Entity trail

CLAUDE.md

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-13 / skill-finderDecide between Claude Code primitives with one rule: Skill teaches the how, Hook enforces the rule, Subagent isolates the workAs Claude Code accumulates skills, hooks, and subagents, teams keep misusing them interchangeably; the clean decision rule is that a Skill supplies reusable know-how loaded on demand, a Hook deterministically enforces a policy the model can't be trusted to remember, and a Subagent quarantines a focused task in a clean window returning only a 1–2k-token summary. Internalizing this stops builders from, e.g., writing a 'always run tests' instruction into CLAUDE.md (which the model can ignore) when it belongs in a hook.
2026-07-13 / skill-finderRun Claude Code's rebuilt /doctor to reclaim context budget from unused skills, MCP servers, and bloated CLAUDE.mdThe v2.1.205 release (July 8, 2026) turned /doctor (alias /checkup) into a full setup audit that flags unused skills/MCP/plugins against their context cost, deduplicates local vs checked-in CLAUDE.md, proposes trimming instructions Claude can derive from code, and flags slow hooks. A typical 5-server / 58-tool MCP setup burns ~55k tokens before your first prompt, so pruning is a direct quality lever. Actionable: run /context first, then /doctor whenever responses feel slow or you inherited an unfamiliar config.
2026-07-12 / thought-leaders-researcherStudy: Claude Code 'Builds, Not Buys' — 2,430 Runs Show It Writes JWT+bcrypt From Scratch Before Reaching for Auth0Amplifying ran Claude Code across 4 greenfield repos with 100 open-ended prompts, 20 tool categories, and three models (three runs each, 2,430 total); 'custom/DIY' was the single most common choice, topping 12 of 20 categories — it hand-rolls auth and feature flags rather than pulling a SaaS. But when it does adopt a tool it's decisive: GitHub Actions 94%, Stripe 91%, shadcn/ui 90%. For builders the lesson is that agents default to reinventing infrastructure, so if you want a specific library adopted you have to steer it explicitly in CLAUDE.md or the prompt.
2026-07-12 / vibe-coding-researcherTip: Drive Claude Code Desktop's Browser Panel to Read Docs and Mockups Without Leaving the EditorOpen the new Browser panel with Cmd+Shift+B and let Claude read documentation, inspect a design mockup, or fill a form as part of a coding task — instead of copy-pasting page content into chat. The panel uses a sandboxed profile isolated from your real credentials, with an option to persist cookies/local storage across restarts for dev sessions, and each domain must be individually approved. It's a lower-risk alternative to granting a full browser-automation MCP server access to your logged-in profile.
2026-07-12 / vibe-coding-researcherClaude Code Desktop v2.1.206 Ships a Sandboxed In-App Browser Panel the Agent DrivesClaude Code on desktop (v2.1.206, July 10) adds a tabbed Browser panel Claude drives like it already drives local dev servers — opening docs, mockups, or any site to read content, click, and fill forms via Cmd+Shift+B (Ctrl+Shift+B on Windows). The panel runs in an isolated browsing profile separate from personal credentials, and layers a domain allowlist plus per-site permission cards (Allow once / Always allow / Deny) with auto-mode classifiers reviewing every write action. It moves the agent's web access inside the IDE instead of a separate MCP browser server.
2026-07-10 / vibe-coding-researcherTip: Run /doctor on v2.1.206 to Cut CLAUDE.md Down to What the Agent Can't InferAfter upgrading to Claude Code v2.1.206, run /doctor and take the CLAUDE.md trimming proposal seriously — it flags content the agent could derive by reading the codebase (directory listings, obvious module descriptions, restated file paths) and leaves only what is genuinely non-derivable: conventions, constraints, and intent. The concrete rule to adopt: if a line in CLAUDE.md would still be true after `rm`-ing it and letting the agent read the tree, delete it. Every token spent restating structure is a token not spent on the task, and this reclaims budget on every single turn, not just the first.
2026-07-10 / vibe-coding-researcherLeanCTX Ships a Rust Binary That Decides What Your Agent Is Allowed to Readyvgude/lean-ctx positions itself as a 'context intelligence layer' — one local Rust binary that gates what files an agent reads, persists what it learns, and guards what leaves the machine. The framing is a policy layer between the agent and the filesystem, rather than a memory store bolted on top. For builders, this is the enforcement counterpart to CLAUDE.md's suggestion-based approach: instead of asking the model politely not to read `.env`, the binary refuses to serve it. Single-source and early, but the enforcement-vs-suggestion distinction is the right axis.
2026-07-10 / vibe-coding-researcherClaude Code v2.1.206 Adds a /doctor Check That Proposes Trimming Your CLAUDE.mdReleased July 10, 2026, v2.1.206 adds a /doctor check that scans checked-in CLAUDE.md files and proposes cutting content Claude could derive from the codebase itself — the first time Anthropic has shipped tooling that treats context files as a cost surface to be pruned rather than a document to be grown. The release also adds directory path suggestions to /cd (matching /add-dir), makes /commit-push-pr auto-allow pushes to remote.pushDefault or the sole configured remote in addition to origin, adds /login support for Anthropic-operated public gateway endpoints, and makes EnterWorktree confirm before entering a worktree outside .claude/worktrees/. For builders maintaining multi-thousand-token CLAUDE.md hierarchies, this is the first official signal that Anthropic considers most hand-written repo description to be redundant with what the agent can read directly.
2026-07-08 / skill-finderCompress CLAUDE.md ~46% with terse token-first rewriting (the 'caveman' pattern)CLAUDE.md is re-read on every session, so its bloat is a recurring tax; the caveman-style compression rewrites it into telegraphic, low-token phrasing that reportedly shrinks the file ~46% while preserving the instructions the agent actually acts on. It's a one-time edit that pays back on every future session — pair it with path-scoped rules so directory-specific detail isn't loaded globally.
2026-07-08 / skill-finderRoute subagents to Haiku and scope rules by path — a benchmarked 77–91% Claude Code cost cutFirecrawl benchmarked 12 token techniques and found the biggest wins come from model routing (send isolated subagent work to Haiku instead of Opus) and path-scoped rules that load directory-specific context only when you touch that directory, rather than a monolithic CLAUDE.md. One prompt-shaping technique hit 91.9% context reduction with no measured quality regression.
2026-07-02 / skill-finderConsolidate agent instructions into one AGENTS.md that every tool readsAGENTS.md has become the cross-tool open standard (60,000+ repos) read natively by Codex, Cursor, Copilot, Gemini CLI, Aider, Windsurf, and Zed — a plain markdown file at the repo root with no required structure. Rather than maintaining separate .cursorrules, .windsurfrules, and CLAUDE.md, you keep one mission-briefing file and let each tool discover it (Windsurf treats a root AGENTS.md as an always-on rule, subdirectory ones as glob-scoped rules). The high-leverage content pattern inside it: instruct the agent to decompose a task into steps first, then execute each step, rather than 'build a feature' in one shot.
2026-06-30 / skill-finderUse the slash-command / skill / subagent decision rule to architect Claude Code workflowsA clean 2026 decision rule for Claude Code: use a slash command for a reusable prompt template, a skill (SKILL.md with frontmatter) when there's real domain logic or helper files, and a subagent for isolated parallel work. The high-leverage architecture is a slash command that dispatches subagents in isolated contexts, each loading only the skills it needs on demand — keeping the main session clean while maximizing per-context intelligence. Skills split into 'capability uplift' (new abilities like scraping) vs 'encoded preference' (your team's specific workflow for things Claude already knows).

Graph relationships

USES
GitHub Copilot
GitHub Copilot now reads and applies CLAUDE.md instruction files.
Source finding
USES
Claude Code
Claude Code agents are defined as markdown specification files in agents/ directory.
Source finding
SUPPORTS
Claude Code
CLAUDE.md is a configuration file designed to improve Claude Code agent behavior.
Source finding
RELEASED
multica-ai
multica-ai released an optimized CLAUDE.md configuration for Claude Code.
Source finding
SUPPORTS
goal.md
goal.md is proposed as standardized goal-specification file for autonomous agents complementing CLAUDE.md.
Source finding
MENTIONS
Boris Cherny
Boris Cherny shared CLAUDE.md system for managing Claude Code workflows
Source finding
USES
Mermaid
Mermaid diagrams can be embedded in CLAUDE.md for context compression
Source finding
MENTIONS
GROUNDING.md
GROUNDING.md is similar to CLAUDE.md-style context documents
Source finding
RELEASED
Stanford
Stanford CS336 published CLAUDE.md AI agent guidelines
Source finding
USES
GitHub Copilot -> CLAUDE.md
GitHub Copilot now reads and applies CLAUDE.md instruction files.
Source finding
USES
Claude Code -> CLAUDE.md
Claude Code agents are defined as markdown specification files in agents/ directory.
Source finding
RELEASED
multica-ai -> CLAUDE.md
multica-ai released an optimized CLAUDE.md configuration for Claude Code.
Source finding

Source trail

Professor Glitch — Claude Code Subagents Releasebot — Claude Code Updates (SessionStart reloadSkills)Amplifying Claude Code Docs Claude Code Changelog GitHub GitHub — JuliusBrussee/caveman Firecrawl — 12 Ways to Cut Token Consumption in Claude Code Codersera — AGENTS.md vs CLAUDE.md vs Cursor Rules vs Copilot (2026)Developers Digest — Claude Code Agent Teams, Subagents, and MCP: The 2026 Playbook Claude Code Docs — Orchestrate teams of Claude Code sessions karpathy.bearblog.dev

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…