Use when creating any AccuKnox external-facing asset (blog, white paper, datasheet, one-pager, deck, social post, landing copy). Tells you where the images live and how to pick the right one, how to build correct help-docs and blog links, the brand colors/fonts/templates to use, and when to scrape with Firecrawl. Trigger when the task is "write/build/draft a blog / white paper / deck / one-pager / datasheet for AccuKnox".
Use when creating any AccuKnox external-facing asset (blog, white paper, datasheet, one-pager, deck, social post, landing copy). Tells you where the images live and how to pick the right one, how to build correct help-docs and blog links, the brand colors/fonts/templates to use, and when to scrape with Firecrawl. Trigger when the task is "write/build/draft a blog / white paper / deck / one-pager / datasheet for AccuKnox".
trigger
/ak-asset
AccuKnox asset kit
Everything you need to attach the right links and images to an AccuKnox
asset, with the brand details and source locations spelled out. Read this before
writing a blog, white paper, deck, datasheet, or one-pager so you don't guess at
URLs or hunt blindly through image folders.
Repo root for all paths below: D:\Atharva\AccuKnox\HelpDocs.
1. Links
Help docs: help.accuknox.com
Every published help page is one entry in mkdocs.yml under nav:. Build the
URL from the file path:
Base: https://help.accuknox.com/
The site uses directory URLs: drop the .md, add a trailing slash.
Some nav entries are already full URLs (e.g. the Compliance matrix points to
https://accuknox.com/compliance). Use those verbatim, don't reconstruct them.
To find a page: open mkdocs.yml, search the nav for the topic, or read the
folder-index docs/<section>/README.md (it lists every page and what it covers).
Fetch a child sitemap (WebFetch or the firecrawl-scrape skill) and filter the
<loc> URLs by slug. Verify a URL resolves before putting it in an asset, because
slugs change.
High-value AI-security / compliance blog URLs (verified from the live sitemap,
June 2026):
Curated demo screenshots in utils/brian-demo-screenshots/
22 clean product screenshots (dashboard, findings, AI red teaming, prompt
firewall, runtime protection, zero-trust discovery). Captions and a
screenshot index are in utils/brian-demo-screenshots/DEMO-SUMMARY.md.
Use these first for polished AI-security assets. They are consistent and
already captioned.
Per-feature doc screenshots in docs/<section>/images/<topic>/
The pictures used on the live help pages. Find the right one without opening
every PNG: each image folder has a generated README.md with a "Shows"
column describing each file (pulled from the alt text on the page that uses
it). Example: docs/use-cases/images/aidr/README.md,
docs/use-cases/images/modelarmor/README.md.
Icons in docs/use-cases/icons/*.svg (e.g. AIML.svg, aidr.svg,
modelarmor.svg, model-safety.svg, mcp-security.svg, zt-security.svg),
plus docs/<section>/icons/.
Logos in docs/assets/images/: logo-white.png, logo-black.png,
ak-logo.png, logo.png, web-logo-dark-back.png.
Diagrams and flowcharts, not only dashboards. Two good sources:
Docs diagrams: e.g. docs/use-cases/images/aidr/1.png (AI-DR workflow),
docs/use-cases/images/ai-overview/1.png (managed vs on-prem AI map),
docs/use-cases/images/modelarmor/ (adversarial diagrams).
The AI Security macro deck: utils/technical-reference/AccuKnox AI Security _ Macro Deck _ June_2026.pptx. Its flowcharts are native slides, so
render them: convert the pptx to PDF with LibreOffice, rasterise pages with
PyMuPDF (fitz), search slide text for the topic, and crop the slide you
want. Strong ones include the stateful inspection pipeline and the AI-SPM
architecture.
Workflow to pick an image
Name the topic (e.g. "AI red teaming results").
Find the doc: mkdocs.yml nav → or docs/<section>/README.md.
Open the topic's docs/<section>/images/<topic>/README.md and pick by the
"Shows" caption. For AI-security assets, check DEMO-SUMMARY.md first.
Copy the chosen image into the asset's own folder. Don't hot-link into the
docs tree, and don't move originals.
Caption rule (always)
Every image in a blog and every image in a white paper gets a caption.
Blogs: caption directly under the image. White papers: images go in the
appendix, each with a figure number and caption. No uncaptioned images.
Make screenshots clean
Product screenshots usually carry junk. Fix it before use:
Crop the browser chrome. Remove the tab strip, the URL bar, and the
bookmarks bar. Cut just above the AccuKnox app header (the logo and
breadcrumb) so the image starts at the product, not the browser.
Trim dead whitespace. If a screenshot has a large empty area at the
bottom or right (a short form, a small diagram on a big canvas), crop it so
the content fills the frame. Keep the top app header and the left nav, only
cut empty space, and leave a small even margin.
Prefer dense screenshots. A sparse diagram on a mostly empty canvas reads
as filler. Pick the view that fills the frame, or crop tightly to the part
that matters.
Vary the imagery. Do not make every image a dashboard. Use flow and
architecture diagrams (source 5 above) for conceptual points such as a
firewall pipeline, an AI-DR workflow, or the discovery footprint, and reserve
console screenshots for showing the product itself.
Quick approach (PIL on py -3.11): find the bottom-most row and right-most
column that still hold content (ignore the left nav and the right scrollbar),
then crop to that plus a ~20px margin. Work on copies, never the originals in
utils/brian-demo-screenshots/ or docs/.
3. Blog deliverable spec
When the task is a blog, it ships as a finished, SEO-ready post, not a working
draft. Every blog has:
Front matter (YAML) with all of:
title:"Punchy display headline"seo_title:"Keyword-first title, <= 60 characters"meta_description:"One sentence, 150 to 160 characters, includes the primary keyword"slug:"lowercase-hyphenated-slug"url:"https://accuknox.com/blog/<slug>"primary_keyword:"the one phrase the post targets"secondary_keywords: ["supporting", "related", "phrases"]
cover_image_prompt_claude:>
A descriptive, natural-language prompt for an image model: subject, style,
AccuKnox palette (#11206D / #003BF6), composition, negative space, no text.
cover_image_prompt_midjourney:>
A keyword-dense Midjourney prompt for the same cover, ending with parameters
such as --ar 16:9 --style raw --v 6 --no text
Body rules:
Punchy headings. Each H2 is a sharp statement, not a label. "You can't
govern a model you can't see" beats "Model inventory".
Links live in the text. Hyperlink help-docs and blog references inline, on
the words that earn them. Do not dump a "further reading" list at the bottom.
Inlined links are what make the post SEO-rich.
Read like a final blog. No internal scaffolding: no source-regulation
paragraph numbers, no "this page", no "(para 60)", no "X asks / AccuKnox
answers" table voice. State the point in prose.
Captioned images, cleaned per the rules above, spread through the post.
Keep to the length the brief sets (default under 1500 words) and the
CLAUDE.md style rules (no dashes, plain words).
White papers are different: there, paragraph references to the source regulation
are wanted, because the document's job is a formal clause-by-clause map.
4. Brand (for designed assets: DOCX, PPTX, HTML)
Templates in utils/doc-ppt-template/:
WORD_TEMPLATE_ACCUKNOX.docx, AccuKnox_Proposal_Template_BLANK.pptx.
Colors (hex, from the real templates):
Use
Hex
Primary navy (headers, badges)
11206D
Deep (dark panels, title bg)
0000A0
Accent purple-blue (CTAs, bars)
4D4DD9
Accent light
5C5CFF
Problem / alert red
C80019
Body text on white
0D1B4B
Caption / secondary text
595959
Fonts: Space Grotesk (headings/labels), Inter (body). Both embedded in the
templates.
PPTX builds (Windows):python-pptx is on py -3.11 only. Render with
PowerPoint COM or LibreOffice. pdftoppm is not installed, so extract PDFs with
PyMuPDF (fitz). (See the accuknox-pptx-build memory for the full recipe.)
5. When to scrape (Firecrawl)
Use the firecrawl-* skills or WebFetch when the repo doesn't have what you need:
Live blog/white-paper content or fresh URLs → fetch from the sitemaps above.
A competitor page, an analyst stat, a news item → firecrawl-search /
firecrawl-scrape.
Bulk-pulling a docs section → firecrawl-crawl / firecrawl-map.
Don't scrape what's already in the repo (docs markdown, screenshots, templates).
6. Pre-ship checklist
Every help-docs link built from mkdocs.yml and resolves on help.accuknox.com.
Every blog link taken from the live sitemap and verified.
A blog has the full SEO front matter (seo_title, meta_description, slug, url, primary/secondary keywords, both cover prompts).
Blog reference links are inlined in the body, not dumped at the bottom; headings are punchy; no paragraph-number scaffolding.
Screenshots have the browser chrome cropped and dead whitespace trimmed.
Every image has a caption (appendix figures for white papers).
Designed assets use the brand colors and fonts above.
Claims are backed by a doc, the demo summary, or a cited source, with no invented features.
Writing follows the no-AI-tells style rules in CLAUDE.md (no dashes, no "leverage/ensure/robust/seamless", plain sentences).
