stackfoundry plugin
Studies your repository, researches each language, app model, capability and tool it uses from current official sources, and writes repo-grounded rulebooks and target profiles. Generic role agents (bug hunter, test writer, fix verifier, security and performance reviewers…) and orchestrator skills then work from those facts, so every specialist knows your real paths, harnesses and gate commands.
Install and start
/plugin marketplace add OleksandrOmelchenko/stackfoundry
/plugin install stackfoundry@stackfoundry
/stackfoundry:init
init surveys the repo, proposes a target map (each deployable or testable unit with its stack of layers), asks you
to confirm it, researches every technology in parallel, grounds the results in your code, proves each gate command
once, and reports what it wrote. It never overwrites existing .claude/ content without asking.
Later:
| Want | Run |
|---|---|
| Find bugs in one target | /stackfoundry:bug-hunt <target> |
| Write tests the right way | /stackfoundry:design-tests <target> then /stackfoundry:write-tests <target> --design <doc> |
| Review uncommitted work | /stackfoundry:review-changes |
| Investigate a failure | /stackfoundry:debug |
| See what is outdated | /stackfoundry:check-updates, then /stackfoundry:act-on-updates |
| Keep the research fresh | /stackfoundry:tech-radar, then /stackfoundry:act-on-radar |
| Pick up a new project or framework | /stackfoundry:sync |
| Take a new plugin version's templates | /stackfoundry:upgrade |
Settings
The repo's settings (the assistants it uses, the docs root, the research refresh window and the default fan-out
effort) are in .stackfoundry/config.json; stackfoundry config prints them with the defaults filled in.
Requirements
Claude Code (or another assistant in reference/clients.md), git, Python 3.11 or newer on PATH as python3,
python or py -3, and sh (Git Bash on Windows) or PowerShell 7. Claude Code's guard hook runs through sh: on
Windows without Git Bash the guard is off, and stackfoundry guard --check says so.