Agent Mentor Learn

Glossary

49 terms from “Claude Code Skills: Build Your Own AI Workflows.” Hover the first occurrence in the lesson for its definition.

TermDefinitionSource
SkillA directory plus a SKILL.md file: a reusable package of workflow instructions that tells Claude how to do one specific task.Anthropic engineering blog: Equipping agents for the real world with Agent Skills
SKILL.mdThe core file of a Skill, containing two parts: YAML frontmatter (metadata) and Markdown instructions (the directions Claude follows).Claude Code official docs: Extend Claude Code with skills
YAML frontmatterThe metadata block at the top of SKILL.md, wrapped in triple-dash (---) markers, holding fields like name and description.Claude Code official docs: Extend Claude Code with skills
nameA required frontmatter field that defines the Skill's unique identifier; it becomes the command and must use lowercase letters, digits, and hyphens.Anthropic engineering blog: Equipping agents for the real world with Agent Skills
descriptionA required frontmatter field that tells Claude what the Skill does and when to use it. It is the sole basis for automatic triggering.Building skills for Claude, hands-on: YAML frontmatter and testing
automatic triggeringWhen what the user says matches a Skill's description, Claude loads and runs that Skill on its own, with no manual command.Building skills for Claude, hands-on: YAML frontmatter and testing
manual invocationCalling a Skill directly with /skill-name, which forces Claude to load it without relying on the description matching.Claude Code official docs: Extend Claude Code with skills
Markdown instructionsEverything in SKILL.md after the frontmatter: the concrete, step-by-step directions Claude follows, written in Markdown.Claude Code official docs: Extend Claude Code with skills
progressive disclosureGiving Claude only what it needs right now instead of dumping every detail at once, so the Skill stays short and loads fast.Anthropic platform docs: Agent Skills overview
supporting filesFiles in a Skill directory other than SKILL.md (checklists, config, scripts) that are loaded on demand and don't consume initial context.Anthropic platform docs: Agent Skills overview
personal skillsSkills stored in ~/.claude/skills/, available only to you but usable across every project. Good for personal habits and general-purpose tools.Teach Claude Code your workflow: a hands-on guide to custom skills
project skillsSkills stored in a project's .claude/skills/ directory, committed to Git so teammates get them automatically when they clone the repo.Teach Claude Code your workflow: a hands-on guide to custom skills
RepetitionOne of the three traits that make a scenario suit a Skill: you do the same thing daily, weekly, sometimes hourly.Claude Code official docs: Extend Claude Code with skills
Fixed stepsOne of the three traits that make a scenario suit a Skill: the process can be written as clear steps rather than changing every time.Claude Code official docs: Extend Claude Code with skills
VerifiableOne of the three traits that make a scenario suit a Skill: you can tell whether Claude got the result right.Claude Code official docs: Extend Claude Code with skills
description formulaThe guideline for writing an effective description: what it does + when to use it + key capabilities.Building skills for Claude, hands-on: YAML frontmatter and testing
trigger keywordsThe words in a description that a user might plausibly say; Claude matches on these to decide whether to auto-load the Skill.Building skills for Claude, hands-on: YAML frontmatter and testing
Input FormatThe instructions section that spells out what kind of input the Skill accepts, including shape, source, and a concrete example.Claude Code official docs: Extend Claude Code with skills
Processing StepsThe ordered, concrete operations the Skill runs; the heart of the instructions, where each step should be specific enough to be unambiguous.Claude Code official docs: Extend Claude Code with skills
Output FormatThe instructions section defining what the result should look like, including structure, symbols, and a worked example.Claude Code official docs: Extend Claude Code with skills
edge casesUnusual or abnormal inputs (empty input, malformed input, over-long content) whose handling must be spelled out explicitly in the instructions.Claude Code official docs: Extend Claude Code with skills
direct invocationThe simplest test: call the Skill with /skill-name and watch whether the output matches expectations.Claude Code official docs: Extend Claude Code with skills
test casesPrepared input-and-expected-output pairs used to test a Skill systematically before relying on it.Claude Code official docs: Extend Claude Code with skills
iteration loopThe test-find-fix-retest cycle a Skill goes through; the path from 'it runs' to 'it runs well.'Claude Code skills: .NET workflows and reusable prompts
A/B itRunning the same task with and without the Skill and timing both, to prove whether the Skill is actually faster and more consistent.Claude Code skills: .NET workflows and reusable prompts
checklist patternOrganizing a complex task as a series of explicit check items, each of which can independently pass or fail.Claude Code official docs: Extend Claude Code with skills
multi-step workflowA Skill that runs several sequential steps to finish a complex task, where one step's output may feed the next.Claude Code official docs: Extend Claude Code with skills
Grading the outputSplitting a Skill's findings by severity (passed, worth a look, must fix) so the reader knows what to act on first.Claude Code official docs: Extend Claude Code with skills
version controlManaging project skills with Git so you can view history, roll back, and compare versions, the same way you manage code.Claude Code skills: .NET workflows and reusable prompts
Reviewing a SkillReviewing a new or changed Skill the way you review code: check the description, the instructions, and the tests before merging.Claude skills as self-documenting runbooks
naming conventionA team-wide, agreed-on style for Skill names: hyphens, verb-noun or noun-verb order, two or three words.Claude Code skills: .NET workflows and reusable prompts
prune on a scheduleReviewing all Skills once a quarter and deleting or improving the ones that barely get used, so an unused pile doesn't crowd Claude's choices.Claude Code skills: .NET workflows and reusable prompts
announce changesAfter changing an existing Skill, post the change and its impact in the team channel so people don't operate the new Skill on old assumptions.Claude Code official docs: Extend Claude Code with skills
Composing SkillsChaining several single-purpose Skills to handle one larger task, or referencing one Skill from inside another.Anthropic engineering blog: Equipping agents for the real world with Agent Skills
allowed-toolsAn optional frontmatter field that restricts a Skill to a specified set of tools, drawing a permission boundary for security-sensitive cases.Claude Skills deep dive (a first-principles view)
disable-model-invocationAn optional frontmatter field that, when set, stops Claude from auto-loading the Skill, leaving manual invocation only, to avoid accidental triggers.Claude Skills deep dive (a first-principles view)
modelAn optional frontmatter field that pins a specific model for the Skill, for example a stronger reasoning model when the task needs it.Claude Skills deep dive (a first-principles view)
MCP serverA Model Context Protocol server that supplies tools (database access, API calls) to Claude; a Skill supplies the workflow knowledge for using them.The complete guide to building skills for Claude (PDF)
Community SkillsSkills built and shared by other developers, findable on GitHub and elsewhere.Claude Code official docs: Extend Claude Code with skills
Documenting SkillsListing every project skill, with usage and examples, in the project README or .claude/README.md.Claude skills as self-documenting runbooks
single responsibilityEach Skill should do one thing and do it well, rather than trying to solve several unrelated problems in one Skill.Anthropic engineering blog: Equipping agents for the real world with Agent Skills
contextWhen Claude loads a Skill, the SKILL.md content occupies the context window; the longer the file, the more it consumes.Anthropic platform docs: Agent Skills overview
loaded on demandSupporting-file content is read only when Claude actually needs it, so it doesn't occupy context at the Skill's initial load.Anthropic platform docs: Agent Skills overview
false positivesWhen a Skill flags normal code or content as a problem; fix it by adding a 'needs confirmation' category or tuning the criteria.Claude Code official docs: Extend Claude Code with skills
MissesWhen a Skill fails to catch a problem it should have found; fix it by adding check rules or adjusting the judgment logic.Claude Code official docs: Extend Claude Code with skills
loading-failureWhen a Skill won't auto-load or a manual call fails, the diagnostic process of checking file path, frontmatter format, and name spelling in that order.Claude Code official docs: Extend Claude Code with skills
ExamplesProviding input/output examples so Claude understands the expected behavior; a worked example does more than prose description.Claude Code official docs: Extend Claude Code with skills
feature branchCreating or changing an experimental Skill on a Git branch and merging to main only once it earns it, so the team isn't disrupted.Claude Code official docs: Extend Claude Code with skills
workflow knowledgeKnowledge about how to organize steps, when to use which tool, and how to judge the result; the core value a Skill supplies.The complete guide to building skills for Claude (PDF)