gupsammy/claudest
30 skills30 installs7.6k starsGitHub
Install
npx skills add https://github.com/gupsammy/claudestSkills in this repo
1Brainstormbrainstorm is a journey-wide thinking-partner skill for solo builders who need to clarify, stress-test, and document ideas before coding or launching. Trigger phrases like interview me about, help me clarify, stress-test my idea, or grill me on route the agent into a deliberate interview loop using Read, Write, AskUserQuestion, and light repo context via Bash, Glob, and Grep. Unlike one-shot brainstorming in chat, the skill emphasizes adaptive questioning styles, automatic saturation when themes repeat, and a tangible deliverable at the end. It is listed under Idea and research as the first shelf moment—when direction is uncertain—but the same ritual applies when validating scope, planning a ship checklist, or reframing a growth bet. Optional --grill adds adversarial dependency-tree depth for strategy-heavy topics.1installs2Claw AdvisorClaw Advisor is a journey-wide agent skill that acts as a topic router for Claw-family documentation: it maps problem domains to slug prefixes and suggested entry pages so your coding agent loads the right manual sections instead of improvising setup steps. Solo builders running personal agents on Telegram, Discord, Slack, or WhatsApp hit overlapping areas—gateway security, sandboxing, cron hooks, provider keys—and broad questions span multiple trees. Invoke it when the question domain is fuzzy or when a single user message touches channels plus gateway plus automation. The skill is procedural research metadata, not an MCP server or installer; it pairs with exec, browser, and skills tools once you know which doc cluster to read. It supports build-time integration, ship security hardening, launch distribution across channels, and operate troubleshooting when sessions or gateways misbehave. Optimized for agents that can fetch docs by slug after routing.1installs3Clean BranchesClean-branches is a git hygiene agent skill built around find-candidates.sh, which scans your repository for branches that are already merged into main (or master) and branches that look stale, without immediately deleting anything. Solo builders accumulate feature branches quickly; this skill gives a labeled, machine-friendly report so you or your agent can decide what to prune. It understands git worktrees, marking any branch currently checked out in another worktree so you avoid destructive surprises. The pattern is deliberately conservative: discovery only, always exit zero, and you pair the output with your own delete or archive step. It is ideal after a merge train or before a release when you want a tidy default branch graph. Compatible with agents that can run bash in a repo with standard git and worktree support.1installs4Commitcommit packages a validate.py pre-commit workflow for Claude-style agents and solo developers who want discipline without running entire monorepo linters on every small change. From the project root it detects Rust, Node, or Python layouts via marker files, maps git staged paths to the right tool, and runs cargo fmt --check, ruff check, or npm run lint only when relevant extensions are staged. That design keeps unrelated legacy debt from blocking commits whose diffs are already clean—aligned with indie velocity while still catching formatting and lint regressions you introduced. Use it in Ship whenever you are about to commit; it also helps during Build iterations when you want the same gate locally instead of waiting for CI. The skill is intermediate because hook wiring, validator tables, and exit-code semantics must match your repo’s scripts. It is a checker-style procedural skill, not a hosted CI product.1installs5Compress VideoCompress-video packages a small Python utility for solo builders who ship screen recordings, trailers, or tutorial footage and need predictable file sizes. Given an input path and `--target-mb`, it probes duration via ffprobe, subtracts the audio bitrate budget (default 128 kbps), and prints the recommended video bitrate in kilobits per second for two-pass FFmpeg `-b:v`. Diagnostics on stderr help verify duration and budget math when encodes overshoot or undershoot targets. The skill assumes FFmpeg/ffprobe are installed locally and fits terminal-driven agent sessions where the coder shells out to encoding pipelines. It is not a full GUI compressor or cloud transcoder—it is a precise bitrate calculator glue step before you run ffmpeg commands.1installs6Convert To MarkdownConvert-to-markdown is an agent skill that wraps the ezycopy CLI so solo builders can turn noisy web pages into clean markdown without ads and clutter. Use it when someone hands you a link and wants an article, doc, or competitor page saved for research, validation notes, or documentation drafts. The skill forces a deliberate mode choice: default HTTP is fast for static HTML, while --browser spins headless Chrome when rendering or auth depends on JavaScript. Execution is a single Bash invocation with optional save or clipboard targets, plus a failure path that escalates to browser mode when output is empty or suspiciously short. It pairs naturally with later writing and planning skills because it produces portable text you can cite, summarize, or drop into repos—without building a custom scraper per site.1installs7Convert Videoconvert-video is a Claude agent skill for solo builders who need general FFmpeg video work—format changes, trims, speed ramps, resize, rotation, flip, and remux—without pulling in sibling skills for compression, GIFs, social presets, or audio/frame extraction. It instructs the agent to classify the operation first, probe with ffprobe when codec or dimensions matter, build a single chained command for multi-step edits, ask for confirmation, then report output path and size. Allowed tools are scoped Bash for ffprobe and ffmpeg plus AskUserQuestion. It fits indie founders prepping launch demos, course creators normalizing footage, and developers embedding MP4 assets in docs or marketing sites. Use when triggers like convert this video, change format to mp4, trim from X to Y, timelapse, or remux appear and the task is not covered by the dedicated media skills in the same family.1installs8CouncilCouncil is a journey-wide Claude-oriented skill that turns fuzzy builder questions into a typed deliberation: you classify the user’s concern, then summon a ordered set of personas to argue from complementary angles. Architecture and design questions prioritize Architect and Skeptic; strategy and roadmap bets lead with Strategist; risk and security lean Skeptic-first; UX and developer experience foreground Advocate; innovation prompts elevate Innovator. Quick, standard, and full council sizes control cost and depth while keeping Advocate in the room unless you explicitly drop it. For solo and indie operators wearing every hat, the skill is procedural insurance against one-voice overconfidence—you get structured disagreement and pragmatic pushback before you refactor, reposition, or ship. Use it whenever a decision has tradeoffs that deserve more than a single chat reply, from API boundaries to growth bets to incident postmortems.1installs9Create AgentThis agent packages a senior reviewer mindset for solo builders who ship without a human pair on every PR. Install it when you have finished a meaningful slice of code and want correctness, clarity, idioms, and light security smell checks before you commit or open a review. The workflow reads the referenced files, infers intent, hunts edge cases and error paths, flags naming and structure problems, and outputs prioritized findings with concrete next steps. It is deliberately not a substitute for a dedicated security or architecture audit—those are separate agents in the same pattern. For Prism’s journey, it sits on the Ship shelf in Review but remains useful during Build when you iterate in tight loops. Intermediate complexity: you need enough code to review and judgment to triage critical versus style noise.1installs10Create Claw SkillCreate-claw-skill helps solo and indie builders package reusable capabilities for the OpenClaw / pi-coding-agent stack rather than default Claude Code layouts. It blends a concrete fetch-page skill template—URL argument, web_fetch, concise summary, error handling—with an OpenClaw ecosystem pattern guide so new skills follow platform conventions. Authors learn to research official behavior through the clawdocs CLI (get, search, fetch) and to design bodies that delegate work via OpenClaw session/sub-agent patterns instead of copying Anthropic-only assumptions. Use it when you are authoring or refactoring SKILL.md files for OpenClaw gateways, groups, exec tools, or when documentation drift would break generated skills. The skill matters because agent marketplaces reward portable, well-scoped skills; getting frontmatter, invocation, and doc lookup wrong wastes agent turns and breaks user-invocable flows.1installs11Create Clicreate-cli is a reference skill that walks solo builders through a complete CLI specification using the `snapr` filesystem snapshot tool as a worked example. It is meant for anyone shipping a command-line tool alongside Claude Code, Codex, or Cursor who needs predictable `--help`, versioning, quiet/verbose modes, and machine-readable list output—not ad-hoc argument parsing buried in chat. The document structures deliverables into named sections (identity, usage block, subcommand matrix, global flags) so an agent can implement or review the binary against a single source of truth. Emphasis on agent-aware patterns (NDJSON, structured stderr errors with remediation hints) reduces rework when scripts parse CLI output in headless environments. Use it when you are scoping a new CLI or refactoring an existing one for automation; it does not replace implementation, testing, or packaging.1installs12Extract Audioextract-audio is an agent skill that turns natural-language requests like “get the mp3” or “strip audio from video” into a repeatable ffmpeg workflow. Solo builders shipping courses, podcasts, or short-form content install it when they need dependable audio exports without memorizing codec tables. The skill opens with a format decision tree—music archive vs voice podcast vs DAW editing—then walks through probing streams with ffprobe and applying the matching ffmpeg flags. It sits in the build phase as a CLI integration pattern: the agent runs constrained Bash, interprets stream metadata, and can ask clarifying questions when multiple audio tracks exist. Compared to one-off chat answers, the procedural structure reduces wrong-bitrate exports and accidental re-encodes when copy would suffice.1installs13Extract Framesextract-frames is an agent skill for solo builders working with video who need reproducible stills at every shot cut without manual scrubbing in an NLE. It parses a required video path plus flags for first-only, last-only, or both frames, and can limit processing to specific 1-indexed shot numbers or override auto-detected scene thresholds. The workflow validates the file, pulls duration, resolution, codec, and frame rate via ffprobe, derives fps for accurate last-frame timing, and writes frames into a sibling directory named from the video basename. Scene detection adapts to the footage so you get a shot breakdown suitable for thumbnails, QC, ML datasets, or social clip planning. Invoke it whenever natural-language requests mention extracting frames, shot breakdowns, or scene detect—especially when you want the agent to own the ffmpeg graph instead of one-off terminal recipes.1installs14Extract LearningsExtract Learnings is an agent skill for Claude-style workflows that turns informal reminders into durable, verified memory. When you invoke phrases like extract learnings, save this for next time, or consolidate memories, it does more than append text: parallel auditor and discoverer work captures new patterns while checking existing memories for staleness. The documented hierarchy separates always-on preferences in L0 (~/.claude/CLAUDE.md), project architecture in L1, and lighter working notes in L2 so sessions stay focused instead of drowning in history. Dream and clean-up paths are explicit maintenance—you retire low-value or outdated facts and merge duplicates so indexes stay small. Solo and indie builders shipping with agents use it whenever a debugging win, API quirk, or team convention should survive the next chat. It needs read/write access to memory files and optional shell helpers; it is not a substitute for version-controlled docs or secrets management.1installs15Generate ImageGenerate Image is an agent skill that encodes capability-specific prompting patterns for image models during prompt crafting—load the relevant section in workflow step 2 when you need consistent visual output without trial-and-error chat. Solo builders use it while building landing pages, app store assets, brand kits, and marketing creatives where lens choice, lighting, isolation vs lifestyle product framing, and quoted text matter. It does not run generation itself; it teaches the agent how to phrase requests for photoreal scenes, e-commerce product isolation, cinematic hero frames, kawaii or poster styles, and advanced text-in-image layouts. Pair it with your image tool or API of choice in the same workflow. Intermediate complexity assumes you already pick a generator and iterate with follow-up edits rather than expecting one-shot perfection.1installs16Get Pr Commentsget-pr-comments packages a small Python utility that uses the GitHub CLI to pull every comment surface on a pull request—issue conversation, review summaries, and inline review lines—and emit JSON or plain text your coding agent can scan. Solo and indie builders use it when a PR has dozens of bot findings plus human notes and manually copying from the GitHub UI is error-prone. The script documents explicit usage examples with repo overrides, relies on subprocess calls to gh with timeouts, and normalizes paginated API payloads so nothing is silently truncated on busy reviews. It assumes gh is installed and authenticated for the target repository. This is an integration skill, not a code review methodology: you still decide what to fix, but you start from one organized feed. Ideal in Ship review before merge or when responding to re-review after CI updates. Intermediate users comfortable with gh auth and Python 3 execution get the most value.1installs17Get Token Insightsget-token-insights is a UserPromptSubmit hook that monitors Claude API prompt cache TTL (5 minutes) and warns users once when idle time will cause cache expiration. A solo builder uses it when working with Claude Code or similar tools to avoid wasting tokens on re-creating cached context. It matters because it prevents expensive token costs and enables proactive session management during development.1installs18Improve SkillImprove-skill is a meta agent skill for solo builders who maintain Claude Code-style capabilities and need more than gut feel to know what is broken. It guides a phased effectiveness review: state the skill’s purpose, simulate realistic requests, and document where agents get stuck, diverge on judgment calls, or hit dead ends with no documented next step. The bundled sample against create-skill shows the expected rigor—calling out Phase 0 subagent failures, inconsistent self-evaluation scores, and absent pointers to repair-skill after delivery. Use it journey-wide whenever a skill misfires in production chat, before publishing to a team catalog, or after a major SKILL.md rewrite. Outcomes are actionable findings and clearer procedural text, not a rewritten skill unless you chain a separate authoring or repair skill afterward.1installs19Make Changelogmake-changelog is a Claude-oriented release helper built around a small Python utility that inspects git history and outputs structured version ranges. Instead of manually scrolling git log between tags, you run list_ranges.py to see how many commits belong in each section—from the oldest tagged release through HEAD as Unreleased. The skill uses that map to decide which changelog blocks the agent should draft next, including incremental fill when you already shipped notes up to a known tag. Solo builders maintaining open-source libraries, CLI tools, or SaaS repos benefit because consistent release notes build trust with users and speed up store or GitHub release pages. The workflow assumes a normal tagged semver-style history; it does not replace semantic commit discipline but makes the mechanical planning step reliable and machine-readable for your agent.1installs20Make Gifmake-gif is an agent skill for solo builders who need dependable video-to-GIF exports without washed-out gradients or banding. It encodes the rule that single-pass GIF encoding is unacceptable: the agent must run palette generation on the trimmed clip, then render with paletteuse. When a user says “gif this” or “export as GIF,” the skill prompts for timing and sizing defaults, probes unusual aspect ratios with ffprobe when needed, and assembles the exact ffmpeg commands. It fits indie demos, changelog loops, and social snippets where you already have an MP4 or screen recording and want a lightweight animated asset. Shell access is limited to ffprobe and ffmpeg via allowed Bash patterns, keeping the workflow repeatable in Claude Code-style environments without guessing codec flags.1installs21Make Readmemake-readme is a documentation agent skill that standardizes how solo builders present GitHub repositories: MIT-style license shields, GitHub Actions build status, Codecov coverage, npm or PyPI version and download counters, and a separate tech-stack row using for-the-badge logos. It exists so agents stop inventing inconsistent badge URLs or nesting stack badges inside centered headers where they render poorly. You invoke it when starting a new OSS lib, CLI, or SaaS repo and want the README to look credible before you ship binaries or publish packages. The skill is intentionally narrow—URL patterns and layout rules—not full narrative README storytelling; pair it with your own features, install, and contributing sections. Because README quality affects launch discovery and grow-stage credibility, the skill spans early build documentation through pre-release polish, but Prism shelves it under Build → docs as the canonical first touchpoint when the file is created.1installs22Push Prpush-pr wraps a small Python utility that solo builders run locally to turn git history into a polished pull request body. You point it at a base branch such as main or feat/auth, and it collects commits between base and HEAD, diff statistics, and per-file change status into markdown sections suited for GitHub. The --output json mode helps agents or scripts pipe the body into gh pr create without manual copy-paste. It assumes a normal git repo on your machine and subprocess access to the git CLI. Use it at the end of a feature branch when you want consistent PR narratives across solo projects, especially when you ship frequently from Claude Code or Cursor. It does not open the PR for you unless you combine it with your host CLI; it focuses on generating the description content reliably from facts in git.1installs23Recall ConversationsRecall Conversations is a journey-wide Claude-oriented agent skill for solo builders who accumulate long coding and planning threads and need structured memory without rereading every message. Command recipes in SKILL.md fetch sessions; this lens reference tells the agent which analytical questions to apply and when to deepen search. Restore-context highlights unfinished work and next steps for fresh sessions. Extract-learnings and find-gaps turn repeated confusion into instruction targets. Review-process and run-retro examine whether planning preceded coding and where time leaked. Extract-decisions and find-antipatterns capture trade-offs, rejections, and corrections the user issues more than once—candidates for durable CLAUDE.md rules. Complexity is intermediate because choosing the right lens and follow-up still requires judgment. Use it whenever agent continuity breaks, you suspect repeated mistakes, or you want decisions persisted—across idea research, validate scoping, build implementation, ship review, and operate iteration. It does not replace git history or production monitoring; it augments conversational evidence from your agent tool.1installs24Repair AgentRepair Agent is an agent skill for solo builders who already ran an agent audit and need a disciplined rewrite instead of vague “make it better” prompts. It lives primarily in Build under agent-tooling but also supports Ship review when quality gates flag routing or instruction-contract failures. The bundled sample shows a log-summarizer agent failing description routing patterns and first-person instructions, then maps each finding to concrete fixes—imperative steps, folded YAML descriptions, and explicit when-to-use openers. Use it when your subagent descriptions say “This agent should be used when” or bodies read like self-narration, because those patterns break tool routing and compliance checks. The outcome is a repaired agent markdown file your orchestrator can trust, with an improvement report that documents what changed and why.1installs25Repair Skillrepair-skill from claudest is meta procedural knowledge for authors and agents running a seven-dimension structural audit on agent skills. It does not replace the audit itself; you load it first so reviewers apply rules correctly. The documented false-positive patterns matter in practice: an entirely missing allowed-tools field is not a security or policy failure because unrestricted default is valid—only an incomplete explicit list should be flagged when the skill uses omitted tools. Likewise, domain orientation that supports in-flight decisions is not the same as a "When to Use This Skill" routing section; conflating them produces bogus violations. Solo builders curating Prism-quality skills or maintaining private skill libraries should invoke this whenever calibration drift causes over-flagging, across build, ship, and operate iterations on skill packs.1installs26Run ResearchRun Research is a claudest agent skill for solo builders who need fast, multi-channel signal before writing specs or code. Instead of bouncing between tabs, the skill documents how to wire Reddit via reddit-cli (script app credentials in ~/.secrets), X via bird using your logged-in browser cookies, YouTube via yt-dlp plus the bundled yt_research.py helper, and the open web via brave-cli with a Brave Search API key. It fits the moment you are exploring niches, complaints, competitors, or content angles—not when you already have a frozen PRD. Expect intermediate setup: shell installs, API keys, and optional manual AUTH_TOKEN/CT0 for X if cookie auth fails. Use across Idea research, Validate competitor checks, and Grow content mining once sources are configured.1installs27Search YoutubeSearch YouTube is an agent skill for solo builders who validate ideas and learn fast by mining practitioner video—not only blog posts. You pose a concrete research question; the skill finds relevant YouTube material and returns synthesized findings, areas of creator agreement, and practical takeaways you can drop into a spec, landing copy, or technical plan. It fits the Idea phase when exploring competitors, audience language, or how complex systems (like browser rendering) actually behave in production talks. It also helps Grow and Build when you need tutorial-backed answers for content or implementation choices. Treat it as structured video research: faster than manual playlist watching, more narrative than raw transcript dumps, and aimed at decision-ready notes for one-person teams.1installs28Share Socialshare-social is an agent skill that prepares master videos for social networks using documented platform presets and ffmpeg/ffprobe. Solo creators and indie marketers invoke it when they need Instagram Reels at 9:16, YouTube Shorts caps, square Twitter assets, or LinkedIn-friendly 16:9 exports without guessing bitrates. The SKILL.md centers on a comparison table for eight destinations—Instagram Feed and Reels, TikTok, YouTube Shorts and standard 1080p, X/Twitter, Facebook, and LinkedIn—each with max resolution, duration limits, video bitrate guidance, and AAC audio targets. When wording is vague, the skill prompts for the target platform because vertical crop alone is not enough to pick encode parameters. It fits builders shipping content-led products, launch teasers, or growth clips who already have a render and need a reliable handoff file rather than a full editing suite in chat.1installs29Update Claudemdupdate-claudemd is a Claude Code skill for solo builders whose project instructions have drifted from the repo. It walks a structured reconciliation: read CLAUDE.md and all `.claude/rules/*.md` files, compare them to the current tree and commits since each file’s last change, remove noise, split overflow into scoped topic files, and ensure every topic file carries `paths:` frontmatter so SessionStart loading stays targeted. The governing principle is behavioral impact—if a line does not change agent decisions, it belongs in code comments or nowhere. Scope is project-level CLAUDE.md only; user-global `~/.claude/CLAUDE.md` stays out unless you ask. Invoke after refactors, when onboarding new contributors, or when sessions feel verbose and contradictory. The outcome is a lean index of invariants plus pointers, with subsystem rules that auto-load only when relevant files are touched—better token efficiency and fewer wrong assumptions across idea through operate work.1installs30Update Readmeupdate-readme is a documentation workflow skill for solo builders and small teams whose README has drifted from the product. It assumes a substantial README already exists; otherwise it hands off to make-readme for greenfield creation. The agent always invokes make-changelog first so version and change sections reflect git reality, then launches parallel research across the tree and history before rewriting sections. Triggers match natural chat phrases like “README is outdated” or “sync README with the codebase.” It fits the Build docs shelf but pays off again at Launch when distribution pages and repo landing copy must match what you ship. Intermediate complexity: you need git history and comfort with delegated sub-agents. Outcome is a current README aligned with changelog intelligence—not a marketing rewrite unless your existing doc already aimed there.1installs