GenAI PMClaude.md
A documentation convention for organizing Claude-related instructions or skills. The newsletter frames it as part of writing lean system prompts and modular skills.
Key Highlights
- Claude.md is a documentation convention for storing reusable instructions and context that steer Claude-based agents.
- Recent coverage positions Claude.md as part of a broader control stack that includes skills, hooks, rules, and subagents.
- The strongest practical lesson is that lean, modular documentation often works better than long, over-engineered system prompts.
- AI PMs can use Claude.md to standardize behavior, improve maintainability, and create feedback loops for agent quality.
Claude.md
Overview
Claude.md is a documentation convention for packaging the instructions, context, rules, and reusable behaviors that guide Claude-based agents. In practice, it acts as a lightweight control layer: teams can store project-specific guidance in a standardized file and combine it with related mechanisms like skills, hooks, rules, and subagents. Across the newsletter coverage, Claude.md is framed less as static documentation and more as an operational interface for steering agent behavior reliably.For AI Product Managers, Claude.md matters because it turns prompt design into a maintainable product artifact. Instead of burying critical behavior in long, fragile system prompts, teams can modularize instructions, version them, import them into tools, and revise them as model behavior changes. The broader lesson from recent mentions is that effective Claude.md usage is usually about clarity and minimalism: lean instructions, targeted conditional logic, and continuous pruning outperform bloated prompt stacks.
Key Developments
- 2026-01-13 — Bolt added support for Claude.md imports, making it easier for agents to ingest project context without manual setup.
- 2026-03-18 — HumanLayer published guidance on getting Claude to reliably read Claude.md, including the use of conditional XML blocks to surface the right instructions at the right time.
- 2026-06-01 — Josh Pigford described using a "/learnings" skill to turn transcripts and code changes into updates for Claude.md, positioning it as a living memory and operating guide.
- 2026-06-18 — Jason Zhou argued that Claude.md assets depreciate as models improve, recommending regular rewrites and removal of rules that no longer materially steer behavior.
- 2026-06-19 — Claude Code announced new steering controls centered in part on CLAUDE.md files, alongside skills, hooks, rules, and subagents.
- 2026-07-25 — Thariq shared lessons from cutting roughly 80% of the Claude Code system prompt, emphasizing lean prompts, modular skills, and standardized Claude.md documentation.
Relevance to AI PMs
- Standardize agent behavior across products and teams. AI PMs can use Claude.md as a shared contract for how assistants should behave in a product, reducing inconsistency across environments, workflows, and contributors.
- Improve prompt operations and maintainability. Rather than continually expanding system prompts, PMs can separate durable guidance into modular documentation, then test, prune, and version it as part of product operations.
- Build feedback loops into agent improvement. Patterns like "/learnings" show how session outcomes can be converted into updates to Claude.md, giving PMs a practical way to operationalize postmortems, QA findings, and user feedback.
Related
- Claude / Claude Code — The primary ecosystem where Claude.md appears as a steering and documentation pattern for agent behavior.
- Skills, hooks, rules, subagents — Complementary control mechanisms often used alongside Claude.md to modularize capabilities and constraints.
- HumanLayer — Published practical tactics for making Claude attend to Claude.md more reliably.
- Bolt — Added Claude.md import support, highlighting interoperability and easier context ingestion.
- Josh Pigford, Jason Zhou, Thariq — Builders and practitioners who shared concrete operating lessons: learning loops, aggressive simplification, and ongoing prompt maintenance.
- Conductor — Mentioned in adjacent workflows where Claude.md is part of a broader autonomous agent stack.
Newsletter Mentions (6)
“𝕏 Thariq cut ~80% of the Claude Code system prompt for the newest models and shares concrete lessons on writing lean system prompts, designing modular skills, and standardizing Claude.MD documentation.”
𝕏 Thariq cut ~80% of the Claude Code system prompt for the newest models and shares concrete lessons on writing lean system prompts, designing modular skills, and standardizing Claude.MD documentation.
“Announces new steering controls for Claude Code, including CLAUDE.md files and mechanisms like skills, hooks, rules, and subagents to guide agent behavior.”
📝 Claude Code Blog Steering Claude Code: CLAUDE.md files, skills, hooks, rules, subagents and more - Announces new steering controls for Claude Code, including CLAUDE.md files and mechanisms like skills, hooks, rules, and subagents to guide agent behavior.
“Jason Zhou warns that CLAUDE.md assets depreciate as the model evolves, so he routinely clears and rewrites them—keeping only the few rules that truly steer behavior—and argues that over-engineering, not lack of context, is most people’s real issue.”
#21 𝕏 Jason Zhou warns that CLAUDE.md assets depreciate as the model evolves, so he routinely clears and rewrites them—keeping only the few rules that truly steer behavior—and argues that over-engineering, not lack of context, is most people’s real issue.
“He uses a “/but-for-real” skill to coerce the AI into rechecking its own output for additional errors and a “/learnings” skill that distills session transcripts and code changes into updates for his CLAUDE.md guidelines.”
#3 ▶️ The Exact AI Skills This Solo Founder Uses to Build 5 Apps at Once | Josh Pigford Peter Yang Josh Pigford demonstrates his autonomous AI stack—combining Conductor-powered 4-step “/build” with Opus, a GPT-3.5 “/adversarial-code-review,” a “/but-for-real” error checker, and a “/learnings” updater of CLAUDE.md—to solo-build and launch five AI products in parallel.
“Getting Claude to Actually Read Your CLAUDE.md - Shows how to use conditional XML blocks inside CLAUDE.md to ensure Claude attends to the right instructions at the right times.”
#11 📝 HumanLayer Blog Getting Claude to Actually Read Your CLAUDE.md - Shows how to use conditional XML blocks inside CLAUDE.md to ensure Claude attends to the right instructions at the right times. These patterns help make agent behavior more reliable by prioritizing context and instructions.
“Bolt @boltdotnew added support for Claude.md imports, allowing agents to ingest project context instantly without manual setup.”
Bolt @boltdotnew added support for Claude.md imports, allowing agents to ingest project context instantly without manual setup. See update .
Related
An AI coding assistant environment used for running evaluation skills and agentic workflows. In this issue it is mentioned as a runtime for ai-evals-course material and as an agent in an OpenRouter-like system.
Anthropic’s assistant, discussed here for shared memory across chat and Cowork. The feature is relevant to PMs because it enables cross-task context reuse and user-controlled memory.
Person who shared an agent skill in the treg repository. Relevant to PMs because it showcases community distribution of reusable agent behaviors.
An AI commentator or builder referenced here for comparing OpenAI’s Computer History with Familiar. He highlights Familiar’s offline, local, and model-agnostic qualities.
A protocol or capability layer mentioned as part of an open, composable extension philosophy for AI tooling. It is grouped with MCP and Plugins.
A commenter who described Claude’s automode as safer than other permission systems and noted its default rollout. The mention is relevant to autonomy and permissions in agent UX.
Specialized subordinate agents used to break down and orchestrate tasks. The newsletter mentions them as part of Claude Code steering controls.
An AI coding IDE and collaboration platform focused on tasks, agent sessions, artifacts, and workflows from questions through implementation. It is positioned as a productivity platform for engineers using AI APIs and agents.
A development tool with a skills system for automating workflows. The newsletter highlights stacked skills, automatic invocation, and playbook loading.
Stay updated on Claude.md
Get curated AI PM insights delivered daily — covering this and 1,000+ other sources.
Subscribe Free