Term Keeper

A reusable Codex skill for maintaining a project glossary/design dictionary.

中文说明 · Agent-readable version

Do you feel that an agent forgets key definitions after context compaction? Do different agents interpret the same term differently from you? Do module names, role boundaries, dashboard metrics, or pipeline terms keep drifting across conversations? This skill gives your project a durable vocabulary layer: Markdown for agents, HTML tables for human review, and confirmation gates before definitions are written.

What It Solves

• Keeps durable project concepts out of transient chat history.
• Prevents agents from treating guesses as your definitions.
• Preserves business wording, role boundaries, deprecated meanings, and reporting semantics.
• Maintains both an agent-readable Markdown glossary and a human-review HTML table.
• Works across projects without hard-coded repository paths.

How It Works

1. The user explicitly defines, corrects, renames, or deprecates a durable concept.
2. The agent drafts a confirmation table.
3. The user confirms or corrects the wording.
4. The agent updates both docs/glossary.md and docs/glossary.html.

Best Fit

Use this skill when your project has terms such as:

• Domain vocabulary and business concepts.
• Module names and ownership meanings.
• Agent, role, or workflow boundaries.
• Dashboard metrics and reporting definitions.
• Pipeline states, handoffs, and artifact semantics.
• Deprecated or renamed concepts that agents must not reuse incorrectly.

Skill Package

The actual skill package is:

skill/term-keeper/
  SKILL.md
  agents/openai.yaml
  examples/demo-glossary.html
  references/dictionary-config-template.md
  references/pressure-test-scenarios.md
  scripts/render_glossary_html.py
  scripts/check_dictionary_sync.py

Quick Start

Install the skill into your Codex skills directory, then say something like:

Please use term-keeper to record this project term:
"Agent Router" means the component that classifies requests and hands them off to the correct agent. It is not a scheduler.

The agent should first show a confirmation table. Only after you approve it should the glossary be updated.

Default Output Files

When no project-specific dictionary exists, the skill defaults to:

docs/glossary.md
docs/glossary.html

The Markdown file is the agent-friendly source. The HTML file is a table-first review page for humans.

Optional Helpers

The bundled Python scripts are optional:

• skill/term-keeper/scripts/render_glossary_html.py: renders default Markdown entries into an HTML review table.
• skill/term-keeper/scripts/check_dictionary_sync.py: checks whether Markdown and HTML have synchronized terms and selected metadata.

Demo

GitHub Pages display page:

https://springrain-i.github.io/term-keeper/

The live page works after GitHub Pages is enabled for main / docs.

Open the bundled HTML demo locally:

skill/term-keeper/examples/demo-glossary.html

It shows the expected table-first review experience with Chinese table headers, last modified date, trigger frequency, confirmation status, and deprecated meanings.

Screenshots

demo-glossary-table.png is captured from examples/demo-glossary.html. confirmation-table-flow.svg explains the confirmation-first workflow.

More

• See INSTALLATION.md for setup.
• See EXAMPLES.md for practical prompts.
• See TESTING.md for minimal verification.
• See README.zh-CN.md for Chinese documentation.
• See README.agent.md for agent-facing instructions.