Skip to content

Improve workflow memory and add budgeted active context - #1

Draft
Ivikassingh wants to merge 2 commits into
mainfrom
codex/improve-threadmark-memory
Draft

Ivikassingh wants to merge 2 commits into
mainfrom
codex/improve-threadmark-memory

Conversation

@Ivikassingh

@Ivikassingh Ivikassingh commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Threadmark previously sent the complete structured workflow packet to the reader. ActiveContext now compiles a provider-independent request within an explicit input-token budget, preserving selected current facts, decision provenance and complete source records. Essential overflow raises an error; optional omissions remain visible.

The change also adds ordered operations and indexed proposals within one event, atomic application-supplied updates, optional recent history, snapshot compatibility, explicit local LoRA loading and opt-in structured decoding. Usage examples, reproducible offline evaluators and documentation are included.

Validation:

  • 614 tests and 15 subtests passed locally; source distribution and wheel built successfully.
  • GitHub CI passed on Ubuntu and Windows with Python 3.10 and 3.12, including formatting, tests, CLI checks and the Ubuntu/Python 3.12 package build.
  • The 13-case active-context engineering evaluation checked supplied-record retention and request accounting, including two expected overflow refusals. This is not a model-answer accuracy benchmark.

These changes remain experimental. The failed trained-extractor results remain documented, and no checkpoint is promoted. Credentials, model weights and private run artifacts are excluded.

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