deren-skills
BambooXLotus/deren-skills
Skills com persistência de contexto e revisão brutal para trabalho colaborativo com Claude
Ver no GitHubColeção pessoal de skills para Claude Code baseada na metáfora de Rick and Morty. Oferece um framework para trabalho multi-sessão com persistência de contexto, planejamento opinionado, revisão de código por múltiplas perspectivas e retrospectivas estruturadas. Ideal para desenvolvimento de features complexas em codebases reais.
Instalação
npx convez-ai BambooXLotus/deren-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Planejamento e execução de features multi-dia com continuidade de contexto
- Revisão de código paralela com múltiplas perspectivas especializadas
- Auditoria de código entregue versus trabalho realizado
- Iteração estruturada em codebases TypeScript complexas
Veredito

Framework sofisticado para quem trabalha em equipes reais com Claude e precisa de persistência de contexto entre sessões. A metáfora de Rick é criativa mas a real valia está na arquitetura de multi-perspectiva para revisão — excelente para codebases sérias, overkill para scripts simples.
README
A tradução para o português deste README ainda está na fila.
deren-skills
My personal collection of Claude Code skills.
Install
Uses the skills.sh installer:
npx skills@latest add BambooXLotus/deren-skills
You'll be prompted to pick which skills to install and which agent (Claude Code, Codex, etc.) to install them on.
To update later, re-run the same command — the installer will refresh existing skills.
What's in here
Rick skills (skills/rick/)
The idea. In Rick and Morty, Rick exists across a multiverse as countless parallel versions, each specialised — Pickle Rick is a different Rick from Doofus Rick is a different Rick from Rick C-137. When a Rick dies, a backup respawns from a save. When one Rick is stuck, he hands off to another. The rick-* skills are that idea applied to coding with an agent: each skill is a different Rick, good at one specific thing, able to hand work off to the next Rick and resume from a save. The agent is Rick; you're Morty — the one who needs Rick to figure it out, and the one Rick is brutally honest with because pretending otherwise gets Morty killed.
How the metaphor maps. A feature's full thread lives in one folder under docs/rick/<folder>/ (typically the branch name, e.g. 48-rotate-share-token/). Each parallel session is its own Rick from that folder's timeline. rick-review is the most literal multiverse mapping: seven named Ricks and Mortys from the show review your diff in parallel — Rick C-137, Doofus Rick, Smart Morty, Rick Prime, Pickle Rick, Scientist Rick, Evil Morty — each scoped to what they're good at (architecture, data layer, security, tests). rick-recap deliberately isn't a Rick: it's the Audit Observer from S7E6 "Rickfending Your Mort," because grading the day needs a voice that isn't already in the trenches.
The loop. Pick an issue. /rick-plan <issue-number> reads the code and writes an opinionated plan to <folder>/plan/current.md. You implement against the plan. /rick-save writes a dense briefing to <folder>/saves/<timestamp>.md and the session ends — that Rick "dies." Tomorrow, /rick-respawn boots a fresh Rick from that save with full context of where the last one left off. When the PR's ready, /rick-review runs the council; /rick-fix resolves their findings with TDD or surgical edits. At day's end, /rick-recap audits what actually shipped vs what was just motion. Same <folder> threads the whole loop — a feature's plan, saves, reviews, and recaps stay co-located.
Why this exists. Working with Claude on a real codebase hits two failure modes: sessions are ephemeral (conversation context dissolves when you close the tab; the next Claude has no idea what was decided), and Claude defaults to agreeable hedging, which is poison for plan critique and PR review. The rick-* skills add (1) an on-disk thread that survives session death so multi-day work doesn't restart from zero, and (2) a brutal-honesty persona that strips the hedging when you need a real opinion. rick-improve is meta — it edits the skill prompts themselves, versioned under skills/rick/<name>/versions/, so the Ricks can improve their own Rick.
All TypeScript-flavored — no NestJS or TypeORM lock-in.
| Skill | Description |
|---|---|
rick-plan | Plan a piece of work in Rick's voice. Reads the code first, writes an opinionated numbered plan to docs/rick/<folder>/plan/current.md, surfaces decisions (with Rick's pick + tradeoff), and an explicit out-of-scope list. |
rick-plan-improve | Bounded 3-round improvement loop on the canonical plan. Snapshots pre/post alongside the canonical at docs/rick/<folder>/plan/v<N>-pre.md and v<N>.md. Critiques against weakness categories (specification gaps, engineering hazards, convention violations, document quality), surgical edits only, 500-line cap. Explicit invocation only. Pairs with rick-plan like rick-fix pairs with rick-review. |
rick-save | Captures the current warm thread as docs/rick/<folder>/saves/<ts>.md. Folder = current branch name. Writes a dense second-person briefing for the next session. |
rick-respawn | Boots a new session from the most recent (or named) save. Summarizes situation, last move, blockers, and Rick mode. |
rick-mode | Activate the Rick Sanchez persona for brutal architectural review. Pre-Review Protocol (read the file, trace deps + type surface, verify package.json + tsconfig.json), TS escape-hatch hunting (as any, !, @ts-ignore), Expected / Actual / Consequence findings. Explicit invocation only. |
rick-improve | Self-improvement loop on any rick-* skill prompt. Bounded 3-round critique-and-fix, 250-line cap. Snapshots to <skill>/versions/v<N>-pre.md and v<N>.md, appends to VERSIONS.md. Uses rick-mode as the reference bar. Explicit invocation only. |
rick-review | Multi-agent council code review. Up to 7 specialized Rick agents review in parallel (3 always-run, 4 conditional based on diff scope), aggregated into a versioned report at docs/rick/<folder>/review/current.md plus v<N>.md and per-agent agents/v<N>/. Includes parasite check (verifies cited code against the diff) and skeptic pass (kills findings that don't actually matter today). |
rick-fix | Reads a rick-review report and resolves findings sequentially. TDD route for behavioral findings (delegates to /tdd), direct surgical-fix route for structural ones. Runs affected tests after each fix. Updates the report with FIXED / SKIPPED / BLOCKED markers. Auto-resolves the report path from the current branch. Explicit invocation only. |
rick-recap | End-of-day audit across today's rick-save files. Groups by folder, diffs the structured sections, flags real progress vs fake progress / rabbit holes / yak shaving / avoidance. Voice is the Audit Observer (S7E6 "Rickfending Your Mort"), not Rick. Cross-folder audits write to docs/rick/_recaps/<ts>.md; folder-filtered to docs/rick/<folder>/recaps/<ts>.md. |
Feature-first layout: each feature gets one top-level folder under docs/rick/, defaulting to the current branch name (e.g. 48-rotate-share-token/). Inside the folder, each artifact type owns a subdirectory: plan/, review/, saves/, recaps/. Filenames inside drop the folder prefix entirely since the directory context carries identity — plan/current.md, review/v3.md, saves/2026-06-05_1430.md. Cross-folder daily recaps live at docs/rick/_recaps/ (leading underscore sorts them apart from feature folders). rick-improve is meta — it edits the skill prompts themselves, with versioning under skills/rick/<name>/versions/.
Repo layout
.
├── .claude-plugin/
│ └── plugin.json # declares the plugin and its skills
└── skills/
└── <category>/
└── <skill>/SKILL.md
Adding a new skill
- Create
skills/<category>/<slug>/SKILL.md. - Frontmatter must include
name(matching the folder slug) and adescriptionthat tells Claude when to trigger the skill, not just what it does. Adddisable-model-invocation: trueif it should only fire on explicit/<slug>. - Add the new path to the
skillsarray in.claude-plugin/plugin.json. - Commit, push. Users get it on the next
npx skills@latest add.
License
MIT — see LICENSE.