This is a highly-opinionated library of philosophy and process for developing software with LLMs, specifically Claude Code.

Copy/clone the contents of out into ~/.claude/commands to install these across all projects.

mkdir -p ~/.claude/commands/terma mkdir -p ~/.claude/agents/terma

ln -s <CHECKOUT>/out/commands ~/.claude/commands/terma ln -s <CHECKOUT>/subagents ~/.claude/agents/terma

Use /orient to begin each session. Use /research :question to probe the codebase and write a report to /research. Then, use /plan or just /feature to plan a change to the application.

Use /implement to spin up one or more well-instructed subagents to implement the plan.

You may find use for /debug, /code-review, /harden after implementation.

When you are at a known good state (i.e. about to commit) use /progress to write a progress report and update LOG.md, then commit w/ the .md file included. /next-up is like /progress but moves straight on to whatever additional task you provide.

You can use /bug-report to interactiely gather and record context for known issues, and use /resolve to resolve them.

We currently assume a protocol of LOG.md, BUGS.md, SPEC.md, CLAUDE.md etc. but this will and should be customized to fit.

• feature dev: /orient, /feature, /implement, /progress, /compact (loop)

  • then: /code-review

• bugs: /bug-report, /debug, /resolve, /code-review

• tech spike: /prototype, /debug

• improve codebase architecture: /orient, /research, /decompose, /code-review

The build depends on deno. You can edit anything in lib or the root and run and use ./build.sh to rebuild all. We use a simple remark transform for text inclusion, nothing fancy.

The subagent.md file encourages "ultrathinking", which may burn through usage quickly. Consider customizing it manually until we have variables.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.