Skip to content

ContextPort 2.7.1 - Safer AI Context Capture

Choose a tag to compare

View all tags
@DeerSpotter DeerSpotter released this 06 Jul 02:31
· 216 commits to main since this release
2.7.1
3b4d4d6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

ContextPort 2.7.1

Safer AI Context Capture

This release hardens Save Context across supported AI providers and fixes a real Claude export failure discovered using ContextPort's own Developer Mode source archive.

ContextPort now prioritizes Memory integrity over blindly completing an export.

What's Changed

Provider aware conversation extraction

Save Context now identifies conversation turns using provider specific UI evidence instead of treating broad page containers as chat messages.

  • ChatGPT uses explicit message author roles.
  • Claude uses its standard chat message structure to positively identify user prompts and Claude responses.
  • Grok uses explicit user and assistant message identifiers.
  • Gemini remains fail closed when ContextPort cannot safely prove message roles.

No more guessed message authors

The old exporter could infer alternating messages as:

You → AI → You → AI

based only on DOM order.

That behavior has been removed.

ContextPort now requires positive evidence that a message belongs to the user or AI.

Protection against corrupted Memory

A real Claude export exposed a case where page navigation content was saved as a user message and Cloudflare challenge JavaScript was saved as a Claude response.

ContextPort now rejects known challenge and interstitial content before it can enter Memory.

Structured export validation

Conversation extraction now returns structured user and assistant turns to the native app.

Before Markdown or PDF generation, ContextPort validates that the conversation contains:

  • At least one positively identified user turn.
  • At least one positively identified AI response.
  • No known challenge or bootstrap content.

When ContextPort cannot safely identify a conversation, it fails instead of saving questionable context.

Claude Save Context Fix

Claude's standard chat renderer is now handled separately from Claude's Cowork and session interfaces.

The final fix was developed from Claude's actual shipped UI source captured directly by ContextPort Developer Mode.

The affected Claude conversation was retested on a physical iOS device and exported cleanly with the correct user and Claude turns.

Why This Matters

Memory is becoming one of ContextPort's core systems.

A failed export is visible.

A successful export containing incorrect context can silently affect every AI conversation that uses that Memory later.

ContextPort now treats the boundary between an AI webpage and saved Memory as a validation boundary.

If ContextPort cannot prove the conversation is safe to save, it does not save it.

Thank you to everyone testing ContextPort and reporting real device behavior. These reports are directly shaping the architecture of the app.

Support continued ContextPort development:

https://buymeacoffee.com/spotterdeer

Assets 4
Loading