By Ruban Shafi
Claude Skills package a proposal shop's format rules, templates and checks into one folder. How the three loading levels work and what must never go in. A skill is a folder holding a SKILL.md file, plus whatever templates and scripts you bundle with it. Here is how the three loading levels work, which field decides whether your house style ever runs, and what a federal contractor must keep out. A skill is a folder on a filesystem. Claude Skills let a proposal shop hand over its format rules, its checklists and its templates once, instead of explaining them in every chat. The folder holds one required file, SKILL.md , plus any templates and scripts you choose to bundle. Claude loads that folder in stages when a request matches what the skill says it is for. So repeatability does not come from a longer prompt. It comes from files that sit on disk until the work calls for them. Shaavir uses Claude in its own proposal work. Every source named here answered on 16 November 2026. What Claude Skills Are Anthropic's documentation describes skills as "reusable, filesystem-based resources that give Claude domain-specific expertise: workflows, context, and best practices". It draws the contrast with a prompt. A prompt is a conversation-level instruction for one task. A skill loads on demand, which means you stop repeating the same guidance across conversations. The open Agent Skills specification (https://agentskills.io/specification) states the minimum. A skill is a directory holding a SKILL.md file, and that file carries YAML frontmatter above its Markdown. Two frontmatter fields are mandatory, name and description . The rest is optional, and scripts/ , references/ and assets/ are conventions rather than rules. That is a short list of requirements. The format is not the hard part. The Three Loading Levels ! Three loading stages of a Claude skill: metadata always loaded, the SKILL.md body on trigger, and bundled templates and scripts only when read (https://mhhifytmrlyksfrjacvi.supabase.co/storage/v1/object/public/blog-images/2026/11/claude-skills-for-proposal-writers-figure.png) Only the name and description load for every skill, so that one field decides whether your house style ever runs. Anthropic's Agent Skills overview (https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) sets out three levels and prices each one. Metadata, meaning the name and description , loads at startup for every installed skill at roughly 100 tokens each. The SKILL.md body loads only when the skill triggers, and the guidance asks authors to keep it under 5,000 tokens. Bundled files cost nothing "until accessed". The documentation calls this progressive disclosure, and it changes where a firm puts its material. Your format specification, your client style sheet and your compliance matrix template are all level three. They can run long. A skill can hold doze
A skill is a folder holding a SKILL.md file, plus whatever templates and scripts you bundle with it. Here is how the three loading levels work, which field decides whether your house style ever runs, and what a federal contractor must keep out.
A skill is a folder on a filesystem. Claude Skills let a proposal shop hand over its format rules, its checklists and its templates once, instead of explaining them in every chat. The folder holds one required file, SKILL.md , plus any templates and scripts you choose to bundle. Claude loads that folder in stages when a request matches what the skill says it is for. So repeatability does not come from a longer prompt. It comes from files that sit on disk until the work calls for them. Shaavir uses Claude in its own proposal work. Every source named here answered on 16 November 2026. What Claude Skills Are Anthropic's documentation describes skills as "reusable, filesystem-based resources that give Claude domain-specific expertise: workflows, context, and best practices". It draws the contrast with a prompt. A prompt is a conversation-level instruction for one task. A skill loads on demand, which means you stop repeating the same guidance across conversations. The open Agent Skills specification (https://agentskills.io/specification) states the minimum. A skill is a directory holding a SKILL.md file, and that file carries YAML frontmatter above its Markdown. Two frontmatter fields are mandatory, name and description . The rest is optional, and scripts/ , references/ and assets/ are conventions rather than rules. That is a short list of requirements. The format is not the hard part. The Three Loading Levels ! Three loading stages of a Claude skill: metadata always loaded, the SKILL.md body on trigger, and bundled templates and scripts only when read (https://mhhifytmrlyksfrjacvi.supabase.co/storage/v1/object/public/blog-images/2026/11/claude-skills-for-proposal-writers-figure.png) Only the name and description load for every skill, so that one field decides whether your house style ever runs. Anthropic's Agent Skills overview (https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) sets out three levels and prices each one. Metadata, meaning the name and description , loads at startup for every installed skill at roughly 100 tokens each. The SKILL.md body loads only when the skill triggers, and the guidance asks authors to keep it under 5,000 tokens. Bundled files cost nothing "until accessed". The documentation calls this progressive disclosure, and it changes where a firm puts its material. Your format specification, your client style sheet and your compliance matrix template are all level three. They can run long. A skill can hold doze
Contact Shaavir · View services · Read the blog · Careers · Case studies · Privacy policy · Terms