Skip to content
kkkwangwPublic

About

kkkwangw-skills

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Answer Quality Skill

English | 简体中文

answer-quality is an explicitly invoked quality constraint for Codex answers. It reduces unsupported certainty, generic advice, source repetition, false precision, and unnecessary length while preserving the active task and any domain-specific skills.

What It Does

  • decides whether the question requires investigation before answering;
  • prioritizes user artifacts and authoritative sources;
  • separates evidence, assumptions, inference, recommendations, and forecasts;
  • requires actionable recommendations and verification criteria when relevant;
  • handles missing evidence and conflicting sources without filling gaps with plausible prose;
  • keeps simple answers concise and avoids rigid response templates.

It is not a research engine or a substitute for domain expertise. It controls answer quality while other tools and skills provide task-specific capabilities.

Installation

Clone the repository and place or link the answer-quality directory under a Codex skills directory:

git clone https://github.com/kkkwangw/skills.git
cd skills
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
ln -s "$PWD/answer-quality" "${CODEX_HOME:-$HOME/.codex}/skills/answer-quality"

Restart or reload Codex if the skill list is already active.

Usage

The skill is explicit-only. Invoke it with $answer-quality when a question needs stronger evidence, uncertainty handling, or an answer audit.

Use $answer-quality. Investigate why this deployment became slower and separate confirmed evidence from hypotheses.
Use $answer-quality. Compare PostgreSQL and MySQL for this workload, state your assumptions, and recommend one.
Use $answer-quality to audit the following answer for unsupported claims, actionability, and unnecessary verbosity: ...

Simple questions remain simple; explicit invocation does not automatically turn them into research tasks.

Behavior

The skill applies these stages:

  1. identify the user's goal and decisive constraints;
  2. decide whether investigation is necessary;
  3. inspect available evidence and authoritative sources when required;
  4. form an independent, evidence-calibrated conclusion;
  5. provide proportionate action and verification guidance;
  6. remove unsupported claims and unnecessary content.

When required evidence is unavailable, the answer identifies the limitation and the next evidence needed instead of guessing.

Repository Layout

.
├── README.md
├── README.zh-CN.md
├── answer-quality/
│   ├── SKILL.md
│   ├── agents/
│   │   └── openai.yaml
│   └── references/
│       ├── evaluation-cases.md
│       └── quality-rubric.md
└── scripts/
    └── validate.py

SKILL.md contains the runtime workflow. The references are loaded only for systematic answer audits or validation of material skill changes.

Validation

Run the repository validation entry point:

python scripts/validate.py

The validator checks SKILL.md frontmatter, agents/openai.yaml, local Markdown links, expected package contents, ignored artifacts, and git diff --check. In this repository's Codex environment, run it without changing the active environment:

conda run -n codex python scripts/validate.py

For behavioral validation, run the exact prompts in answer-quality/references/evaluation-cases.md in fresh sessions and score the outputs with answer-quality/references/quality-rubric.md.

After committing the release candidate, verify that the worktree and Git archive represent the reviewed files:

python scripts/validate.py --release

Release Practice

Use Git commits and tags to record released versions. Before tagging a release:

  1. run python scripts/validate.py --release on the release commit;
  2. run the applicable behavioral regression cases;
  3. confirm the packaged skill contains no local metadata or unrelated files;
  4. review agents/openai.yaml against SKILL.md;
  5. record material evaluation results in the release or pull request.

About

kkkwangw-skills

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages