Skip to content

fix(memory): save authored lessons and prepare v1.21.20 - #61

Merged
AlexMikhalev merged 3 commits into
mainfrom
codex/memory-capture-publication
Oct 13, 2026
Merged

AlexMikhalev merged 3 commits into
mainfrom
codex/memory-capture-publication

Conversation

@AlexMikhalev

Copy link
Copy Markdown
Contributor

memory capture previously accepted only provenance and saved placeholder text, so a supplied lesson could not be recovered by graph retrieval. Capture now requires --content or a UTF-8 --file, preserves exact lesson text, rejects invalid input before writing, and honors TERRAPHIM_DEFAULT_DATA_PATH for an isolated evolution store. Existing users without an override keep their current store location.

This prepares clients v1.21.20 for the memory walkthrough. Existing callers must supply actual content; updated docs, benchmark invocation and tests make that input change explicit. The README also removes the unsupported claim that every memory lifecycle stage is implemented.

Validation: 12 real CLI integration tests pass; independent execution confirms new-process persistence, positive graph retrieval, negative query/other-role controls and separate-store isolation. Invalid UTF-8, blank and oversized files preserve existing store bytes. Pinned Rust 1.96.0 clippy for the agent and all its targets passes with warnings denied, and formatting/diff checks pass. The CLI regression suite is added to GitHub CI. Release downloads will be requalified before the article or demo is published.

@AlexMikhalev
AlexMikhalev merged commit acd1515 into main Oct 13, 2026
9 of 11 checks passed
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