Claude Patterns
adelaidasofia/claude-patterns
Detecta padrões implícitos em sessões e os transforma em regras e capturas explícitas
Ver no GitHubPlugin que escaneia sessões recentes, decisões e anotações para identificar padrões recorrentes e os transforma em capturas concretas: regras CLAUDE.md, notas de conceito, sementes de escrita e melhorias de skills. O "Instinct Engine" torna explícito o reconhecimento de padrões implícito e armazena em um lugar acessível à próxima sessão.
Instalação
npx convez-ai adelaidasofia/claude-patterns1 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
- Consolidar decisões e lógicas recorrentes em regras reutilizáveis
- Capturar insights e metáforas que ressurgem sem você notar
- Automatizar a geração de sementes de escrita baseadas em padrões reais de uso
- Manter CLAUDE.md e vault sincronizados com aprendizados implícitos
Veredito

Ferramenta criativa para quem trabalha com Claude Code e mantém vault pessoal. Resolve um problema real: capturar padrões antes que evaporem. Requer setup manual de caminhos e integra bem com ecossistema de plugins (daily-journal, insights), mas tem curva de onboarding e depende de disciplina para manter artefatos estruturados.
README
A tradução para o português deste README ainda está na fila.
claude-patterns
A Claude Code plugin that scans recent sessions, journal entries, and decisions for recurring patterns — then turns them into concrete captures: CLAUDE.md rules, concept notes, writing seeds, skill improvements.
The Instinct Engine. It catches what's hardening into real insight before it evaporates.
What it does
You run /patterns and Claude:
- Pulls in your last session, last 10-15 decisions, recent drafts, and last 7 journal entries
- Scans for recurring themes, repeating decision logic, recurring metaphors, behavioral loops, and vault gaps
- Surfaces up to 5 specific findings, with the actual quotes and the proposed capture
- After you confirm, writes the rules, drafts, concept notes, or skill notes in one pass
- Logs the run
It also runs semi-autonomously: at the end of every session, Claude can silently check four auto-detection triggers (tool-call friction, corrections, dead-end recovery, non-trivial discoveries) and surface a one-line suggestion if any fire.
The point is to make implicit pattern recognition explicit — and to capture the pattern in a place the next session can see, not in your head where the next session can't.
Install
Open Claude Code, paste:
/plugin marketplace add adelaidasofia/claude-patterns
/plugin install claude-patterns@claude-patterns
Then in any Claude Code session:
/patterns
On first use, Claude asks where your sessions, decisions, journals, drafts, and concept notes live. Takes 60 seconds, never asks again.
claude plugin add github.com/adelaidasofia/claude-patterns
Requirements
- Claude Code
- A vault or notes folder with at least one of: a last-session file, a decision log, a journal, a drafts folder
- Nothing else
Optional integrations:
- The daily-journal plugin —
/patternsreads the last 7 entries automatically if your journal lives in the path you configured - The insights plugin — if a weekly review was just run,
/patternsuses that output instead of re-scanning journals
What gets captured
Writing seed → a draft with the exact phrasing + context of where the pattern appeared. The most original material is usually the metaphor that resurfaced three times across two months without you noticing.
CLAUDE.md rule → a new rule added to the relevant section. Decision frameworks that repeat ("when X, I always do Y because Z") become rules so the next session doesn't re-derive them.
Concept note → for ideas that keep appearing in passing but never got their own page.
Skill improvement → flagged for the next skill update. If a workflow keeps requiring manual steps, the workflow itself needs to change.
The four auto-detection triggers
At session end, Claude silently checks for:
- High tool-call friction — 5+ steps for something routine? Missing shortcut or rule.
- User correction — you said "no, not that way"? Worth checking if this echoes prior corrections.
- Dead-end recovery — Claude backtracked before finding the working path? Preserve the working path.
- Non-trivial discovery — found something non-obvious about a tool, API, or workflow? May warrant a rule.
If any fire, Claude offers a one-line prompt — never a wall of text. You can say "yes" (run full /patterns), "just save it" (capture the specific finding), or "no" (skip).
The full scan never auto-runs. Only suggestions.
How this differs from the insights plugin
- insights is periodic review (weekly, monthly). It looks at floor trends, behavioral patterns, and panel commentary across a defined time window.
- patterns is signal extraction. It looks at what's repeating across the most recent ~2 weeks and proposes concrete captures.
Use insights for "how was this week." Use patterns for "what's hardening into something I should write down."
Submitting to the Cowork marketplace
Once the repo is on GitHub, submit at: clau.de/plugin-directory-submission
For the community marketplace:
claude plugin marketplace add anthropics/claude-plugins-community
Telemetry
This plugin sends a single anonymous install signal to myceliumai.co the first time it loads in a Claude Code session on a given machine.
What is sent:
- Plugin name (e.g.
slack-mcp) - Plugin version (e.g.
0.1.0)
What is NOT sent:
- No user identifiers, names, emails, tokens, or API keys
- No file paths, message content, or anything from your work
- No IP address is stored after dedup processing
Why: Helps the maintainer know which plugins people actually install, so attention goes to the ones that get used.
Opt out: Set the environment variable MYCELIUM_NO_PING=1 before launching Claude Code. The hook will skip the network call entirely. Already-pinged installs leave a sentinel at ~/.mycelium/onboarded-<plugin> — delete it if you want to reset state.
License
MIT