curator: ingest oscal-presence-gate + 12 new articles; refresh driftlab-dgx
- New stub oscal-presence-gate (oscal-compliance) — presence-aware OSCAL
guardrails for delegated agent traffic; category page, Home, Index updated
- 12 new article pages (5 ai-security, 5 agentic-frameworks,
1 oscal-compliance, 1 developer-tools); timeline +July 2026;
Articles sections updated on 4 category pages
- driftlab-dgx stub: v2.1.1, last_commit_date 2026-07-28
- articles-published stub: 34 articles, last_commit_date 2026-07-29
- Sitemap.md + Sitemap-Stacks.md regenerated; Log.md appended
curate: ingest DGX Spark wave — new local-inference-dgx category, dgx stack, 6 new repos, 2 renames, projectpulse purged
- SCHEMA: +category local-inference-dgx, +stack dgx, tags += cuda/vllm/dspy
- New stubs: bonsai-ternary-27b-dgx, screen-lens-dgx, driftlab-dgx, ralph-dgx, tslit-dspy-ar, tslit-dspy-dgx
- Renames: driftlab -> driftlab-mlx, screen-lens -> screen-lens-mlx (stub ids stable)
- Deleted projectpulse stub + references (user-approved; repo no longer public)
- New wiki/local-inference-dgx/ overview page; 4 category pages updated
- Home 50->56 repos + DGX section; Index +7/-1; sitemaps regenerated (both views)
- nemotron-parse-spark: stacks += dgx
wiki-curator 2026-07-09: ingest 8 new repos, refresh tars-ai/regintel, resolve rename
New repos: STTbench, ace-step-1-5-mlx, ltx-2-3-mlx, longcat-video-avatar-1-5-mlx,
driftlab, langgraph-checkpoints-vs-stores, venture-pathfinder, strix-omlx.
Updated stubs: tars-ai (now OMLX-server-backed for LLM+TTS), pentest-regulatory-intel
(periodic monitoring + CLI package), screenlens (date). Repointed
local-mlx-responsesAPI-server stub to the renamed mlx-responses-api-server.
Refreshed 5 category/pattern pages + Home + Index; regenerated both sitemaps.
Normalized version-dotted stub IDs and fixed a pre-existing broken citation.
Known-missing remote: project-pulse-sap (projectpulse stub retained).
Ingest oscal-skills-guardrails, skillspector-trial, dflash-mlx-trial
feat(wiki): ingest 6 new repos, update RepoBundle + NVD-Extractor + add local-mlx-wrapper-pattern
Ingest wiki-vs-rag and nemotron-parse-spark; refresh affected pages
Two new repos appeared in the sw30labs org since 2026-04-15:
* wiki-vs-rag (agentic-frameworks, stacks: [agentic, rag])
Reproducible benchmark of single-shot RAG vs agentic-RAG vs
wiki-navigation vs QMD over the sw30labs wiki corpus pinned at
c130a3b. Agentic-RAG wins Pareto on quality vs cost; QMD is
cheapest but collapses on branch / cross-branch synthesis.
* nemotron-parse-spark (developer-tools, stacks: [converter, cli])
Test harness for NVIDIA Nemotron Parse v1.2 on DGX Spark
(Grace Blackwell GB10). CLI + REPL + batch + benchmark; both
transformers and vLLM backends share a single postprocessing path.
Touched:
- raw/wiki-vs-rag.stub.md, raw/nemotron-parse-spark.stub.md (new)
- wiki/agentic-frameworks/agentic-frameworks.md (6 -> 7 repos, +3 claims)
- wiki/developer-tools/developer-tools.md (7 -> 8 repos, +3 claims)
- Home.md (counts 32 -> 34, two new table rows)
- Index.md (two new stub rows)
- Sitemap.md, Sitemap-Stacks.md (regenerated)
- Log.md (run entry)
No updated_repos, no stale_pages, no orphan_stubs.
Articles repo unchanged (still 22, all already mapped).
Sitemap-Stacks: split into per-stack diagrams
Same density problem as Sitemap.md - all 10 stacks crammed into one
mermaid graph renders as an unreadable horizontal smear with a giant
grey Unclassified block. Replaced with:
- Adoption table kept at the top, sorted by repo count
- One LR diagram per stack, showing only that stack's repos
- Multi-stack repos appear in each of their stacks, with the primary
stack using solid arrows + solid stroke, secondary stacks using
dashed arrows + dashed stroke (via classDef)
- Articles excluded from this view (stacks view is about tech stacks,
not prose); Unclassified now only shows repos lacking a stacks: field
Largest sub-diagram is langgraph at 17 repos; smallest is mcp at 1.
Taxonomy migration: SCHEMA.md yaml block + stub classification
- SCHEMA.md now carries a fenced yaml taxonomy block with 5 categories
and 10 stacks, parseable by scripts/taxonomy.py. Kept tag whitelist,
voice, freshness threshold (90d), and directory-structure prose
below the yaml fence.
- Added category: + stacks: frontmatter to all 55 stubs (33 repo
stubs, 22 article summary stubs). Article stubs generally declare
stacks: [] when the article is opinion/analysis with no tech-stack
tie; the validator treats those as informational, not errors.
- Regenerated Sitemap.md (category view) and new Sitemap-Stacks.md
(cross-cutting stack view, with adoption table + mermaid graph).
detect_changes.py post-flight: taxonomy.ok=True, 0 bad/missing
categories, 0 bad stacks, 14 stubs with empty stacks (informational).