project-instructions

Writes and audits project instructions for Claude Chat Projects and Cowork Projects.

1|Updated Apr 19, 2026
One-click install
npx skills add https://github.com/fercosnt/fernando-claude-marketplace --skill project-instructions-fercosnt
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: project-instructions
Source: https://github.com/fercosnt/fernando-claude-marketplace/tree/main/prompt-suite/skills/project-instructions
Command: npx skills add https://github.com/fercosnt/fernando-claude-marketplace --skill project-instructions-fercosnt

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Project instructions often fail because rules are written in the wrong layer: facts pasted into instructions instead of knowledge, one-off tasks frozen as permanent rules, or conversational tone guidance given to an agentic Cowork project that needs deliverables and save paths. This Skill routes every rule to the correct layer (profile, project instruction, knowledge, memory, or skill) before writing, and audits existing instructions that Claude ignores. ## Core Features & Use Cases - Layer routing: Classifies each rule against a durability test (durable, non-obvious, no better home) and assigns it to the right layer for Chat Projects or Cowork Projects, which have opposite mental models. - Three modes: CREATE (interview of max 7 questions, routing table, then drafting within a 1,500-2,500 character budget), AUDIT (named-failure catalog such as instruction-resume, dead rule, encyclopedia, silent conflict, orphaned negative), and PACKAGE (full project setup with folders, connectors, skills, and a versioned source file for teams). - Platform-accurate guidance: Encodes verified facts such as Cowork not reading ~/.claude, Cowork projects being local-only, 50 MB file limits, and the documented organization-over-individual precedence. - Use Case: A user pastes a 9,000-character instruction that Claude ignores; the Skill counts and classifies every fragment, names each failure, moves the pricing table to knowledge, and delivers a full rewrite with concrete test prompts. ## Quick Start Ask the assistant to write or review the instructions for your Claude project, pasting any existing instruction text and mentioning whether it is a Chat Project or a Cowork Project.

Frequently Asked Questions about project-instructions

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write good project instructions for Claude?

Start by identifying the target (Chat Project or Cowork Project), then interview for the work, deliverable, audience, and known errors. Route every rule through a durability test, keep Chat Project instructions between 1,500 and 2,500 characters, and include an owner and review date header.

Why does Claude ignore my project instructions?

The most common causes are instructions that are too long, silently conflicting rules, or rules that belong in knowledge or a skill instead. Audit by counting characters and rules, naming each failure pattern, and rewriting the full text rather than patching fragments.

What is the difference between Chat Project and Cowork Project instructions?

Chat Project instructions govern how responses are written, while Cowork Project instructions govern what work gets done: deliverables, definition of done, save locations, and permission boundaries. Copying one style into the other produces a project that talks well but does not deliver.

Can I share a Cowork project with my team?

No. Cowork projects live only on the local machine and do not sync or share. The workaround is a versioned source file (such as INSTRUCOES.md in a repo) plus a replicable setup checklist covering folders, connectors, and skills added via Customize.

What is the character limit for Claude project instructions?

Anthropic has not published a limit for project instructions; the only official limit is 3,000 characters for organization instructions. Practical guidance is to target 1,500-2,500 characters for Chat Projects because instructions consume context window space.

When should a rule become a skill instead of an instruction?

A rule becomes a skill when it is a repeatable process for one task with steps, exceeds roughly 15 lines, is used across multiple projects, or runs only occasionally. Instructions load in every task, so rarely-triggered processes waste context window.