| name | skill-validator |
| description | Validating agent skills against the agentskills.io specification and generating compliance reports. |
| metadata | {"triggers":"skill-audit, validate-skill, check-compliance"} |
Skill Validator Standards
Purpose
This skill provides the logic required to audit and enforce compliance for ANY agent skill within the serenity-android project according to agentskills.io standards.
The "Strict" Protocol (LLM-Driven Audit)
As a strict validator, I assume every skill is in violation until proven otherwise. I perform an exhaustive scan via the following protocols:
1. Mandatory/Optional Metadata Audit
- Mandatory:
name: 1-64 chars, lowercase alphanumeric + hyphens only. No consecutive hyphens, no start/end hyphens. MUST match parent folder name.description: 1-1024 chars, non-empty.
- Optional:
license: Short description or reference to file (e.g., LICENSE.txt).compatibility: 1-500 chars (if present).metadata: Arbitrary key-value map (string:string).allowed-tools: Space-delimited list of pre-approved tools (experimental).
- Token Limit: Metadata MUST be < 100 tokens.
2. Structural & Reference Integrity
- Mandatory:
tmp_SKILL.md present.
- Assets/References:
assets/ MUST ONLY contain operational files (templates, resources).references/ MUST contain structured markdown or readable text.
- Reference Management:
- All file references inside
tmp_SKILL.md MUST use relative paths (e.g., ./references/doc.md).
- Constraint: No parent traversal (
../). Chain depth MUST be 1 level.
- Audit: Identify absolute paths or invalid relative links as violations.
3. Progressive Disclosure & Token Economy
- Size:
tmp_SKILL.md MUST be < 500 lines.
- Heuristic for References:
- Flag long code blocks (> 50 lines), large tables, or deep technical manuals.
- Suggestion: Propose moving flagged content to
references/ to preserve context economy.
- Token Limit: Instructions (body) MUST be < 5000 tokens.
Reporting & Output
Upon completion, I generate an HTML report with a table of results that includes both compliance findings (violations) and structural strengths (best practices/adherence). The report uses the template at assets/audit_report_template.html.
Operational Workflow
- Exhaustive Scan: The agent reads the target
tmp_SKILL.md, references/, and assets/.
- Strict Reporting: The agent generates the HTML report using the template and writes it to
${PROJECT_ROOT}/skill_audit.html.
- Compliance Level: Strict (Pass/Fail).
Compliance
- For details, see
https://agentskills.io/specification.
- Variable Resolution Policy:
${PROJECT_ROOT} refers to the absolute path of the project root. Defaults to current working directory if not set.
