
nixtla-universal-validator
by intent-solutions-io
Claude Code plugin concepts for Nixtla - Generate TimeGPT pipelines, model benchmarks, and FastAPI services from natural language
SKILL.md
name: nixtla-universal-validator description: "Validate Nixtla skills and plugins with deterministic evidence bundles and strict schema gates. Use when auditing changes or enforcing compliance. Trigger with 'run validation' or 'audit validators'." allowed-tools: "Read,Write,Bash(python:),Bash(bash:),Bash(pytest:*)" version: "1.0.0" author: "Jeremy Longshore jeremy@intentsolutions.io" license: MIT
Nixtla Universal Validator
Purpose
Produce deterministic, reviewable validation evidence (reports + JSON + logs) for a repo, plugin, or skill.
Overview
This skill combines two layers:
- A multi-phase subagent workflow (for human-readable analysis + reconciliation)
- A deterministic validator runner (for ground-truth logs and machine-readable summaries)
Validation runs as a pipeline with deterministic gates:
- Discover what changed and what should be validated
- Validate schemas/structure (skills + plugins) using canonical repo validators
- Run behavioral checks (tests) when requested
- Reconcile results into a single evidence bundle with pass/fail and next actions
This pattern generalizes beyond Nixtla by swapping the check catalog (a list of commands + expected artifacts).
Prerequisites
- Python 3.11+
- Repo validators available:
004-scripts/validate_skills_v2.py004-scripts/validate-all-plugins.sh
- Optional for plugin validation:
jq
Instructions
Step 1: Create a run directory
Use the built-in runner to create a timestamped evidence bundle under reports/<project>/<timestamp>/.
Step 2: Pick a target scope
Choose one:
- Repo root: validate everything
- A plugin folder:
005-plugins/<plugin> - A skill folder:
.claude/skills/<skill>or003-skills/.claude/skills/<skill>
Step 3: Run the deterministic validator suite
python {baseDir}/scripts/run_validator_suite.py \
--target . \
--project nixtla \
--out reports/nixtla
List built-in profiles:
python {baseDir}/scripts/run_validator_suite.py \
--list-profiles \
--target . \
--project nixtla \
--out reports/nixtla
To validate a single plugin:
python {baseDir}/scripts/run_validator_suite.py \
--target 005-plugins/nixtla-baseline-lab \
--project nixtla-baseline-lab \
--out reports/nixtla-baseline-lab
Step 4: (Optional) Include tests
python {baseDir}/scripts/run_validator_suite.py \
--target . \
--project nixtla \
--out reports/nixtla \
--run-tests
Step 4b: (Optional) Run an enterprise profile
python {baseDir}/scripts/run_validator_suite.py \
--target . \
--project nixtla \
--out reports/nixtla \
--profile enterprise \
--fail-on-warn \
--run-tests
Step 5: (Optional) Use the multi-phase subagent workflow
Run phases in order using the prompts in {baseDir}/agents/ and procedures in {baseDir}/references/.
Each phase must write a report file under the run directory and return strict JSON per the phase contract.
Output
Each run creates a timestamped evidence bundle:
reports/<project>/<timestamp>/summary.jsonreports/<project>/<timestamp>/report.mdreports/<project>/<timestamp>/checks/*.log
Error Handling
-
Error: Validator command not found
Solution: Confirm repo scripts exist and run from the repo root. -
Error: Plugin validation fails due to
jq
Solution: Installjqor run only skill validation. -
Error: Tests fail after schema passes
Solution: Treat this as a behavioral regression; fix tests or code, then re-run.
Examples
Common validations:
# Strict schema/structure gates
python 004-scripts/validate_skills_v2.py --fail-on-warn
bash 004-scripts/validate-all-plugins.sh .
Generate an evidence bundle (profile-driven):
# Generate a single evidence bundle for a PR
python {baseDir}/scripts/run_validator_suite.py \
--target . \
--project pr-1234 \
--out reports/pr-1234 \
--run-tests
Resources
- Subagent orchestration pattern:
000-docs/000a-planned-skills/templates/verification-pipeline/README.md - Canonical skills validator:
004-scripts/validate_skills_v2.py - Canonical plugin validator:
004-scripts/validate-all-plugins.sh - Subagent prompts:
{baseDir}/agents/ - Phase procedures:
{baseDir}/references/
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon