Mary Olowu, a software developer, shifted her approach to using Claude Code from relying on large, repetitive prompts to establishing a project operations layer for ongoing development. She identified that the default interaction pattern—opening the tool, pasting context, explaining a task, and repeating the cycle the next day—fails to sustain context for long-term project needs like backlog triage, bug capture, and status reporting. To address this, she developed a starter repo available at restofstack/claude-project-ops-starter, which utilizes structured files and defined workflows rather than manual session renegotiation.

The implemented structure consists of five core components. A short CLAUDE.md file establishes repository-level guardrails, such as prioritizing existing systems of record over invented state. Maintainer documentation located in a docs/maintainers/ directory provides durable project context, including service boundaries, runbooks, and release notes, ensuring both AI and human collaborators have access to stable information. For working memory, Olowu uses a small local JSON file instead of complex databases, which remains cheap and easy to inspect. She mandates that real systems of record—such as Jira, GitHub Issues, or Linear—must remain the authoritative sources for backlogs and PR data, with the AI tasked only to read and synthesize information from them.

The most significant usability improvement involved creating repeatable commands for common tasks. By defining workflows such as /standup, /bug, /rfe, /reflect, and /weekly-report, she reduced prompt thrash and eliminated the need to treat every interaction as a blank slate. The /standup command, for example, aggregates memory, git state, and backlog data to recommend specific next actions. This setup is particularly effective for managing the complexity of a solo-developer monorepo (a software development strategy where many projects are stored in one repository). Olowu emphasizes that this ops-focused design enables Claude Code to function as a practical interface for maintaining project momentum, filling a gap distinct from feature-specific tools like Spec Kit.

Mary Olowu, a software developer, shifted her approach to using Claude Code from relying on large, repetitive prompts to establishing a project operations layer for ongoing development. She identified that the default interaction pattern—opening the tool, pasting context, explaining a task, and repeating the cycle the next day—fails to sustain context for long-term project needs like backlog triage, bug capture, and status reporting. To address this, she developed a starter repo available at restofstack/claude-project-ops-starter, which utilizes structured files and defined workflows rather than manual session renegotiation.

The implemented structure consists of five core components. A short CLAUDE.md file establishes repository-level guardrails, such as prioritizing existing systems of record over invented state. Maintainer documentation located in a docs/maintainers/ directory provides durable project context, including service boundaries, runbooks, and release notes, ensuring both AI and human collaborators have access to stable information. For working memory, Olowu uses a small local JSON file instead of complex databases, which remains cheap and easy to inspect. She mandates that real systems of record—such as Jira, GitHub Issues, or Linear—must remain the authoritative sources for backlogs and PR data, with the AI tasked only to read and synthesize information from them.

The most significant usability improvement involved creating repeatable commands for common tasks. By defining workflows such as /standup, /bug, /rfe, /reflect, and /weekly-report, she reduced prompt thrash and eliminated the need to treat every interaction as a blank slate. The /standup command, for example, aggregates memory, git state, and backlog data to recommend specific next actions. This setup is particularly effective for managing the complexity of a solo-developer monorepo (a software development strategy where many projects are stored in one repository). Olowu emphasizes that this ops-focused design enables Claude Code to function as a practical interface for maintaining project momentum, filling a gap distinct from feature-specific tools like Spec Kit.