Catch-up prompt

Paste into an agent session at the root of an existing repo.

Bring this repo in line with the project-knowledge rules at https://mojzis.github.io/clauding/rules/v1/ (rationale: https://mojzis.github.io/clauding/llms.txt).

1. Read the rules. Check AGENTS.md / CLAUDE.md for a project-knowledge rules vN marker. - No marker: full adoption (steps 2–8). - Older version: apply only what changed between versions, then bump the marker.
2. Add the rules section to AGENTS.md. If CLAUDE.md exists, have it reference AGENTS.md rather than duplicate it.
3. Copy the scaffold: QA.md, docs/style.md, docs/tags.md, docs/adr/, docs/papercuts/. Read docs/style.md now. It governs everything you write below.
4. QA.md (one per subproject in a monorepo): - Find the commands in CI config, pyproject.toml, Makefile/justfile. CI is the reference. - Run every tier you can. Fill expect only from what you observed. Never write an expectation you didn't see. - A tier that fails or can't run now becomes a papercut. Don't fix it in this pass. - Propose oracles from the code and data: invariants, bounds, candidate reference cases. All go in as status: proposed. List them in the report for a human to confirm.
5. docs/architecture.md: if missing, write it from the code as it is today. This is the baseline. Describe only what you can verify in the code and don't guess at intent. Do not write ADRs for past decisions. ADRs start from here.
6. Existing docs: split into one-concept files, add frontmatter, add code: refs you've verified exist, and link related files. Tighten them to the style, but keep the facts.
7. Tags: derive 10–20 tags from what the docs actually cover. Write them into docs/tags.md, one line each.
8. Log as papercuts anything that confused you during this pass.
9. Report: files created or split, tiers run and their observed results, proposed oracles, tags chosen, papercuts logged, open doubts.

Make it one commit (or a small series), separate from code changes.

raw markdown