Paper Poster (HTML): measurement-gated poster generation
Unverified●36/40Claude Code●WorksValid SKILL.md that declares allowed-tools
Cursor·UnknownWe have not crawled the repo tree, so we will not guess
Codex·UnknownWe have not crawled the repo tree, so we will not guess
Gemini CLI·UnknownThe spec defines no detection rule for Gemini
Copilot·UnknownWe have not crawled the repo tree, so we will not guess
npx agentalley add paper-poster-htmlWho is stuck, and on what
DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the
The whole source
Frontmatter — 4 properties
| name | paper-poster-html |
|---|---|
| description | DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the user says \"做海报\", \"poster\", \"conference poster\", \"paper poster\", or asks to design/redo a research poster. Supersedes the retired LaTeX /paper-poster. |
| argument-hint | [paper-dir-or-pdf] [— venue: ICLR, canvas: 185x90cm landscape, venue-colors: true] |
| allowed-tools | Bash(*), Read, Write, Edit, Grep, Glob, WebFetch, WebSearch, AskUserQuestion, mcp__codex__codex |
| 1 | --- |
| 2 | name: paper-poster-html |
| 3 | description: "DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the user says \"做海报\", \"poster\", \"conference poster\", \"paper poster\", or asks to design/redo a research poster. Supersedes the retired LaTeX /paper-poster." |
| 4 | argument-hint: "[paper-dir-or-pdf] [— venue: ICLR, canvas: 185x90cm landscape, venue-colors: true]" |
| 5 | allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, WebFetch, WebSearch, AskUserQuestion, mcp__codex__codexA4 — This skill pulls in web or user content but never says to treat that content as data. A signal, not proof. |
| 6 | --- |
| 7 | |
| 8 | # Paper Poster (HTML): measurement-gated poster generation |
| 9 | |
| 10 | One HTML file styled for an exact print canvas (`@page { size: W H }`), rendered to PDF |
| 11 | via Playwright print emulation. **Iterate by measuring, not eyeballing** — the screen |
| 12 | preview lies; only print emulation at the correct viewport tells the truth. Core gate |
| 13 | machinery is adapted from [posterly](https://github.com/Chenruishuo/posterly) (MIT, © |
| 14 | 2026 Ruishuo Chen — see `NOTICE.md` and `LICENSES/posterly-MIT.txt`); ARIS adds style |
| 15 | discipline gates, figure-provenance gates, the cross-model review loop, and the |
| 16 | anti-patch-loop fix vocabulary. |
| 17 | |
| 18 | ## Why this skill exists (the failure it prevents) |
| 19 | |
| 20 | A predecessor pipeline produced a poster with **30+ colors, zero real paper figures, a |
| 21 | screen-pixel canvas, and tiny formulas floating in oversized boxes**, then spent 12+ |
| 22 | review rounds making it *worse* — each round added a new badge color or bespoke SVG |
| 23 | patch. The cure is structural, not exhortative: |
| 24 | |
| 25 | 1. **Hard gates run before any aesthetic opinion** (alignment, style, assets must PASS |
| 26 | first — a reviewer never sees an unmeasured poster). |
| 27 | 2. **A closed fix vocabulary** — visual-review fixes can only touch design tokens, |
| 28 | whole catalogued components, content rebalance, assets, or canvas choice. New inline |
| 29 | styles / new hex values / bespoke decorations are structurally forbidden. |
| 30 | 3. **Two-hue discipline as a machine check**, not a style suggestion. |
| 31 | 4. **Real paper figures with provenance manifest**, or the gate fails. |
| 32 | |
| 33 | ## Mental model |
| 34 | |
| 35 | ``` |
| 36 | paper (.tex / PDF) ──► content plan + claim→evidence audit (codex, fresh) |
| 37 | │ |
| 38 | figures extracted ─────────┤ FIGURE_MANIFEST.json (provenance, sha256) |
| 39 | (real paper figures ONLY) ▼ |
| 40 | template scaffold ──► fill ──► run_gates.py ◄─── HARD, loop here |
| 41 | preflight → style → asset → measure → polish |
| 42 | │ all hard gates PASS |
| 43 | ▼ |
| 44 | Claude visual review (≤3 issues × ≤3 rounds, fix-vocabulary only) |
| 45 | │ score ≥ 9 |
| 46 | ▼ |
| 47 | codex final cross-model review (fresh thread, full HTML+PDF) |
| 48 | │ pass |
| 49 | ▼ |
| 50 | verify-final → poster.pdf + GATE_REPORT.json |
| 51 | ``` |
| 52 | |
| 53 | ## Constants |
| 54 | |
| 55 | - **SKILL_SCRIPTS** = `${CLAUDE_SKILL_DIR}/scripts` — all helpers are single-owner and |
| 56 | ship inside this skill (Arch C). If the directory is missing the install is broken: |
| 57 | abort and tell the user to re-install the skill (Policy A — the gates ARE the skill; |
| 58 | never improvise replacements). |
| 59 | - **REVIEWER_MODEL** = `gpt-6-astra`, reasoning `xhigh`, **fresh thread per review call** |
| 60 | (`mcp__codex__codex`, never `codex-reply` across review boundaries). |
| 61 | - **CANVAS** — from the venue's official spec, looked up live in Phase 0. Never assume. |
| 62 | (Known anchor: ICLR 2026 main = 185×90 cm landscape per its official printing |
| 63 | service; ICML/NeurIPS commonly 60×36 in landscape; workshop posters often 61×91 cm |
| 64 | portrait. Specs change yearly — verify.) |
| 65 | - **PALETTE** — default = `templates/tokens/generic.json` (slate-blue `#2D5F8B` accent |
| 66 | + gold `#C9A24A` highlight + neutrals) for **all** venues. Venue packs are opt-in via |
| 67 | `— venue-colors: true`. Purple-dominant accents (hue 250–285) are banned unless the |
| 68 | user passes `— allow-purple: true`. |
| 69 | - **AUTO_PROCEED = false** — wait for explicit confirmation at every 🚦 checkpoint. |
| 70 | - **OUTPUT_DIR** = `poster_html/` in the working directory. |
| 71 | |
| 72 | ## Workflow |
| 73 | |
| 74 | ### Phase 0 — Resume, dependencies, venue spec |
| 75 | |
| 76 | 1. **Resume**: if `poster_html/POSTER_STATE.json` exists with `status: in_progress` |
| 77 | (< 24 h), resume from the saved phase. |
| 78 | 2. **Dependencies** (degradation chain, in order): |
| 79 | - Playwright + bundled Chromium → if missing, `python3 -m playwright install |
| 80 | chromium` → if install fails but system Chrome exists, scripts fall back to |
| 81 | `channel="chrome"` → if all fail: you may produce the content plan and scaffold |
| 82 | only, label everything **"not print verified"**, and must NOT emit a final PDF. |
| 83 | - `pdfinfo` missing → PyMuPDF reads PDF dimensions. At least one of |
| 84 | pdftoppm / PyMuPDF must exist for PNG review renders. |
| 85 | - MathJax: download `tex-svg.js` once into `poster_html/assets/mathjax/` and |
| 86 | reference it locally in the HTML. CDN is acceptable only for drafts; the measure |
| 87 | gate hard-fails on unrendered MathJax either way. |
| 88 | 3. **Venue spec lookup (live)**: consult the venue's official poster-instructions page |
| 89 | (search + fetch). Extract dimensions, orientation, font floor, logo policy, |
| 90 | anonymity rules, file format. Record `{spec, source_url, retrieved}` into |
| 91 | `POSTER_STATE.json` — specs change yearly; never reuse a cached spec silently. |
| 92 | |
| 93 | **🚦 Checkpoint**: echo the venue spec table (canvas, orientation, source URL) and the |
| 94 | chosen template. Wait. |
| 95 | |
| 96 | ### Phase 0.5 — Design discovery (one AskUserQuestion batch) |
| 97 | |
| 98 | Ask once, ≤4 questions: layout template (from `templates/README.md`), palette |
| 99 | (default generic pack / venue pack / custom within constraints), logos + venue mark |
| 100 | (paths or "none" — never fabricate; check the venue's logo policy), QR target (paper / |
| 101 | code / project page / none — generate **offline** with `qrencode` or python-`qrcode`; |
| 102 | never a remote QR-service URL). Persist answers in `POSTER_STATE.json` as |
| 103 | `design_decisions` — re-read before any later "improvement" so deliberate choices are |
| 104 | never reverted. |
| 105 | |
| 106 | ### Phase 1 — Paper ingest, content plan, claim audit |
| 107 | |
| 108 | 1. Read the paper source (`.tex` ideal; PDF otherwise). Extract: title/authors/affils, |
| 109 | the 3–5 headline numbers, core method (equations verbatim), main results |
| 110 | (tables/figures and what they show), takeaways. Build |
| 111 | `poster_html/POSTER_CONTENT_PLAN.md` — what goes in which column, word budget per |
| 112 | card. **Target density** (excluding table cells, captions, author line, footer): |
| 113 | standard poster **550–850 words**; dense theory+empirical poster **750–1050 words**, |
| 114 | allowed only when ≥2 compact components are used (`eqn-anatomy`, `flow-strip`, |
| 115 | `derived-col`, `claim-pills`, `keybox--4`). Warn yourself below 500 words on a |
| 116 | 4-column landscape (it will read as sparse next to professionally dense posters) |
| 117 | unless the template is hero/visual-first; warn above 1100 unless the user asked for |
| 118 | dense mode. Bullets ≤ 8 words when possible — density comes from *structure*, not |
| 119 | long prose. **Prefer compact structure over prose**: if the paper contains an |
| 120 | explicit objective, algorithm, theorem mechanism, or baseline comparison, extract at |
| 121 | least two of: (1) empirical objective / loss stack; (2) term-by-term equation |
| 122 | anatomy; (3) a method-flow strip grounded in paper variables; (4) a derived-Δ column |
| 123 | for method-vs-baseline rows; (5) a 4-up implementation/theory keybox; (6) a |
| 124 | claim/evidence pill table for numeric-heavy posters. **Do not invent an algorithm.** |
| 125 | If the paper has only an objective, label the component "objective flow" or "loss |
| 126 | anatomy", never "algorithm". |
| 127 | 2. **Cross-model content audit** (fresh codex thread, `xhigh`): give it the content |
| 128 | plan path + paper source path(s) — paths only, no summaries — and ask for a |
| 129 | claim→evidence table: `| claim on poster | paper file:line | paper says (verbatim) | |
| 130 | match? |` with match ∈ {OK, NUMERIC-MISMATCH, OVERCLAIM, MISSING-PRECONDITION, |
| 131 | NOT-IN-PAPER, SCOPE-NARROWED}. Save to `poster_html/CLAIM_EVIDENCE.md`. |
| 132 | 3. Fix every non-OK row or record it as a user-acknowledged tradeoff. |
| 133 | |
| 134 | **🚦 Checkpoint**: content plan + audit summary. Wait. |
| 135 | |
| 136 | ### Phase 2 — Real paper figures (provenance-gated) |
| 137 | |
| 138 | Source preference chain: |
| 139 | 1. Paper source `figures/` (vector SVG/PDF → convert to SVG via |
| 140 | `inkscape`/`pdf2svg` if available, else rasterize ≥ 2× rendered px). |
| 141 | 2. PDF-only: `extract_pdf_figures.py contact-sheet` + `auto` to list candidate |
| 142 | regions → pick crops (**🚦 human confirms crop choices**) → `crop` at 300–450 DPI. |
| 143 | 3. Last resort: user supplies explicit `page,x0,y0,x1,y1` bboxes. |
| 144 | |
| 145 | Then `preprocess_figures.py --autocrop` every asset. Every paper-derived image gets a |
| 146 | `FIGURE_MANIFEST.json` entry (source hash, page, bbox, dpi, sha256, natural_px) and is |
| 147 | embedded as `<img data-source="paper" data-asset-id="...">`. |
| 148 | |
| 149 | **Hard rule**: ≥ 2 paper-derived visuals or the asset gate fails. Theory-only papers |
| 150 | may waive the *total-area* rule (`--waive-total-area`) at a human checkpoint — never |
| 151 | silently. Never draw bespoke decorative SVG "figures" as substitutes. |
| 152 | |
| 153 | **Figure-area bands** (asset gate, fractions of *body*): total target **14–22 %** |
| 154 | (warn < 12 % / > 24 %, hard < 10 % / > 28 %); per ordinary figure target 4–8 % (warn |
| 155 | > 10 %, hard > 13 %); `figure--duo` combined 8–12 %. Hero templates pass `--hero` |
| 156 | (centerpiece may take 30–40 %). The failure mode is symmetric: too small reads as |
| 157 | decoration, too big crowds out content. Sibling figures that share axes or tell a |
| 158 | before→after story belong in one `figure--duo` card, not two cards. |
| 159 | |
| 160 | ### Phase 3 — Scaffold + tokens |
| 161 | |
| 162 | `cp templates/<chosen>.html poster_html/poster.html`; retarget `@page` + `.poster` |
| 163 | dims to the venue canvas (two edits, same values); apply the chosen token pack onto the |
| 164 | `:root` DESIGN TOKENS block; fill content per the plan; embed manifest figures. |
| 165 | Run `preflight` + `style_check` — both must PASS before any layout iteration. (A fresh |
| 166 | scaffold is *expected* to fail `measure` — that gate judges a filled poster.) |
| 167 | |
| 168 | ### Phase 4 — Layout hard loop |
| 169 | |
| 170 | After every layout change: |
| 171 | |
| 172 | ```bash |
| 173 | python3 "$SKILL_SCRIPTS/run_gates.py" poster_html/poster.html \ |
| 174 | --tokens <pack.json> --manifest poster_html/FIGURE_MANIFEST.json \ |
| 175 | --report poster_html/GATE_REPORT.json |
| 176 | ``` |
| 177 | |
| 178 | Canonical order: preflight → style → asset → measure → polish. Targets: column-bottom |
| 179 | **spread < 5 px** (aim < 3), footer gap ∈ [30, 50] px, intercard gap ∈ [12, 50] px, |
| 180 | canvas-fill ∈ [95, 101] %, poster bbox aligned to page within ±2 px. Fix guidance for |
| 181 | each failure mode lives in the gate output and `templates/COMPONENTS.md`. **Do not |
| 182 | proceed while any hard gate fails. Do not let a reviewer see an unmeasured poster.** |
| 183 | Balance under-filled columns with *content from the paper* (Gate C), never with |
| 184 | whitespace, `space-between`, or stretched cards. |
| 185 | |
| 186 | ### Phase 5 — Claude visual review (gated aesthetics) |
| 187 | |
| 188 | Render and read the result yourself: |
| 189 | |
| 190 | ```bash |
| 191 | python3 "$SKILL_SCRIPTS/render_preview.py" poster_html/poster.html |
| 192 | pdftoppm -r 100 poster_html/poster_preview.pdf poster_html/review_full -png -f 1 -l 1 |
| 193 | # plus 2-4 region crops at higher res (header / one column / equations) via PIL |
| 194 | ``` |
| 195 | |
| 196 | **Calibrate first** (`../shared-references/taste-calibration.md`): if |
| 197 | **human-curated** `references/good/` + `references/bad/` exist under this skill |
| 198 | dir (or the project supplies its own pair), score those 3+3 reference posters |
| 199 | on the axes below BEFORE the target, anchoring the scale. Never select, search |
| 200 | for, or generate anchors yourself; if no reference sets exist, proceed |
| 201 | uncalibrated and mark `CALIBRATION: none` — never fabricate anchor scores. |
| 202 | Axes (weights sum 1.0): Design 0.35 · Craft 0.30 · Functionality 0.20 · |
| 203 | Originality 0.15. Mapping: `SCORE = min(round(1 + 9 × COMPOSITE), lowest |
| 204 | triggered cap)` — caps apply AFTER the mapping, and the loop's `Score ≥ 9` |
| 205 | threshold below always reads this final capped `SCORE`, never the raw |
| 206 | composite. |
| 207 | |
| 208 | Score strictly 1–10. **Critical caps** (hard floors — a calibrated composite |
| 209 | never overrides them): < 2 real paper figures → ≤ 3; broken canvas / |
| 210 | clipped content / unreadable math → ≤ 4; ≥ 4 visible hue families or gradient-heavy |
| 211 | header → ≤ 4; large blank cards or columns → ≤ 5; fabricated visual claim → ≤ 3. |
| 212 | Checks: posterly-showcase gestalt (would this hang next to a professionally designed |
| 213 | poster without looking like a patched dashboard?), single-accent discipline, real |
| 214 | figures readable and central, print hierarchy (title → headline stats → figures → |
| 215 | detail), column fill, **equation prominence** (no tiny math in oversized boxes), |
| 216 | serif-body/sans-display pairing, no gradient kitsch, component consistency, 60-second |
| 217 | narrative. Output format: |
| 218 | |
| 219 | ``` |
| 220 | SCORE: N/10 (= min(round(1 + 9 × COMPOSITE), lowest cap); drives the loop) |
| 221 | COMPOSITE: 0.xx (weighted; list the four per-axis scores) |
| 222 | CALIBRATION: anchored | none |
| 223 | GAP: <which reference poster the target falls short of / exceeds, on which axis, and why — one paragraph; omit only when CALIBRATION: none> |
| 224 | CAPS_TRIGGERED: ... |
| 225 | TOP_ISSUES: (max 3) |
| 226 | ALLOWED_FIX_TYPE per issue: token | component | rebalance | asset | template/canvas |
| 227 | PATCH_LOOP_RISK: low | medium | high |
| 228 | ``` |
| 229 | |
| 230 | Loop: fix (fix vocabulary below) → re-run Phase 4 gates → re-score. **≤ 3 issues per |
| 231 | round, ≤ 3 rounds.** Score ≥ 9 → Phase 6. Still < 9 after 3 rounds → STOP patching; |
| 232 | escalate to template / canvas / content re-choice (back to Phase 3) or a human |
| 233 | decision. Never enter round 4 of cosmetic patching. |
| 234 | |
| 235 | #### Fix vocabulary (closed set — the anti-patch-loop core) |
| 236 | |
| 237 | Allowed: **(a)** edit a `:root` token value; **(b)** swap/remove/add a whole component |
| 238 | instance from `templates/COMPONENTS.md`; **(c)** content rebalance (move a card across |
| 239 | columns, trim/grow text *from the paper*, resize a figure within its AR band); |
| 240 | **(d)** template/canvas re-choice; **(e)** global edits to an existing component's CSS |
| 241 | that reference only tokens; **(f)** switching predefined variants (`.eqn--large`, |
| 242 | `.card--compact`, `.figure--wide`, `.nowrap`, …); **(g)** asset fixes (re-crop, swap |
| 243 | for a clearer figure from the same paper, re-preprocess). |
| 244 | |
| 245 | Forbidden: new inline styles, new hex values anywhere, bespoke decorative SVG, |
| 246 | per-element font-size overrides. **A new component may not be born inside the visual |
| 247 | loop** — stop, get a human checkpoint, add it to `COMPONENTS.md`, re-run from Phase 3. |
| 248 | |
| 249 | ### Phase 6 — Codex final review (fresh thread, cross-model) |
| 250 | |
| 251 | All hard gates PASS + polish warnings zero-or-waived + visual ≥ 9 first. Then a fresh |
| 252 | codex thread (`xhigh`) reviews the **final artifacts** (not the content plan): |
| 253 | `poster.html`, the rendered PDF/PNG, the paper source, `GATE_REPORT.json`, |
| 254 | `CLAIM_EVIDENCE.md` — paths only, no executor framing. It checks: (1) fidelity & |
| 255 | overclaims **re-checked on final text** (polish introduces new claims), (2) residue |
| 256 | (`\ref{`, `TODO`, raw `<` in math, missing images, remote URLs), (3) visual rhetoric |
| 257 | (headline numbers prominent, banner readable from 2 m), (4) gate-log coherence. The |
| 258 | reviewer recommends; it does not edit. Any fix → back through Phase 4/5 gates — never |
| 259 | straight to re-review. |
| 260 | |
| 261 | ### Phase 7 — Final verification + report |
| 262 | |
| 263 | ```bash |
| 264 | python3 "$SKILL_SCRIPTS/poster_check.py" verify-final poster_html/poster_preview.pdf \ |
| 265 | --from-html poster_html/poster.html --max-size-mb 20 |
| 266 | ``` |
| 267 | |
| 268 | Page count 1, dimensions match `@page`, size ≤ 20 MB, no TODO/residue, no remote |
| 269 | assets. Report: PDF path, final spread px, footer-gap range, gate summary table, |
| 270 | unresolved waivers, codex verdict. Update `POSTER_STATE.json` → `done`. |
| 271 | |
| 272 | ## State persistence |
| 273 | |
| 274 | `poster_html/POSTER_STATE.json`: `{phase, venue, canvas{w,h,orientation,source_url, |
| 275 | retrieved}, template, token_pack, design_decisions{...}, figures_selected[], |
| 276 | visual_rounds, codex_threads{audit, final}, status, timestamp}` — written after every |
| 277 | phase; enables compact-recovery resume. |
| 278 | |
| 279 | ## Key rules |
| 280 | |
| 281 | - **Measure, don't eyeball.** No layout claim without `run_gates.py` output. |
| 282 | - **Gates before aesthetics.** Claude/codex review only ever sees a poster whose hard |
| 283 | gates PASS. This ordering is what kills the patch-loop death spiral. |
| 284 | - **Never invent paper numbers or figures.** Numbers come from the paper source; |
| 285 | visuals carry manifest provenance. Fabrication = critical cap ≤ 3. |
| 286 | - **Two hues, one system.** Accent + gold + neutrals. The style gate enforces it; |
| 287 | don't negotiate with the gate. |
| 288 | - **Real figures are the poster.** A poster without the paper's own figures is a |
| 289 | dashboard, not a poster. |
| 290 | - **Fix vocabulary is closed.** If a fix isn't expressible as token / component / |
| 291 | rebalance / asset / canvas, it's the wrong fix. |
| 292 | - **Cross-model verdicts.** Claude drives the loop and scores visuals; acceptance of |
| 293 | content fidelity comes from the fresh codex thread (a loop can drive, never acquit). |
| 294 | - **Preserve user decisions.** Re-read `design_decisions` before "improving" anything. |
| 295 | - **Vendor boundary.** `poster_check.py`, `render_preview.py`, `_posterly/` are |
| 296 | vendored from posterly — keep diffs minimal; ARIS-side logic goes in the new |
| 297 | scripts, not in vendored files. |
| 298 | |
| 299 | ## Review tracing |
| 300 | |
| 301 | Save every codex reviewer call's trace per `shared-references/review-tracing.md` to |
| 302 | `.aris/traces/paper-poster-html/<date>_run<NN>/` (audit + final threads, raw responses). |
| 303 | |
| 304 | ## Output contract |
| 305 | |
| 306 | ``` |
| 307 | poster_html/ |
| 308 | ├── poster.html # single-file source of truth |
| 309 | ├── poster_preview.pdf # print-emulated, verify-final-checked |
| 310 | ├── poster_preview.png # thumbnail |
| 311 | ├── POSTER_STATE.json # resume state |
| 312 | ├── GATE_REPORT.json # canonical gate ledger (schema v1) |
| 313 | ├── POSTER_CONTENT_PLAN.md # what-goes-where + word budgets |
| 314 | ├── CLAIM_EVIDENCE.md # codex claim→evidence audit |
| 315 | ├── FIGURE_MANIFEST.json # figure provenance (sha256, page, bbox, dpi) |
| 316 | └── assets/{paper_figures,logos,qr,mathjax}/ |
| 317 | ``` |
| 318 | |
| 319 | ## When NOT to use |
| 320 | |
| 321 | - Slides, not a poster → `/paper-talk` / `/slides-polish`. |
| 322 | - The paper's headline isn't stable yet — fix the paper first; a poster amplifies |
| 323 | whatever story it's given. |
| 324 |
Reviews
Installed this one?Write the first review and take the Trailblazer badge.
Alternatives
Brand Monitoring 📡Brand monitoring tool for tracking mentions across social media platforms. Monitor Reddit, Google News, YouTube, and DuckDuckGo for brand mentions. Includes sentiment analysis, trend tracking, crisis detection, and competitor comparison. No API key required fo◐····●34/40Spark Memory & Thermal OpsManage unified memory and thermals during long-running ML jobs on NVIDIA DGX Spark. Use when planning memory headroom for a training run on GB10, when a job OOMs on unified memory, or when monitoring temperature and power during multi-hour training.◐····●32/40Secrets ManagementImplement secure secrets management for CI/CD pipelines using Vault, AWS Secrets Manager, or native platform solutions. Use when handling sensitive credentials, rotating secrets, or securing CI/CD environments.◐····●32/40Deployment Pipeline DesignDesign multi-stage CI/CD pipelines with approval gates, security checks, and deployment orchestration. Use this skill when designing zero-downtime deployment pipelines, implementing canary rollout strategies, setting up multi-environment promotion workflows, or debugging failed deployment gates in CI/CD.◐····●31/40