knowledge-base
Use this skill whenever you work in this repository — before answering questions about how anything in the project works, before writing or modifying code…
Skill by YamiHilik
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/yamihilik-knowledge-base
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: writing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/knowledge-base
What's inside
- Skill name
knowledge-base- Description
- Use this skill whenever you work in this repository — before answering questions about how anything in the project works, before writing or modifying code, and after finishing any non-trivial change. The project keeps its source of truth in `.docs/knowledge/` (a wiki-style knowledge base with `README.md` as the map) and `.docs/knowledge/adr/` (architecture decision records). Trigger this skill on phrases like "how does X work", "where is Y", "add feature", "refactor", "fix bug", "what's the convention for", or any request that touches the codebase's behavior, architecture, or conventions. Also trigger after making changes — the knowledge base must stay in sync with the code, and new architectural decisions must be captured as ADRs. Do NOT skip this skill assuming you can figure things out from the code alone; the knowledge base captures intent and decisions that the code does not.
- License
- Apache-2.0
- Instructions
- 158 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Project Knowledge Base
This project maintains a living knowledge base. Your job is to **read it before acting**, **keep it in sync** when you change things, and **record architectural decisions** as ADRs.
## Where things live
```
.docs/
└── adr/
├── <YYYY-MM-DD>-<short-title>.md
├── ...
└── outdated/ ← superseded ADRs go here
└── knowledge/
├── README.md ← map of the entire knowledge base
├── <topic>.md ← one document per topic
├── ...
```
- `.docs/knowledge/README.md` is the **entry point**. It links to every document. Read it first.
- Each topic (design system, components, routing, data models, etc.) is its own `.md` file.
- Documents reference each other with wiki-style `[[link]]` syntax — the link text matches the filename without extension.
- `adr/` holds Architecture Decision Records. **ADRs are immutable.**
## When to consult the knowledge base
**Before answering or coding, always read `.docs/knowledge/README.md` first.** Then read the specific documents relevant to the task. The README is small and gives you the map; skipping it means you'll miss context that's already been written down.
Concretely, consult before:
- Answering questions about how the project works ("how does the kanban board behave?", "what's the data model for tasks?")
- Implementing a feature — read the relevant component, design system, and routing docs
- Fixing a bug — read the doc for the affected area to understand intended behavior
- Refactoring — check ADRs in `.docs/adr/` for decisions that constrain your options
If a document is missing for the area you need, that itself is a signal: either the area is undocumented (note it, and consider adding a doc) or the README is out of sync.
## When to update the knowledge base
After making any change that affects what's documented, update the knowledge base in the same task. Treat the docs the way you treat tests — they're part of "done".
Update when you:
- Change behavior, structure, or conventions described in an existing document → edit that document
- Add a new feature, component, or subsystem that no existing document covers → create a new document and add a link in `README.md`
- Rename or delete something referenced in the docs → fix every reference (search for the old name across `.docs/knowledge/`)
- Make a non-trivial architectural choice → write an ADR (see below)
Do not let docs drift. If the code says one thing and the doc says another, the doc is a bug.
### What belongs in knowledge base docs vs ADRs
База знаний и ADR описывают **систему**. Всё, что описывает сам документ или
историю его правки, в них не попадает.
Не добавлять:
- обоснования собственных правок: «здесь не используется термин X», «переименовано, потому что»,
«раздел переписан», «уточнено по итогам ревью»;
- повторное изложение того, что уже сказано в контракте, таблице ответственности или диаграмме
этого же документа — ради связности или подстраховки;
- определения понятий, очевидных из контракта.
**Проверка перед добавлением абзаца или раздела:** проверяется ли это утверждение по коду
и поведению системы? Если оно проверяется только по истории правки документа или по переписке
о нём — это не содержание документа.
**Что и почему изменено в документе** — это сообщение коммита.
**Knowledge base doc** — только актуальное описание системы:
- как работает фича / поток / компонент прямо сейчас
- бизнес-правила и их условия
- нетривиальные ограничения или хрупкие места ("GitLab присылает X в ситуации Y — код делает Z")
**ADR** — всё остальное про "почему":
- история расследований и отладки
- хронология багов и инцидентов
- альтернативы, которые рассматривались
- причина, по которой текущий код написан именно так
Конкретный тест: если удалить секцию из knowledge base doc, станет ли непонятно как работает система? Если нет — это ADR, не knowledge base.
### Adding a new document
1. Pick a clear, lowercase, hyphenated filename matching the language already used in the knowledge base (e.g., `card-component.md`, `карточка-дела.md` — match the existing convention).
2. Create the file in `.docs/knowledge/`.
3. Add a bullet for it in the appropriate section of `.docs/knowledge/README.md` using the same `[[link]] — short description` format as the surrounding entries.
4. If it fits no existing section, add a new section to the README rather than burying it.
### Editing an existing document
- Keep the document focused on its topic. If the change is really about a different topic, edit that other document (or create a new one) instead of stuffing unrelated content in.
- Preserve the existing structure and tone. The knowledge base is meant to be browsable; consistent shape matters.
- Update cross-references if you rename headings other documents link to.
## Architecture Decision Records (ADRs)
ADRs live in `.docs/adr/` and capture *why* a decision was made, not just *what* the code does.
### Hard rules
- **ADRs are never edited.** Once written, the file stays as it is — it's a historical record.
- **To change a decision, write a new ADR** that supersedes the old one, then move the old one to `.docs/adr/outdated/`. Do not delete it.
- **One decision per ADR.** Keep the scope minimal — a single, narrow concern. If you find yourself writing about two things, split them.
- **Filename format:** `<YYYY-MM-DD>-<short-meaningful-title>.md` (e.g., `2026-05-10-use-postgres-jsonb-for-task-metadata.md`).
### When to write an ADR
Write one when you make a choice that:
- Constrains future work (technology choice, schema shape, API contract, library selection)
- Has reasonable alternatives that someone might later wonder about
- You'd want a future developer (including yourself in six months) to understand the reasoning behind
Don't write ADRs for trivial choices (variable names, file layout within a feature, fixing typos). Reserve them for decisions worth re-explaining.
### ADR template
Keep it short. A good ADR is half a page, not three pages.From YamiHilik/SDLCRoom, Apache-2.0.
More from YamiHilik/SDLCRoom
- 0.524score
simplify
YamiHilik
Simplifies code for clarity.
SkillBFreshness: BClaude Code +9 - 0.521score
domain-modelling
YamiHilik
Строй и уточняй доменную модель проекта.
SkillBFreshness: BClaude Code +9 - 0.517score
commit-message
YamiHilik
ALWAYS load this skill as the final step after making any code changes (edit, create, or delete source files).
SkillBFreshness: BClaude Code +9 - 0.516score
estimate-scrum-points
YamiHilik
Используй этот скилл для оценки задач в стори-поинтах SCRUM.
SkillBFreshness: BClaude Code +9 - 0.513score
do-task-one
YamiHilik
Используй этот скилл, когда пользователь дает задание реализовать задачу (программно).
SkillBFreshness: BClaude Code +9 - 0.512score
send-push-to-user
YamiHilik
Отправка пуша пользователю.
SkillBFreshness: BClaude Code +9 - 0.504score
make-functional-spec
YamiHilik
This skill should be used when the user asks to "create a functional spec", "write a system design", "describe the architecture of a feature", "make a…
SkillBFreshness: BClaude Code +9 - 0.504score
make-plan
YamiHilik
Use this skill when the user asks to "create an implementation plan", "make a plan", "plan the work", "make implementation-plan.md", "создать план…
SkillBFreshness: BClaude Code +9 - 0.504score
make-research
YamiHilik
Use this skill when the user asks to "do research", "research the approach", "compare alternatives", "choose a library", "make a research.md", "исследовать…
SkillBFreshness: BClaude Code +9 - 0.504score
make-tasks
YamiHilik
Use this skill when the user asks to "create tasks", "break down the plan into tasks", "generate tasks", "make tasks", "create issues", "make-tasks", "создать…
SkillBFreshness: BClaude Code +9 - 0.504score
make-ui-spec
YamiHilik
This skill should be used when the user asks to "create a UI spec", "write a UI/UX spec", "describe the interface", "make a ui-spec.md", "document the…
SkillBFreshness: BClaude Code +9 - 0.504score
review-plan
YamiHilik
Use this skill when the user asks to "review the plan", "validate implementation plan", "check the plan", "review implementation-plan.md", "проверить план"…
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 22, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/YamiHilik/SDLCRoom
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from YamiHilik
- 0.524score
simplify
YamiHilik
Simplifies code for clarity.
SkillBFreshness: BClaude Code +9 - 0.521score
domain-modelling
YamiHilik
Строй и уточняй доменную модель проекта.
SkillBFreshness: BClaude Code +9 - 0.517score
commit-message
YamiHilik
ALWAYS load this skill as the final step after making any code changes (edit, create, or delete source files).
SkillBFreshness: BClaude Code +9 - 0.516score
estimate-scrum-points
YamiHilik
Используй этот скилл для оценки задач в стори-поинтах SCRUM.
SkillBFreshness: BClaude Code +9 - 0.513score
do-task-one
YamiHilik
Используй этот скилл, когда пользователь дает задание реализовать задачу (программно).
SkillBFreshness: BClaude Code +9 - 0.512score
send-push-to-user
YamiHilik
Отправка пуша пользователю.
SkillBFreshness: BClaude Code +9
Cite this page
knowledge-base skill. Agent Mart. https://agentmart.dev/skills/yamihilik-knowledge-base (last updated 2026-10-24)