~/groundwork

Other tools

The core is plain markdown, so any agent that can read files can follow Groundwork. Install with the none adapter:

npx groundwork-ai init --adapter none

You get AGENTS.md plus the full .groundwork/ folder — nothing tool-specific. Most modern agents read AGENTS.md on their own; if yours doesn't, point it there.

Working without native commands

Ask your agent:

Read .groundwork/commands/gw.md and follow it.

Every command works the same way — for example "read .groundwork/commands/gw-next.md and follow it" builds the next card. The commands are written to be followed by any capable model, and their descriptions say when each one applies, so the agent can also choose the right one from a plain request.

What changes without an adapter

With an adapterPlain markdown
CommandsNative /gw-*Read the command file and follow it
RolesSeparate subagents, fresh contextOne session plays each role in turn
GuardsWired into tool hooksInstructions only — nothing blocks automatically
Session startWhere things stand is added automaticallyThe agent reads HANDOFF.md as rule 1

The workflow covers the single-session mode: the agent reads each role file before playing that role, finishes it completely before switching, and as reviewer re-reads the diff from scratch and re-runs every check itself.

Honest trade-off: without hooks, a RULE is still an instruction, and a GUARD can't block by itself. That's the one thing an adapter adds that markdown alone can't.

The CLI still helps

Even with no adapter, Node users can use the terminal side:

npx groundwork-ai status    # where things stand, from files
npx groundwork-ai doctor    # install health and token budget
npx groundwork-ai retro     # signals for /gw-retro
npx groundwork-ai upgrade   # refresh Groundwork's core files

Nothing in the workflow depends on the CLI — it's the deterministic helper, not the engine.

On this page