RomanNum3ral
/
paperclip
mirror of https://github.com/paperclipai/paperclip.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix(e2e): remove all readable preview text

This commit is contained in:
Dotta 2026-09-04 13:39:26 -05:00
parent a4be68bae0
commit cadd50c566
3 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/runner-e2e/README.md
View File

@ -207,8 +207,8 @@ cost. The CI report job stages the same portable site at
Permanent public history has a narrower boundary. A trusted job without
provider or AWS credentials copies successful declared PNG screenshots into a
separate publication tree. It validates the PNG container, reduces each image
to at most 160 pixels on either edge, applies a strong blur, limits the image to
24 colors, and removes metadata. It then runs OCR and rejects the whole public
to at most 96 pixels on either edge, applies a 3.5-sigma blur, limits the image
to 16 colors, and removes metadata. It then runs OCR and rejects the whole public
bundle if any letter or digit remains readable. These layout previews preserve
coarse UI state while making rendered task and provider text unreadable. The job then
removes the full-resolution raster files, failure screenshots, video, archives,

6
tests/runner-e2e/SECURITY.md
View File

@ -197,9 +197,9 @@ packaging because it is active content.
Before permanent publication, the trusted report job creates a separate tree.
It accepts only declared screenshots from passing results. It validates a
bounded, non-interlaced PNG container before invoking ImageMagick with strict
memory, disk, thread, and time limits. It reduces each image to at most 160
pixels on either edge, applies a strong blur, limits the palette, removes alpha
and metadata, and validates the new PNG again. Tesseract must then find no
memory, disk, thread, and time limits. It reduces each image to at most 96
pixels on either edge, applies a 3.5-sigma blur, limits the palette to 16
colors, removes alpha and metadata, and validates the new PNG again. Tesseract must then find no
readable letter or digit. OCR output is never logged. Any readable text or OCR
failure blocks the whole publication. This is a layout preview, not diagnostic
evidence. The job then prunes full-resolution raster files, all failure images,

6
tests/runner-e2e/history-public-bundle.ts
View File

@ -31,7 +31,7 @@ const PNG_SIGNATURE = Buffer.from([
const MAX_PRIVATE_PNG_BYTES = 32 * 1024 * 1024;
const MAX_PRIVATE_PNG_PIXELS = 64 * 1024 * 1024;
const MAX_PRIVATE_PNG_EDGE = 32 * 1024;
const MAX_PUBLIC_PREVIEW_EDGE = 160;
const MAX_PUBLIC_PREVIEW_EDGE = 96;
const MAX_PUBLIC_PREVIEW_BYTES = 256 * 1024;
const PRIVATE_PNG_CHUNKS = new Set([
"IHDR",
@ -320,9 +320,9 @@ export async function createPublicLayoutPreview(
"-resize",
`${MAX_PUBLIC_PREVIEW_EDGE}x${MAX_PUBLIC_PREVIEW_EDGE}>`,
"-blur",
"0x2.5",
"0x3.5",
"-colors",
"24",
"16",
"-strip",
"-define",
"png:exclude-chunks=all",