CtrlK
BlogDocsLog inGet started
Tessl Logo

ai-quality

AI quality: testing strategy, assisted testing, code review, safety, content detection, docs, workflow automation. Topics: ai-assisted-testing, ai-code-review, ai-content-detection, ai-documentation, ai-safety, ai-testing-strategy, ai-workflow-automation, code-review, llm-evaluation.

SKILL.md
Quality
Evals
Security

ai-quality

Router for AI/code quality work. Resolve ONE topic, Read ONLY its playbook from routes:, then execute. [DOC]

When to use

Any request to assess, test, review, evaluate, or harden code or AI output: test plans, AI-assisted/LLM-judge evaluation, human or AI code review, model safety, AI-vs-human content detection, quality docs, or QA automation. [DOC] Not for: building features, prompt authoring, deployment — route elsewhere.

Inputs / Outputs

  • In: topic (required; infer from intent, ask only if two routes tie), depth (quick=essentials, deep=exhaustive with verification per step), plus the artifact under review (code, model output, spec, pipeline). [CONFIG]
  • Out: the playbook's deliverable, every non-obvious claim carrying one Alfa-set tag ([CÓDIGO] [CONFIG] [DOC] [INFERENCIA] [SUPUESTO]). [DOC]

Routing

  1. Map intent → exactly one topic (enums are fixed; never invent one). [CÓDIGO]
  2. Disambiguate near-collisions: AI judges/scores output → llm-evaluation; AI writes/runs tests → ai-assisted-testing; AI reviews a diff → ai-code-review; human/tool review of a diff → code-review. [INFERENCIA]
  3. Read that single playbook. Never preload the cluster — token waste and cross-topic bleed are the failure modes. [INFERENCIA]
  4. Apply the spine: Discover → Analyze → Execute → Validate. [DOC]

Routing/gate contracts live in assets/ (routing-matrix.json, quality-rubric.json); see assets/README.md. [CONFIG]

Validation gate (before "done")

  • Resolved topic ∈ enum; exactly one playbook Read. [CÓDIGO]
  • Quality gates honored: constitution v6.0.0, evidence tags, script-first. [CONFIG]
  • One tag family throughout (Alfa set here); no [CÓDIGO] without an in-repo referent — else downgrade to [SUPUESTO]. [DOC]
  • deep ran verification at each step; quick stayed at essentials.

Anti-patterns & self-correction

Avoid: loading >1 route; answering from this router without a playbook; guessing an ambiguous topic; untagged prose; mixed tag families. [SUPUESTO] On any of these → stop, re-resolve topic via the disambiguation rules, re-read the one playbook. [INFERENCIA]

Repository
JaviMontano/jm-adk-beta
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.