Knowledge management plugins for Claude Code — planning, research, and project initialization
npx claudepluginhub kevinswiber/gumboKnowledge management for Claude Code — planning, research, and project initialization
kevinswiberVersion-controlled development thoughts, plans, research, and issues -- separated from code repos.
This is my personal workflow tool. If you're here, you should watch a clip from this cooking show I grew up with:
Keep your inner dev loop artifacts (implementation plans, research, issue tracking) out of your code repos while still version-controlling them. Each project gets a .gumbo symlink pointing to a per-project directory under ~/.gumbo.
The file system acts as scratch space for each phase of the inner dev loop so you don't lose context between /clear calls. Each skill runs independently -- /clear between steps.
Run /research-create to create a research plan. This produces a list of research tasks -- things like searching the web for an issue, reviewing codebases, combing through git logs, or going through issues and PRs. Research is instructed to find info on the what, where, how, and why.
Run /research-resume to execute the plan. This launches subagents in parallel to carry out each research task. Each subagent writes its findings to a file in the research subdirectory. From there, you can edit, refine, have conversations, or do more research.
Run /plan-create to create an implementation plan in .gumbo/plans/. Each plan gets its own subdirectory containing implementation-plan.md, task-list.md, and a tasks/ directory with a file per task. Tasks are grouped into phases, with commits after each phase for more atomic changes.
You can base a plan on previous research: /plan-create research 0044.
Run /clear if you haven't already, then /plan-resume to start implementation. Along the way, it records findings in the plan's findings/ subdirectory -- deviations from the plan, new information, things that came up.
After everything's committed, run /plan-findings-resume to create issues in .gumbo/issues/ from the findings. Those issues might spawn more research plans or direct fixes.
When done, run /research-archive or /plan-archive to move completed work into the .gumbo/{research,plans}/archive/ directory.
No phase depends on another. You can have multiple research and implementation plans running at the same time, or nest them -- research informing plans informing more research. Each has its own state file tracking progress.
Run /gumbo-init from your code directory to set things up. Each code repo gets a .gumbo symlink that points to a project directory under ~/.gumbo:
~/src/myapp/
├── .gumbo -> ~/.gumbo/projects/myapp # Symlink to data directory
├── .gitignore # Contains "/.gumbo"
└── ...
The data lives outside the code repo so it can be version-controlled separately:
~/.gumbo/ # Data root (version-controllable)
└── projects/
└── myapp/
├── config.json # Project metadata and backlink
├── AGENTS.local.md -> <plugin-root>/plugins/gumbo/AGENTS.local.md
├── plans/
├── research/
└── issues/
config.json stores the project name, working directory, and data root so you can trace in both directions.
Add the marketplace and install plugins:
claude plugin marketplace add kevinswiber/gumbo
claude plugin install gumbo@gumbo
Run /gumbo-init from your code directory. You can optionally pass a project name:
/gumbo-init myapp
Without a name, it uses the directory name. You can also run the script directly:
plugins/gumbo/skills/gumbo-init/scripts/init.sh ~/.gumbo ~/src/myproject
plugins/gumbo/skills/gumbo-init/scripts/init.sh ~/.gumbo ~/src/myproject myapp
This will:
~/.gumbo/projects/myapp/ (if it doesn't exist)plans/, research/, issues/) with their CLAUDE.md filesconfig.json with the project name, working directory, and data rootAGENTS.local.md to the shared copy in the gumbo plugin~/src/myproject/.gumbo pointing to the data directory/.gumbo to the project's .gitignore/plan-create -- Create an implementation plan/plan-resume -- Resume an in-progress plan/plan-archive -- Archive a completed plan/plan-cancel -- Cancel a plan/plan-findings-create -- Extract findings from completed phases/plan-findings-resume -- Triage findings into issues/research/research-create -- Create a research plan with parallel investigation/research-resume -- Resume or synthesize research/research-archive -- Archive completed research/research-cancel -- Cancel research