Add the SQA and QC framework and project instructions

Mount the framework as a submodule at framework/ (v0.4.0), add AGENTS.md and
the installed skill copies for Codex CLI (.agents/skills) and Claude Code
(.claude/skills).

.agents/skills/artifact/SKILL.md

@@ -0,0 +1,131 @@
+---
+name: artifact
+description: Create, edit or review any project document artifact — Business Case, Stakeholder Analysis, KPI, Business Model Canvas, BPMN, milestones/gateways, use case diagram, user stories, use cases, domain model, SSD, operation contracts, sequence diagrams, DCD, ERD, ADRs, SQA review records, traceability matrix, governance, QC checklists. Scaffolds the file with the correct ID, CrossReference and links, then gives the required sections for that type.
+---
+
+# Artifact
+
+Every artifact is a markdown file with `## Metadata` and `## Version History`
+tables, a registered short-name ID, and links only at the bottom. Types and
+their related types are in `framework/registry/artifact-catalog.md`; this
+project's file locations and next versions are in `docs/artifact-registry.md`.
+
+## Creating an artifact
+
+1. Find the type's short name (`BC`, `SA`, `MIL`, `DM`, `ADR`, `RC`, …) in
+   the catalog. If the type is new, add it to the catalog and the registry.
+2. Scaffold — this sets the ID, `CrossReference` (only artifacts that exist),
+   the bottom link block and the registry version:
+
+   ```bash
+   bash framework/scripts/new-artifact.sh <SHORT> [--file <path>] [--title "<text>"] [--cite <ID>=<path>]...
+   ```
+
+   `--file` is required for multi-document types (`MIL`, `UC`, `ADR`, `RC`,
+   `QC`); single-document types default to the registry's `Primary File`.
+   The script refuses to overwrite an existing file.
+3. Read `framework/.agents/skills/artifact/references/<SHORT>.md` (cite-for
+   hints and required sections), then fill in the file. Keep the sections in
+   order and replace every `<placeholder>`.
+
+## Editing an artifact
+
+Append a row to `## Version History` on every change (a status change such as
+`Proposed` → `Accepted`, or a content change). To re-check `CrossReference`,
+run `bash framework/scripts/find-crossreferences.sh <SHORT>`. When the first
+instance of a type is created, add it to the `CrossReference` of every
+existing artifact that lists that type as a candidate (with a new Version
+History row).
+
+### Version History rule
+
+The table keeps only the **two latest** changes; git holds the rest. Columns:
+
+| Date | Status | Author | Reviewer | Change | Commit |
+| --- | --- | --- | --- | --- | --- |
+| 2026-10-01 | Accepted | Jens Tirsvad Nielsen | S02 | Added Risks section<br>Fixed scope wording | [a1b2c3d] |
+
+- **Status** — `Proposed`, `Accepted`, `Rejected` or `Deprecated` for every
+  artifact type (`ADR` also has `Superseded by ADR-NNNN`); `Approved` is not
+  used. A new row starts `Proposed`. When the review gives Go, that row
+  becomes `Accepted` and the row before it becomes `Deprecated`, so at most
+  one row is `Accepted`: the latest reviewed one. If the reviewer refuses the
+  change for good (not a No-Go that returns it for rework), the row becomes
+  `Rejected`; the earlier `Accepted` row stays `Accepted` and nothing is
+  deprecated.
+- **Change** — a short summary of what this row changed; several lines are
+  separated with `<br>`.
+- **Commit** — the commit that made the change, as a reference-style link
+  defined at the bottom of the file
+  (`[a1b2c3d]: https://<host>/<owner>/<repo>/commit/<full-hash>`; Gitea and
+  GitHub use the same `/commit/` form). A row cannot contain its own commit
+  hash, so a new row says `pending` until the commit exists. Leave it
+  `pending` and do not commit: only when the user asks for a commit, then:
+  1. commit the edited documents;
+  2. run `bash framework/scripts/resolve-pending-commits.sh <file>...`, which
+     replaces `pending` with the link to that commit and adds the definition;
+  3. commit the result as a follow-up commit, before opening the PR. Do not
+     amend: an amend changes the hash, so the link would point at a commit
+     that is never pushed.
+- When adding a third row, delete the oldest row and its now-unused commit
+  link definition. Never rewrite the content of the two retained rows other
+  than resolving `pending` and setting Status to `Accepted` / `Deprecated`
+  after a Go review.
+- Every document uses this format, including QC checklists. A file still in the
+  old four-column format is converted when next edited (old row kept as
+  `Initial version`, plus a new row for the conversion).
+
+## Reviewing an artifact
+
+1. Take the QC checklist named in the catalog (`framework/qc/qc-*.md`).
+2. Create the review record: `new-artifact.sh RC …`, following
+   `references/RC.md`.
+3. Add or update the instance's row in the Traceability Matrix.
+
+## Rules
+
+- **Links:** all links are reference-style, defined once at the bottom of the
+  file after a final `---`, labelled by the target's ID (`[SA-001]`), never
+  inline. No links → omit the block.
+- **CrossReference:** cite only artifacts that exist now. Empty if none.
+- **People:** use exact stakeholder IDs from the project's Stakeholder
+  Analysis (`S01`) for owners, reviewers and RACI. Never invent role names.
+- **Diagrams** are PlantUML blocks; check them with
+  `bash framework/scripts/render-diagrams.sh --server <url> <file>` (or set
+  `PLANTUML_URL`) before review.
+- **QC checklists** are framework files: every criterion is tagged with an
+  ISO/IEC 25010:2023 characteristic, and they never mention real instances.
+- **IDs:** `<SHORT>-<version>`, 3 digits (`BC-001`); `ADR` uses 4 digits;
+  `RC` is sequential across all types; QC is `QC-<SHORT>-<version>`.
+- **Language:** the PO language and the register of each artifact type are in
+  the `Languages` section of `docs/artifact-registry.md` (registers:
+  `IT Executive English` for high-level, `IT Professional English` for
+  technical).
+- **Language and Domain rows:** every artifact of a type written in the PO
+  language has `Language` and `Domain` rows in its Metadata table, so a reviewer
+  sees at once how to read it. `Language` is a BCP 47 code (`da`, `en`);
+  `Domain` is a value from the domain list in the registry's `Languages`
+  section (for example `it`, `medical`, `construction`), because the same
+  language can carry a different professional vocabulary. `new-artifact.sh`
+  fills both from the registry's `PO language` and `PO domain` settings and
+  leaves a placeholder (with a warning) when a setting is missing. Technical
+  types (OC, SD, DCD, ERD, ADR, TM, RC, QC, source code) have no such rows:
+  they are always professional IT English. To list every document's language
+  and domain, see `check-languages.sh --list`.
+- **One file per artifact:** each type the registry marks "Written in the PO
+  language" exists once, in that language, under its normal name
+  (`business-case.md`). There is no translated twin (`business-case.da.md`)
+  and no authoritative English source; two files drift apart. Use the PO terms
+  from the dictionary (`DICT`). Types marked "No" (OC, SD, DCD, ERD, ADR, TM,
+  RC, QC, source code) stay in professional IT English.
+- **Structural vocabulary stays English:** Metadata keys, section headings,
+  IDs and statuses are the same in every language, because the scripts read
+  them (`find-crossreferences.sh`, `new-artifact.sh`,
+  `resolve-pending-commits.sh`, `sync-project.sh`, `check-plan.sh`). Only the
+  content (prose and table cells) is in the PO language.
+- **Changing an artifact's language** is a material change: add a Version
+  History row ("language en to da") and review it again. Git history keeps the
+  earlier language.
+- **Dictionary:** the Domain Model uses the PO term; the Operation Contract,
+  Sequence Diagram, Design Class Diagram and ERD use the IT term. Every pair is
+  recorded in `docs/dictionary.md` (`DICT`).