A Claude Code starter for running project operations through a few repeatable workflows instead of one-off prompts.
This starter is opinionated about the pattern, not the products:
- Your backlog can be Jira, GitHub Issues, Linear, Asana, or a local JSON file.
- Your published context can be Confluence, a
docs/folder, MkDocs, Obsidian, Notion, or dated markdown files. - Your working memory can stay as a local JSON file until you outgrow it.
.claude/commands/for reusable Claude Code workflowsCLAUDE.mdfor repo-level operating rulesdocs/maintainers/for curated durable contexttemplates/for cross-session memory and reporting templatesdocs/adapters/for swapping tools without changing the overall pattern
Claude works better on real projects when it has four stable context layers:
- Repo guardrails in
CLAUDE.md - Curated technical context in
docs/maintainers/ - Lightweight working memory in
.workspace-temp/context-db.json - Real systems of record for backlog, docs, PRs, and releases
The commands in .claude/commands/ sit on top of those layers and turn them into repeatable workflows like:
- planning and capture:
/standup,/bug,/rfe - memory and learning:
/reflect,/checkpoint,/root-cause,/docs-sync - publishing and sharing:
/weekly-report,/release-notes,/sanitize
- Copy this repo or the parts you want into your project.
- Create
.workspace-temp/context-db.jsonfromtemplates/context-db.template.json. - Customize the command files to match your backlog and publishing targets.
- Add or trim
docs/maintainers/until Claude has the minimum useful durable context.
- Replace placeholder project keys, labels, and URLs in
.claude/commands/ - Decide where weekly reports should land by default
- Decide whether release notes and incident notes should live in
docs/or an external system - Adjust the priority labels to match your team
- Keep
CLAUDE.mdshort and move heavier context intodocs/maintainers/
Before making your setup public, read docs/sanitize-before-sharing.md.