Every initialization step, as the declarations name them

You are making sure a harness's initialization is complete: right after installing it (initialize-harness calls complete-initialization), or because you were dispatched to do exactly this with nothing but the diagram and the repository. Either way the steps are the same, and they are not listed in this file. They are named by the declarations.

Where the steps come from

Read <name>.json at the root, then each name in its needs, from a sibling checkout at ../<name>/<name>.json. Each step below names the declaration that names it, so a reader can see why it is there.

The two schemas: steps are the primary ones (owner, 2026-09-30: "json(ld) is primary step in initializing KG harness"). They are checked first once the declarations are read, and the report leads with them: publish-documents. Everything after them is either how those documents are reached (the site) or how a person reads about them (the README).

A harness above bootstrap names more steps the same way. A work-plan store it declares as a directory is a directory: step, and a harness's own set-up instructions are its instructions: step. Bootstrap does not need to know which harness declares what.

The site steps are in publish-site.

Four states, not two

• done: checked, and it holds.
• not done: checked, and it does not. Say exactly what to do next.
• could not determine: the check could not run (no forge CLI, no network). Never reported as done. A check that could not look is not a pass.
• stated: the declarations name it but nothing can observe it (reading and following a harness's instructions). Follow it, say you did, and leave it stated, as a stated precondition in bootstrap's diagrams is never reported as satisfied.

Idempotent by construction

Check before doing, every time. A done step is left alone, so running the process twice changes nothing the second time, and an interrupted run is resumed by running it again. Performing is not the same as done: after you do a step, check it again, and only the check says done.

Replace nothing. A step that would overwrite what somebody wrote (an existing README, a Pages site built from a branch, a declaration that disagrees with the directories on disk) is a step for the person. Ask through human-agent-discussion, with the exact step written out.

Do not create an empty declared directory to make its step pass. Git does not track an empty directory, so the check would pass on your machine and fail on every clone. Put its first file in it, or ask whether the entry should go.

Where tools are available

A toolset may perform the checks for you — for example https://github.com/litlfred/bootstrap-tools, whose init command reads the same declarations, does what a tool can (regenerating README sections; switching Pages on when an authenticated gh is present), and prints each step in these four states with what to do next:

bun run scripts/init.ts --root ../<instance>        # from a toolset checkout beside it
bun run scripts/init.ts --root ../<instance> --dry-run   # check only

Where no tools are available, do the same by reading files. The diagram is the same either way, and so is the report.

The report

One report through log-message, led by the primary step's state (the JSON Schemas and JSON-LD at their IRIs): then every step, the declaration that named it, its state, what was seen, and for anything not done, what to do and who does it. End with the counts (done, not done, could not determine, stated). An initialization that finished with steps open says which. It is not hidden behind "installed".