telegram source, agent sends, buttons, reply routing, placeholders, notices, or the Telegram prompt suffix, before close-out.

Send messages and read updates from Telegram via Bot API.

Send messages and read updates from Telegram via Bot API.