Authoring pilot: al-api-page-author skill + code-artifact output kind

[JesperSchulz]

Summary

This is the first authoring action skill for BCQuality. Every action skill today is a review skill that consumes a diff and emits a findings-report; this PR adds the authoring half — a skill that consumes an abstract spec and generates Business Central AL code.

The pilot is an API-page author: given an object-spec (target table, entity naming, fields to expose, read-only flag), it generates a correct PageType = API page applying the existing curated web-services API guidance.

What's in this PR

• microsoft/skills/author/al-api-page-author.md (new author/ family) — a leaf authoring skill mirroring the canonical al-web-services-review Source → Relevance → Worklist → Action → Output shape, but emitting generated AL. It applies the existing web-services knowledge (required API properties, SystemId-as-key, read-only lockdown, committed-data reads, bound actions, additive versioning) and emits unresolved spec ambiguities as open-questions / per-artifact notes with clearly-labeled placeholders rather than invented values.
• skills/do.md (⚠️ stable contract change — requires both-maintainer approval) — additive only, the existing findings-report schema is untouched:
  • Adds object-spec to the standard inputs and defines it (abstract description of the BC object(s) to generate, supplied by the orchestrator for authoring tasks).
  • Updates outputs prose to state two output kinds are defined — findings-report (review) and code-artifact (authoring) — keeping the single-element list.
  • Adds a new ### Code-artifact output (authoring skills) subsection: a JSON schema that reuses the findings-report envelope (skill, outcome enum, outcome-reason, JSON-validity/escaping rules, suppressed, sub-results, skipped-sub-skills) and swaps findings/counts for artifacts/open-questions, with field semantics for artifacts[].content (full AL as one escaped JSON string), path, references (empty ⇒ confidence capped at medium, mirroring the additive agent-findings principle), notes, and open-questions. Documents that a code-artifact maps to file creation/scaffolding rather than PR comments.
• .github/scripts/validate_frontmatter.py — allow-list updates only, no rule-logic changes: object-spec added to STANDARD_INPUTS (R17 no longer warns) and code-artifact added to ALLOWED_OUTPUTS (R18 no longer errors).
• README.md / agent-consumption.md — minimal docs: note that the action-skill family also includes an author group that generates code (emitting code-artifact), with al-api-page-author.md as the first example; mermaid diagram unchanged.

No duplicated knowledge

This pilot reuses the existing microsoft/knowledge/web-services/* articles — no new knowledge articles were added (curating duplicates would fail the repo's remedial-admission test). Article count is unchanged.

Note: the local index reports 192 articles before and after this change (the task brief referenced 191; the invariant — no knowledge articles added, count unchanged — holds regardless of the exact baseline).

Validation

• python .github/scripts/validate_frontmatter.py --root . → 0 errors / 0 warnings
• pwsh .github/scripts/Test-KnowledgeIndex.ps1 -Root . → PASSED, 192 articles, deterministic, full coverage (unchanged)
• pwsh ./tools/Build-KnowledgeIndex.ps1 → builds, 192 articles
• R26 unaffected — the author skill lives in author/, is not an al-*-review.md leaf, and is not registered in al-code-review.md sub-skills.

Do not merge — opening for review.