Skip to content

Docs: list remaining bypass gaps as known limitations (#203)#219

Merged
twschiller merged 1 commit into
mainfrom
docs/known-limitations-203
Jun 7, 2026
Merged

Docs: list remaining bypass gaps as known limitations (#203)#219
twschiller merged 1 commit into
mainfrom
docs/known-limitations-203

Conversation

@twschiller
Copy link
Copy Markdown
Contributor

Summary

  • Audits the Audit: rule-bypass findings (red-team pass) #203 red-team checklist against merged fixes and surfaces the four items that won't be closed in code as documented limitations.
  • Rule-specific gaps are collocated with their rules; cross-cutting extension-presence note lives under Coverage scope.
  • Phrasing stays abstract per repo convention — no injection examples, no roadmap language.

Gaps documented

  • Coverage scope — extension presence is observable. Sites that fingerprint ABS artifacts (placeholders, landmarks, chips, neutralized labels) can cloak content; counter-cloaking from a content script is out of scope. (Audit ci: workflow-driven CalVer release for the extension #17.)
  • prompt-injection-redact — closed pattern bundle. Catalog is finite; novel framings pass through. Limitation propagates to the five rules sharing the bundle (meta-injection-strip, attribute-injection-sanitize, json-ld-sanitize, html-comment-strip, svg-text-strip). (Audit docs: scaffold Astro Starlight documentation site #8.)
  • countdown-timer-redact — reset-on-tick and canvas timers. Snapshot-and-confirm needs a parseable text representation whose value strictly decreases. (Audit docs: add rules reference page #20.)
  • confirmshame-sanitize — English-only, text-only. Phrase set ships in English; icon-only buttons with no accessible name are out of scope. (Audit docs: show sidebar on homepage and drop alpha warning #22.)

Items #1#7, #9#16, #18, #19, #21 are already closed by merged PRs.

Test plan

  • CI: Pre-commit hooks (mdformat / markdown lint) pass.
  • Spot-check rendered output on the Starlight site — new paragraphs render under Coverage scope and under the three rule subsections.

🤖 Generated with Claude Code

Add limitation language for the four items from the red-team audit that
aren't being closed by code fixes: closed prompt-injection pattern set,
extension-fingerprint cloaking, countdown reset/canvas timers, and
English-only / text-only confirmshame coverage. Rule-specific gaps are
collocated with the rule that owns them; the cross-cutting extension-
presence note lives under Coverage scope.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@vercel
Copy link
Copy Markdown

vercel Bot commented Jun 7, 2026

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
agent-browser-shield-demo-site Ready Ready Preview, Comment Jun 7, 2026 10:11pm

Request Review

@twschiller twschiller merged commit 5157948 into main Jun 7, 2026
7 checks passed
@twschiller twschiller deleted the docs/known-limitations-203 branch June 7, 2026 22:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant