AGENT SKILL

Validation Engine

Checks outputs from other skills for consistency, correctness, and adherence to AGENTS.md rules

License 许可
MIT
最近更新
2026-06-22
Stars
★ 1
Forks
0

01 前言 Front Matter

name
Validation Engine
tags
[ "validation", "quality", "review" ]
author
ForgeWeave Core
version
1.0.0
skill_id
validation-engine
description
Checks outputs from other skills for consistency, correctness, and adherence to AGENTS.md rules
tui_compatibility
[ "opencode", "claude", "gemini", "qwen" ]

02 资源文件 2

03 SKILL.md 正文

Validation Engine

Purpose

Systematically inspect outputs from any skill or agent for correctness, consistency, completeness, and compliance with project rules defined in AGENTS.md. Acts as a quality gate before outputs are presented to the user or used by downstream skills.

When to Use

  • After any skill produces output that needs quality assurance
  • Before presenting research results to the user
  • When cross-checking claims across multiple sources
  • After code generation, before presenting the code

When Not to Use

  • The output is trivial and doesn't need validation
  • The user is iterating quickly and doesn't want validation overhead
  • The output is an intermediate step that will be further processed

Inputs

Input Type Required Description
output string Yes The content to validate
type enum Yes research, code, documentation, plan
rules array No Specific rules to enforce (defaults from AGENTS.md)

Expected Outputs

Output Description
Validation report Pass/fail per rule, with details
Issues found List of problems with severity and location
Suggestions How to fix each issue

Exact Workflow Steps

  1. Load validation rules from AGENTS.md and the specific skill's requirements
  2. For research: check every claim has a source URL, flag contradictions, check no blog sources
  3. For code: check syntax, check against project conventions, verify tests pass
  4. For documentation: check completeness, accuracy, clarity
  5. For plans: check steps are ordered, dependencies identified, risks documented
  6. Compile validation report with pass/fail per rule

Required Checks

  • [ ] All required fields are present in the output
  • [ ] No contradictions across the output
  • [ ] Output follows applicable rules from AGENTS.md
  • [ ] All claims are traceable to sources

Failure Modes

Failure Condition Response
Validation rules are ambiguous Ask for clarification
Output fails critical rules Reject and report specific failures
Output fails minor rules Accept with warnings

References

Reference Path
AGENTS.md ./AGENTS.md
Internal validator component Used by the deep-research skill pipeline

04 同仓库更多 Skills