supaschema init installs or merges its Claude, Codex, and AGENTS-compatible surfaces into the consuming project.
For a new project setup, copy the agent prompt from Installation. The important instruction is simple: install the package with the consuming project’s package manager, run supaschema init, inspect the generated config and install manifest, then run schema commands through that same manager’s local runner.
Agent workflow
1
Install the package
Install
supaschema in the package or workspace that owns the schema
workflow. Do not clone the supaschema repository into the project unless you
are intentionally developing supaschema itself.2
Review installed context
supaschema init installs or merges the active agent files and records the
result in .supaschema/install.json. The raw package copies remain under
node_modules/supaschema/agent-bundle/ for audit and repair. When hosted
docs are unavailable, start with agent-bundle/docs/index.md; the offline
MDX corpus remains inside node_modules and is not copied by init.3
Edit schema intent
Change the configured declarative SQL tree, such as
database/schemas/**,
neon/schemas/**, aws-postgresql/schemas/**, or supabase/schemas/**.4
Run the workflow commands
Run
diff, check, and types through the local runner from the install
prompt. diff renders the migration, check gates replay safety, and
types regenerates TypeScript and Zod outputs.5
Report the result
Summarize the generated migration, generated outputs, target runner
expectations, and any
SUPA_* diagnostic. Apply only through an explicit
user request or a config-gated supaschema apply workflow with resolved
targets and required remote approval.Installed agent context
The public npm package includes the consumer-safe raw surfaces needed for that workflow:Workflow and policy
Install prompt
Copyable setup instructions for another coding agent.
Rule
Durable migration policy: generated files are not hand-edited, destructive
intent must be explicit, and apply is config-gated.
Skills
Three curated workflows cover core policy, migration execution, and ongoing
maintenance.
Runtime and reference
Hooks
Write-time guardrails that block generated migration edits, run diff or
config-gated apply after schema edits, and sync installed LLM surfaces.
Offline docs
A byte-exact MDX corpus with a sorted canonical-URL index, available inside
the installed package without adding documentation files to the project.
Optional discovery surfaces
Use these only when the agent or documentation client supports them.Hosted skill
Public agent skill metadata for clients that can read
https://supaschema.com/docs/skill.md.llms.txt
Compressed docs context for agents that load
https://supaschema.com/docs/llms.txt.MCP docs
Mintlify MCP endpoint for documentation-aware clients at
https://supaschema.com/docs/mcp.npx skills
Installs reusable skill context only; project setup still comes from the
package manager plus
supaschema init.The npm package is the project integration path.
npx skills is useful for
reusable agent context, but it cannot replace project config,
path-confirmation state, or the active agent surfaces installed by supaschema init.