≈1.6k tokens on demand. Your agent reads this file only when SKILL.md points to it.
Anthropic Skill Standards Reference
Source: "The Complete Guide to Building Skills for Claude" (Anthropic, 2025)
Reference used by Canon in the ASSESS phase to evaluate SKILL.md compliance with the official specification.
1. Standard Overview
| Field |
Value |
| Standard Name |
Anthropic Agent Skill Specification |
| Publisher |
Anthropic |
| Version |
2025 (The Complete Guide to Building Skills for Claude) |
| Scope |
Skill design, structure, and distribution across Claude Code / Claude.ai / API |
| Category |
AI Agent Quality |
| Canon Category Code |
SKILL |
2. Requirements Matrix
2.1 Structural Requirements (STR)
| ID |
Requirement |
Level |
Evidence |
| STR-01 |
A SKILL.md file exists (case-sensitive) |
CRITICAL |
File existence check |
| STR-02 |
YAML frontmatter is enclosed by --- delimiters |
CRITICAL |
YAML parse validation |
| STR-03 |
The name field is kebab-case |
HIGH |
Regex: ^[a-z0-9]+(-[a-z0-9]+)*$ |
| STR-04 |
name matches the folder name |
HIGH |
String comparison |
| STR-05 |
name does not contain "claude" / "anthropic" |
CRITICAL |
String search |
| STR-06 |
A description field is present |
CRITICAL |
Field presence check |
| STR-07 |
description is 1024 characters or fewer |
HIGH |
Character count |
| STR-08 |
description contains no XML tags (< >) |
CRITICAL |
Character search |
| STR-09 |
The skill folder contains no README.md |
MEDIUM |
File absence check |
| STR-10 |
compatibility field is 500 characters or fewer (when used) |
LOW |
Character count |
2.2 Description Quality Requirements (DSC)
| ID |
Requirement |
Level |
Evidence |
| DSC-01 |
WHAT (what it does) is described |
HIGH |
Semantic analysis |
| DSC-02 |
WHEN (when to use it / trigger conditions) is described |
HIGH |
Trigger phrase presence |
| DSC-03 |
Concrete tasks/phrases are included |
MEDIUM |
Actionable keyword detection |
| DSC-04 |
Relevant file types are mentioned when applicable |
LOW |
Context-dependent |
| DSC-05 |
Not vague (excludes generic wording like "Helps with projects") |
HIGH |
Anti-pattern matching |
2.3 Instruction Quality Requirements (INS)
| ID |
Requirement |
Level |
Evidence |
| INS-01 |
A step-by-step structure is present |
HIGH |
Heading/list structure |
| INS-02 |
Instructions are concrete and actionable |
HIGH |
Imperative verb presence |
| INS-03 |
An Examples section is present |
MEDIUM |
Section heading detection |
| INS-04 |
Troubleshooting / error handling is described |
MEDIUM |
Section heading detection |
| INS-05 |
Links from reference/ are appropriate |
MEDIUM |
Link validity check |
| INS-06 |
Critical instructions are placed near the top of the document |
LOW |
Position analysis |
2.4 Progressive Disclosure Requirements (PD)
| ID |
Requirement |
Level |
Evidence |
| PD-01 |
1st level: frontmatter is minimal yet sufficient for triggering decisions |
HIGH |
Frontmatter content analysis |
| PD-02 |
2nd level: SKILL.md body focuses on core instructions |
MEDIUM |
Word count ≤ 5000 recommended |
| PD-03 |
3rd level: details are separated into reference/ |
MEDIUM |
Directory structure check |
| PD-04 |
Reference links from SKILL.md are explicit |
LOW |
Reference link presence |
2.5 Composability Requirements (CMP)
| ID |
Requirement |
Level |
Evidence |
| CMP-01 |
Designed to coexist with other skills |
LOW |
Exclusive capability claims absence |
| CMP-02 |
Environment dependencies are recorded in the compatibility field |
LOW |
Field content check |
3. Troubleshooting Compliance Check
In the ASSESS phase, evaluate skill compliance against the following 6 categories:
| Category |
Check |
Compliant Criteria |
| Upload Failure |
SKILL.md naming, YAML format, name format |
STR-01 through STR-05 all PASS |
| Undertriggering |
Description quality |
DSC-01 through DSC-05 all PASS |
| Overtriggering |
Scope clarity, negative triggers |
DSC-02 + presence of scope-limiting statements |
| Instructions Not Followed |
Instruction structure quality |
Key items in INS-01 through INS-06 PASS |
| MCP Connection Issues |
Explicit MCP dependencies and error handling |
INS-04 + MCP-related troubleshooting |
| Large Context Issues |
Progressive Disclosure implementation |
PD-01 through PD-04 all PASS |
4. Compliance Level Determination
Overall Determination Criteria
| Level |
Criteria |
Action |
| Compliant |
All CRITICAL PASS + 80%+ of HIGH PASS |
Document and maintain |
| Partial |
All CRITICAL PASS + 50-79% of HIGH PASS |
Enhancement recommended |
| Non-compliant |
1+ CRITICAL FAIL |
Remediation required |
Severity Timeline
| Severity |
Timeline |
Examples |
| CRITICAL |
Immediate |
Missing SKILL.md, broken YAML, XML in frontmatter |
| HIGH |
1 sprint |
Vague description, no trigger phrases, no steps |
| MEDIUM |
1 month |
No examples, no troubleshooting, inline-heavy |
| LOW |
Backlog |
Missing compatibility field, no negative triggers |
5. Role Boundary with Gauge
| Aspect |
Canon |
Gauge |
| Target standard |
Anthropic official skill specification + industry standards |
Ecosystem's internal 16-item normalization checklist |
| Evaluation perspective |
Compliance with external standards |
Conformance with internal templates |
| Output |
Compliance report with citations |
PASS/PARTIAL/FAIL with fix snippets |
| Collaboration |
Canon detects official-standard violations → Gauge performs detailed verification against the internal checklist |
Gauge detects structural issues → Canon cross-checks against official standards |
6. Evidence Format
Standard: Anthropic Agent Skill Specification (2025)
Requirement: [ID] [requirement description]
Evidence: [file:line or structural observation]
Status: [Compliant | Partial | Non-compliant]
Finding: [specific observation]
Recommendation: [actionable fix]
Priority: [CRITICAL | HIGH | MEDIUM | LOW]
Remediation Agent: [Sigil | Architect | Gauge]
