Skip to content

Log

Jump to bottom
Nicolas Cravino edited this page Jul 16, 2026 · 13 revisions

Change Log — sw30labs Wiki

All significant additions and updates to the knowledge base.

2026-07-04

Ingest 3 new public repos: oscal-skills-guardrails, skillspector-trial, dflash-mlx-trial

Incremental update run. Three repos added to the org since mid-June (all previously unstubbed) ingested and wired into their categories.

New stubs (raw/):

  • oscal-skills-guardrails.stub.md — category oscal-compliance; stacks [oscal, agentic, langgraph, mcp, compliance, cli]. OSCAL-as-policy carried from tools to Agent Skills; dual-evidence admission, digest lockfile, CI gate, assessment-results audit. Direct descendant of oscal-agent-guardrails.
  • skillspector-trial.stub.md — category ai-security; stacks [agentic]. Single-file offline scanner grading Agent Skills A–F; the static evidence stream consumed by oscal-skills-guardrails.
  • dflash-mlx-trial.stub.md — category local-inference-mlx; stacks [mlx]. DFlash block-diffusion speculative decoding for Qwen3.6-27B on Apple Silicon; ~3.37× measured, identical output.

Category pages updated: oscal-compliance (6→ six OSCAL repos, new key claim + cross-ref), ai-security-pentesting (added Skillspector, supply-chain framing), local-inference-mlx (14→15 repos, DFlash claim). All updated: bumped to 2026-07-04.

Home.md / Index.md: added all three repos to their category tables and the index; refreshed header counts (43 repos) and cross-cutting stack tallies.

Cross-references established: oscal-skills-guardrails ↔ skillspector-trial (consumer/engine), oscal-skills-guardrails → oscal-agent-guardrails (ancestor), dflash-mlx-trial ↔ qwenbench-mlx / local-mlx-responsesAPI-server, and a Qwen3.6-27B-bf16 + oMLX thread linking dflash and oscal-skills-guardrails.

Environment note: the Cowork sandbox mount could not run git or the sitemap generator over the wiki repo (fuse layer blocks lockfiles and deadlocks on bulk reads). Wiki files were written host-side and are on disk; Sitemap.md / Sitemap-Stacks.md still need regenerating locally (one command — see the run summary), and the commit/push is a manual step.

2026-04-15 (afternoon)

Taxonomy migration — SCHEMA.md yaml block + stub classification

Completed the structural migration flagged in the morning run. SCHEMA.md now carries a fenced ```yaml taxonomy block (5 categories, 10 stacks) parseable by scripts/taxonomy.py, and every stub declares its `category:` and `stacks:` in frontmatter.

SCHEMA.md:

  • Added canonical yaml block with categories: (ai-security, oscal-compliance, agentic-frameworks, local-inference-mlx, developer-tools) and stacks: (mlx, oscal, agentic, pentest, compliance, rag, mcp, langgraph, converter, cli).
  • Kept tag whitelist, voice, freshness threshold (90 days), and directory-structure prose below the yaml fence.
  • Added "Rules" and "How scripts use this file" sections per template.

Stub frontmatter (55 stubs):

  • All 33 repo stubs got category: and stacks: assigned.
  • All 22 article summary stubs got category: (single, by primary topic). Article stubs generally get an empty stacks: [] — 14 of them carry no stack since the article is opinion/analysis rather than tied to a specific tech stack. The validator treats missing stacks as informational, not an error.
  • Injection respects remote_url, repo_path, and all other prior fields.

Stack adoption after migration: agentic 21, langgraph 17, converter 8, compliance 7, mlx 7, oscal 6, pentest 6, cli 3, rag 2, mcp 1. 14 stubs are unclassified (mostly article stubs without tech-stack ties).

Sitemaps regenerated:

  • Sitemap.md — category view, updated in place.
  • Sitemap-Stacks.md — new file, stacks view. Shows adoption table
    • mermaid graph grouped by stack with dashed edges for secondary memberships.

Post-flight detect_changes.py reports taxonomy.ok: True, zero bad/missing categories, zero bad stacks.

2026-04-15

Incremental update — stub hygiene pass

Change detection found no new repos on GitHub, no new articles in articles-published, and no stale wiki pages. Cleanup focused on stub frontmatter drift accumulated from a previous session's paths and a batch of upstream repo renames.

Stub path normalization (23 stubs): All stubs pointing at /sessions/epic-magical-hawking/Documents/sw30labs/... migrated to the canonical ~/Documents/sw30labs/repos/<dir> form required by the skill's path policy. Session paths are ephemeral; the ~/Documents form survives across runs.

Stub remote_url fixes (8 stubs): These stubs pointed at GitHub URLs for repos that have since been renamed. Updated to the current canonical names so remote_url-keyed matching in detect_changes.py stops flagging them as "new":

  • mlx-youtubescribe: MLX-YouTubeScribe.git → youtube-scribe-mlx.git
  • screenlens: screenlens.git → screen-lens.git
  • strixresearch: strixresearch.git → strix-research.git
  • n8n2langraph: N8n2langraph.git → n8n-to-langraph.git
  • projectpulse: projectpulse.git → project-pulse-sap.git
  • local-mlx-responsesAPI-server: → mlx-responses-api-server.git
  • qwen3-vl-python-ocr-script-mlx: → qwen-ocr-mlx.git
  • tslit: TSLIT.git → tslit.git (case change)

last_commit_date refreshed (2 stubs):

  • mlx-youtubescribe: 2026-04-11 → 2026-04-12
  • qwen3-vl-python-ocr-script-mlx: 2026-04-11 → 2026-04-12

Articles: All 22 articles remain in 1:1 sync with articles-published (latest commit 2026-04-11). No timeline churn.

Open item (not applied): SCHEMA.md does not yet contain the fenced ```yaml taxonomy block that the newer detect_changes.py / `generate_sitemap.py` scripts expect (for `categories:` / `stacks:` validation). The current schema still documents the taxonomy in prose only. Post-flight detect_changes reports `category_count: 0` and `stack_count: 0` for that reason. Left untouched in this pass — migrating to the YAML schema is a structural change that should be done deliberately, probably alongside adding `category:` / `stacks:` to every stub's frontmatter.

2026-04-13

Incremental update — 9 new repos ingested

New stubs created (9):

  • autogen-doc-generator → Developer Tools (AutoGen-based code documentation generator)
  • autogen-mermaid-generator → Developer Tools (AutoGen GroupChat Mermaid diagram generator)
  • autogen-requirements-agent → Agentic Frameworks (experimental strict JSON inter-agent messaging)
  • nvd-extractor → AI Security (NVD API vulnerability extraction with network attack vector filtering)
  • ollama-pdf2markdown → Developer Tools (PDF→Markdown via Ollama multimodal, Mistral Small 3.1)
  • repobundle → Developer Tools (Git repo export/import as single text files, stdlib only)
  • word-to-markdown-converter → Developer Tools (.docx→Markdown converter)
  • animated-gif-creator → Developer Tools (image folder / MOV → animated GIF)
  • deepseekvl2-pdf-ocr → Local Inference & MLX (local PDF OCR via DeepSeek-VL2 on NVIDIA CUDA)

Path fixes:

  • Updated 7 stubs with stale session paths (n8n2langraph, local-mlx-responsesAPI-server, projectpulse, screenlens, strixresearch, mlx-youtubescribe, qwen3-vl-python-ocr-script-mlx) to point to current ~/Documents/sw30labs/repos/ locations.

Category pages updated:

  • Developer Tools: expanded from 1 repo to 7, rewrote summary
  • AI Security: added NVD-Extractor claim
  • Agentic Frameworks: added AutoRequirementsAgent claim, updated summary (5→6 repos)
  • Local Inference & MLX: added DeepSeek-VL2 OCR claim, updated summary (8→9 repos)

Articles: All 22 articles already synced — no new articles since last run.

Home.md: Updated repo count (23→32), added all new repo rows to category tables.

2026-04-11

Bootstrap: Created wiki schema and directory structure for sw30labs organization. Initialized with taxonomy covering AI security, OSCAL compliance, agentic frameworks, local inference (MLX), and developer tools. Established tag whitelist, freshness thresholds (90 days), and voice guidelines (concise, technical, present tense, practitioner-oriented).

  • Created: SCHEMA.md, Index.md, Log.md
  • Created: raw/, wiki/ai-security/, wiki/oscal-compliance/, wiki/agentic-frameworks/, wiki/local-inference-mlx/, wiki/developer-tools/

Bulk Ingest (23 repositories): Cloned and ingested all 23 public repos from github.com/sw30labs. Created 23 raw stubs and 5 category wiki pages. Updated Index.md and Home.md.

  • Stubs created (23): tslit, pentest-regulatory-intel, strixresearch, agentic-ai-pentesting, oscal-agent-guardrails, oscal-digital-twin-playground, oscal-zero-trust-lens, oscal-agent-lab, oscal-cac-playgd, agent-stack, deepagent-azure-cli, n8n2langraph, sst-autoresearch, projectpulse, tars-ai, screenlens, qwenbench-mlx, mlx-distillation-explained, local-mlx-responsesAPI-server, audiobook-generator, qwen3-vl-python-ocr-script-mlx, mlx-youtubescribe, gitnexus-fleet
  • Wiki pages created (5): ai-security-pentesting, oscal-compliance, agentic-frameworks, local-inference-mlx, developer-tools
  • Contradictions surfaced: 0
  • Index.md updated: yes
  • Home.md updated: yes

2026-04-11 — Incremental Update (wiki-curator first run)

Result: No changes required. All 23 repos pulled — all already up to date. Change detection confirms:

  • 23/23 GitHub repos cloned and stubbed
  • 5/5 category wiki pages present and indexed
  • 0 new repos, 0 updated repos, 0 stale pages, 0 orphan stubs, 0 unindexed pages

Observations for next run:

  • 2 repos inactive >90 days: TSLIT (last commit 2025-12-05), N8n2langraph (2025-11-29) — will trigger stale-page warnings after next freshness check
  • Most active repos: projectpulse (Apr 10), sst-autoresearch (Apr 9), screenlens (Apr 8)
  • Tag qwen appears in 4+ stubs but is not in SCHEMA whitelist — consider adding
  • MLX-YouTubeScribe uses both chromadb and langgraph but isn't cross-referenced with pentest-regulatory-intel (also uses both)

2026-04-11 — Articles Timeline Integration

Articles Ingest (22 published articles): Cloned github.com/sw30labs/articles-published and created article stubs + timeline page. Articles span 2023-04-23 to 2026-03-08 across all 5 wiki categories.

  • Stubs created (22): revolutionizing-cyber-risk-advisory-ai, safeguarding-datasets-llm-era, transforming-crypto-mining-hardware, ai-generated-code-zero-trust, digital-twins-incident-response, asics-dethrone-gpus-llm, osi-model-natural-language-layer, ai-supply-chain-governance, asics-vs-gpus-grid, custom-silicon-gpt-story, rise-agentic-security, agentic-pentesting-layer7, 72-hour-civilization-agents, oscal-phantom-compliance, agentic-pipeline-code-optional, automated-ai-news-workflow, intelligence-event-horizon, understanding-agent-security-analogies, ai-agent-security-governance, one-person-conglomerate-reality, best-ai-unrestricted-access, zero-trust-ai-coding-boom
  • Wiki pages created (1): articles-timeline (wiki/articles/timeline.md)
  • Category pages updated (5): ai-security-pentesting (+11 articles), oscal-compliance (+2), agentic-frameworks (+4), local-inference-mlx (+4), developer-tools (+1)
  • Index.md updated: yes — added Published Articles section
  • Contradictions surfaced: 0

2026-04-11 — Wiki Curator Update (merged skill)

Operation: Full update using merged wiki-curator skill (articles integration built-in).

Repo sync: All 24 repos pulled (including articles-published). 0 new repos, 0 updated repos.

Articles sync: 22/22 articles already ingested. No new articles found.

New content:

  • Created raw/articles-published.stub.md (repo-level stub for the articles collection — was previously missing)
  • Updated Index.md to include the articles-published stub

Issues: None. Wiki fully current.

Observations:

  • 2 repos remain inactive >90 days: TSLIT (last commit 2025-12-05), N8n2langraph (2025-11-29)
  • Tag qwen still not in SCHEMA whitelist despite appearing in 4+ stubs
  • All article stubs present but no individual article detail pages exist yet (only timeline + stubs)

2026-04-11 — Article Sync + Sitemap Generation

Operator: wiki-curator (automated)

Changes

  • Articles synced: Created 22 article detail pages in wiki/articles/ from articles-published repo
    • AI Security: 16 articles
    • Local Inference & MLX: 4 articles
    • OSCAL & Compliance: 1 article
    • Agentic Frameworks: 1 article
  • Timeline regenerated: wiki/articles/timeline.md now lists all 22 articles chronologically
  • Index updated: Published Articles section now includes all article wiki pages (not just raw stubs)
  • Sitemap generated: Sitemap.md created with Mermaid flowchart showing full wiki structure

Repo status

  • 24 repos on GitHub, 24 cloned, all up to date (no new commits)
  • No new repos to ingest
  • No updated repos to refresh
  • No stale pages detected

Notes

  • articles-published stub flagged as orphan (expected — it's a meta-stub not cited by wiki pages)
  • Many slug-based article filenames lack embedded dates; used git commit dates as fallback

2026-04-22

Update run — two new repos ingested

Picked up two new public repos in the sw30labs org since the last run: wiki-vs-rag (agentic-frameworks) and nemotron-parse-spark (developer-tools). All 35 GitHub repos cloned; no updated_repos, no stale_pages, no orphan_stubs. Articles repo unchanged (still 22 articles, all already mapped to wiki pages).

New stubs:

  • raw/wiki-vs-rag.stub.md — Reproducible benchmark of single-shot RAG vs agentic-RAG vs wiki-navigation vs QMD over the sw30labs wiki corpus pinned at c130a3b. Headline result: agentic-RAG is the Pareto winner (4.77 quality vs 4.69 for wiki-nav at 45% of the cost; 11–2 vs 17 ties against single-shot RAG). QMD is cheapest ($0.005/query, no embedding API) but collapses on branch / cross-branch synthesis (2.83 / 3.37 vs 4.5+ for the others). Categorized agentic-frameworks, stacks: [agentic, rag].
  • raw/nemotron-parse-spark.stub.md — Test harness for NVIDIA Nemotron Parse v1.2 (<1B-param ViT-H + mBART document-understanding model) on DGX Spark (Grace Blackwell GB10, 128 GB unified memory). CLI + REPL + batch processor + benchmark; transformers and vLLM backends share a single postprocessing path. Notes the load-bearing 4-token v1.2 prompt and the 1024×1280 / 1664×2048 image-sizing contract. Categorized developer-tools, stacks: [converter, cli].

Pages updated:

  • wiki/agentic-frameworks/agentic-frameworks.md — bumped summary from 6 to 7 repos; added 3 key claims for wiki-vs-rag (the four-arm setup, the agentic-RAG Pareto result, the QMD branch/cross-branch collapse); added 2 open questions; tags now include rag, benchmark; updated to 2026-04-22.
  • wiki/developer-tools/developer-tools.md — bumped summary from 7 to 8 repos; added 3 key claims for nemotron-parse-spark (NVIDIA Nemotron Parse on DGX Spark, the 4-token v1.2 prompt contract, the image-sizing envelope); added 1 open question (Nemotron vs DeepSeek-VL2 vs Qwen3-VL); tags now include benchmark; updated to 2026-04-22.
  • Home.md — repo count 32 → 34, last-updated 2026-04-15 → 2026-04-22; added rows for wiki-vs-rag (Agentic Frameworks table) and nemotron-parse-spark (Developer Tools table).
  • Index.md — added wiki-vs-rag and nemotron-parse-spark stub rows to the relevant category tables.
  • Sitemap.md and Sitemap-Stacks.md — regenerated to pick up the new stubs (2 nodes added to agentic-frameworks and developer-tools diagrams; new entries in the agentic / rag / converter / cli stacks).

Repo status

  • 35 repos on GitHub, 35 cloned, all up to date.
  • 2 new repos ingested (wiki-vs-rag, nemotron-parse-spark).
  • 0 updated repos.
  • 0 stale pages.
  • 0 orphan stubs.
  • Taxonomy validation: all 57 stubs have valid category: and known stacks: entries. 14 article stubs still report stubs_missing_stacks — informational, expected for opinion/analysis pieces.

Notes

  • Articles repo unchanged since the previous run; no new article pages needed.
  • The wiki-vs-rag repo benchmarks this very wiki as its corpus (pinned at commit c130a3b) — the reflexive loop is a nice symptom that the wiki is now substrate for downstream agent work, not just documentation.
  • The nemotron-parse-spark stub deliberately calls out the prompt / image-size landmines from the model card so future readers don't silently degrade output by tweaking either.

2026-05-19

Sync run — rename drift fix + local-clone audit

Routine sync. No new repos, no new articles, no content updates to wiki pages. The bulk of this run was untangling stub/repo state drift that had accumulated since the last sync.

Stub frontmatter fixed:

  • raw/local-mlx-responsesAPI-server.stub.md — corrected remote_url and repo_path from mlx-responses-api-server to local-mlx-responsesAPI-server. The repo was renamed on GitHub (the old URL still redirects), but the stub was pointing at the pre-rename path. Without this fix, detect_changes.py reported the cloned dir as a "new repo" and the stub as a "missing repo" at the same time.

Articles:

  • 22 articles in articles-published ✓ 22 wiki pages with matching source_file. Nothing to ingest. Timeline updated: left at 2026-04-11 per the timeline-stability rule (no content changed).

Pages updated: none.

Sitemap: not regenerated — no pages added or removed; structure unchanged.

Repo status

  • 34 repos on GitHub, 24 cloned locally, 0 updated.
  • 11 stub missing_repos (stubs whose repo_path directory doesn't exist on the host — same set listed in the rename audit below).
  • 17 uncloned_repos (on GitHub, not cloned locally) — overlaps heavily with the missing list.
  • 0 stale pages, 0 orphan stubs, 0 unindexed pages.
  • Taxonomy validation: 0 missing/bad categories, 0 bad stacks. 14 article stubs still flag stubs_missing_stacks — informational, unchanged from prior runs.

Rename audit

The following GitHub repos appear to have been renamed since the prior clones were taken. Local clone directories still use the old names but their origin remotes silently redirect, so git pull works:

Old (local dir + stub repo_path) New (current GitHub) HEAD parity
MLX-YouTubeScribe youtube-scribe-mlx host behind by 1 commit
QWEN3-VL-Python-OCR-Script-MLX qwen-ocr-mlx host behind by 1 commit
N8n2langraph n8n-to-langraph in sync
strixresearch strix-research in sync
screenlens screen-lens in sync
mlx-responses-api-server (stub only) local-mlx-responsesAPI-server stub fixed this run

Two local clones (MLX-YouTubeScribe, QWEN3-VL-Python-OCR-Script-MLX) fetched the new commit objects but git pull couldn't update HEAD because the macOS host blocked the sandbox from removing .git/index.lock (Cowork mount permissions). On the host, a plain git pull resolves it. The stubs already record the up-to-date last_commit_date, so no wiki action is needed.

Sandbox limitations encountered (for future runs)

  • git clone into the mounted repos/ failed for all 17 uncloned repos with could not lock config file ... File exists. The host mount allows file creation but blocks the lock-file dance git uses during clone init. Worked around by cloning to /tmp inside the sandbox for introspection; the 17 broken half-clones the failed attempts left behind were cleaned up via allow_cowork_file_delete.
  • projectpulse (private repo) requires interactive auth; skipped.

Action items the user should run on the host

cd ~/Documents/sw30labs/repos
# Fast-forward the two stuck clones
git -C MLX-YouTubeScribe pull
git -C QWEN3-VL-Python-OCR-Script-MLX pull

# Clone the 17 repos the sandbox couldn't (all stubs already exist):
for r in OllamaPDF2Markdown AutogenMermaidGenerator AutogenDocGenerator \
         animated-GIF-Creator AutogenRequirementsAgent \
         Word-to-Markdown-Converter youtube-scribe-mlx \
         deepseekvl2-PDF-OCR-private NVD-Extractor RepoBundle \
         qwen-ocr-mlx n8n-to-langraph mlx-responses-api-server \
         strix-research screen-lens wiki-vs-rag nemotron-parse-spark; do
  git clone --depth 1 https://github.com/sw30labs/$r.git
done

Once those land, the next sync will have a real cloned working tree for introspection (today's run could only see what shipped in the existing stubs). diff a/Log.md b/Log.md (rejected hunks) @@ -273,3 +273,32 @@ no stale_pages, no orphan_stubs. Articles repo unchanged

  • The nemotron-parse-spark stub deliberately calls out the prompt / image-size landmines from the model card so future readers don't silently degrade output by tweaking either.

+--- + +## 2026-05-29 — wiki-curator v2.7 run + +New repos ingested (6): +- bonsai-image-ternary-4b-mlx-2bit → local-inference-mlx [mlx, cli] +- genai-regulatory-intel → oscal-compliance [agentic, langgraph, compliance, rag] +- lance-3b-video-bf16 → local-inference-mlx [mlx, cli] +- stable-audio-3 → local-inference-mlx [cli] +- sulphur-2-base → local-inference-mlx [mlx, cli] +- supertonic-3-mlx → local-inference-mlx [mlx, cli] + +Updated stubs (1): +- repobundle — last_commit_date bumped 2025-05-27 → 2026-04-26; GUI (tkinter, scripts/gui.py) added + +Taxonomy fix (1): +- nvd-extractor — stacks assigned [pentest, cli] (was []) + +Wiki pages updated: +- wiki/local-inference-mlx/local-inference-mlx.md — summary updated (9→14 repos), 5 new claims + sources +- wiki/oscal-compliance/oscal-compliance.md — summary updated (5→6 repos), genai-regulatory-intel claim + source +- wiki/developer-tools/developer-tools.md — RepoBundle claim updated to reflect GUI + +Other: +- Index.md: 6 new entries added +- Sitemap.md and Sitemap-Stacks.md regenerated +- Missing repos (private, expected): project-pulse-sap, deepseekvl2-PDF-OCR-private +- Articles: 22 in repo, all 22 already indexed — no new articles

2026-07-09 — wiki-curator update run

New repos ingested (8):

  • STTbench → local-inference-mlx [mlx, converter, cli] — STT cost/speed/accuracy benchmark (OpenAI vs local MLX Whisper)
  • ace-step-1.5-mlx → local-inference-mlx [mlx, cli] — text-to-song wrapper (follows local-mlx-wrapper-pattern)
  • ltx-2.3-mlx → local-inference-mlx [mlx, cli] — LTX 2.3 text/image/audio-to-video (shares ltx-2-mlx runtime with Sulphur)
  • longcat-video-avatar-1.5-mlx → local-inference-mlx [mlx, cli] — talking-avatar video wrapper
  • driftlab → oscal-compliance [oscal, compliance, agentic, langgraph] — runtime OSCAL drift observer with sandboxed micro-experiments
  • langgraph-checkpoints-vs-stores → agentic-frameworks [langgraph, agentic] — runnable offline checkpoints-vs-stores reference
  • venture-pathfinder → agentic-frameworks [langgraph, agentic, mlx] — repo-scanning venture pathfinder on Neo4j + OMLX
  • strix-omlx → ai-security [pentest, mlx, agentic] — points Strix pentest agent at a local OMLX server (companion to strix-research)

Rename resolved (1):

  • local-mlx-responsesAPI-server stub → remote_url/repo_path repointed to mlx-responses-api-server (old URL 301-redirects); stub id kept stable to preserve Index/citation links; last_commit_date corrected to 2026-03-07

Updated stubs (3):

  • tars-ai — architecture rewrite: LLM + TTS now served by a local OMLX server over the OpenAI protocol (/v1/chat/completions + /v1/audio/speech, Qwen3-TTS Voice Design), replacing in-process mlx-lm/mlx-audio; last_commit_date → 2026-07-06
  • pentest-regulatory-intel — CLI refactored into a package + periodic regulation-monitoring mode noted; last_commit_date → 2026-07-06
  • screenlens — trivial gitignore chore; last_commit_date → 2026-06-27 (no content change)

Wiki pages updated:

  • wiki/local-inference-mlx/local-inference-mlx.md — summary 15→19 repos, 4 new claims + sources, stale TARS claim refreshed to OMLX
  • wiki/oscal-compliance/oscal-compliance.md — summary 7→8 repos, driftlab claim + source
  • wiki/agentic-frameworks/agentic-frameworks.md — summary 7→9 repos, 2 new claims + sources, removed a duplicated "Open questions" block
  • wiki/ai-security/ai-security-pentesting.md — strix-omlx claim + source
  • wiki/local-inference-mlx/local-mlx-wrapper-pattern.md — +3 repos in pattern list and port table (ace-step 8789, ltx 8788, longcat 8796); also newly added to Index (was unindexed)
  • Home.md — header 43→50 repos + date; 8 new repo rows; tars-ai description de-staled; renamed server link; cross-cutting stack counts refreshed

Other:

  • Index.md: 9 new entries (8 stubs + local-mlx-wrapper-pattern page)
  • Sitemap.md and Sitemap-Stacks.md regenerated (both views)
  • Missing repos: project-pulse-sap (projectpulse stub) — remote 404s; known-private/expected per prior runs, stub retained
  • Articles: 22 in repo, all 22 already synced — no new articles; timeline updated: left unchanged per stability rule
  • Taxonomy validator: clean (0 bad category, 0 bad stacks, 0 missing category)

Post-commit cleanup (same run):

  • Normalized 3 new stub IDs to drop version dots (tooling CITATION_RE excludes ., which caused false orphans): ace-step-1.5-mlx→ace-step-1-5-mlx, ltx-2.3-mlx→ltx-2-3-mlx, longcat-video-avatar-1.5-mlx→longcat-video-avatar-1-5-mlx (remote_url/repo_path keep the real dotted repo names)
  • Fixed a pre-existing broken citation: [^src:asics-dethrone-gpus-llm] was used in the Local Inference articles list but never defined in Sources — added the missing definition
  • Regenerated both sitemaps; detect_changes now reports 0 orphans, 0 unindexed, clean taxonomy

2026-07-16 — wiki-curator update run

Updated stubs (3):

  • driftlab — v2.1.0: control family DL-1..7 → DL-1..8 (Resource Budget, SC-6/CA-7); new driftlab.sensor plane (process meter, multi-currency budget ledger, BudgetExceeded breaker) with the two-planes rule (resource telemetry on its own lossy-permitted JSONL stream, never in decision traces); enforcement delegated to platform (cgroup v2 / OMLX gateway quotas); proof engine dogfoods the sensor; Apache-2.0 since v2.0.0 RELEASE 2; tests 92 → 51 post-restructure; last_commit_date → 2026-07-15
  • screenlens — architecture rewrite: captioning moved off in-process mlx-vlm onto a local oMLX server via OpenAI-compatible /v1 (omlx_client, Ollama fallback); v0.2 reconstruction pipeline (Classify → Plan → Reconstruct → QA) rebuilds code/docs/demo artifacts; transcribe + OCR modules; idempotent Conda launcher setup_and_run.sh; mlx-vlm + huggingface-hub dropped from deps; last_commit_date → 2026-07-13
  • langgraph-checkpoints-vs-stores — chapters 2–3 landed same local day as ingest (caught by content diff, not date diff): production persistence (kill-and-resume across PIDs, SQL peek, offline semantic search, backend matrix over memory/SQLite/Postgres/Redis on compose ports 5442/6390) and HITL + time travel; backend-integration CI vs real service containers; Apache-2.0

No-op verifications:

  • oscal-skills-guardrails — 07-05Z commits (initial release, skill-gate CI, ROADMAP) already fully captured in stub
  • RepoBundle, animated-GIF-Creator, genai-regulatory-intel, qwen-ocr-mlx — pushed_at vs stub gaps are timezone skew; HEAD local dates match stubs, content verified current
  • youtube-scribe-mlx — 0 commits after stub date
  • Articles: 22 sources ↔ 22 pages, all matched via source_file; timeline untouched per stability rule

Wiki pages updated:

  • wiki/oscal-compliance/oscal-compliance.md — DL-8 certified-resource-budget claim
  • wiki/local-inference-mlx/local-inference-mlx.md — ScreenLens claim refreshed to oMLX seam + reconstruction; open question refined (TARS-AI + ScreenLens now share the OMLX pattern)
  • wiki/agentic-frameworks/agentic-frameworks.md — chapters 2–3 claim for checkpoints-vs-stores
  • Home.md — screenlens repo URL fixed (screenlens → screen-lens), 3 rows refreshed, date bump

Other:

  • Fresh-session selective sync: 9 repos re-cloned (4 flagged + 4 borderline + articles); remaining 41 screened via GitHub API pushed_at vs stub dates — no new repos
  • No pages added/removed → Index.md and sitemaps unchanged
  • Known issues carried: project-pulse-sap remote still 404 (private/expected); 23 article pages >90d freshness threshold (static content, expected); 13 article stubs without stacks (informational)

Home | Index | Sitemap | Sitemap-Stacks | SCHEMA | Log

Clone this wiki locally