Repository navigation
docs(blog): add the version rule to the agent-memory post - #744
Merged
Merged
Conversation
A tour of how agent tools shape memory today (instruction files, the Anthropic memory tool, Copilot Memory, Letta blocks, Mem0, LangGraph's store, MCP and temporal knowledge graphs), what they agree on, why LoreKit's record looks the way it does, and how the same record carries a CI job's versioned JSON state. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Fix claims that did not match the code: TTL is opt-in and only a write carrying a TTL resets it (LoreKit is permanent by default), rewrites drop unknown frontmatter fields, kind is null unless inferred, and the :: rationale (git forbids ':' in branch names). Add the existing portability proposals (Portable Memory, Open Memory Protocol, COGX), LangGraph's read-refreshed TTL, the GitHub-shaped repo grammar, and a section answering the objections a skeptic would raise. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Four SVGs in the blog's figure style: a map of memory formats by who writes them and how structured they are, export vs live addressing, how Copilot, LangGraph and LoreKit decide when a memory expires, and one record envelope carrying a prose lesson and CI JSON state. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
…ng posts Captures the workflow used for the open memory format post: check LoreKit claims against code, outside claims against sources, run a skeptic's review and plain-language rewrite, draw SVG figures in the blog style (with a render script to inspect them), register the post, and compile it through the real MDX pipeline. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Retitle to 'Agent memory needs its OpenTelemetry moment'. Lead with the OpenTracing/OpenCensus to OpenTelemetry history and its four layers, present LoreKit as one open-source proposal, and add a section listing seven gaps in LoreKit's own format (stable id, open attributes, content type, verifiable anchors, explicit lifecycle, supersession, host-neutral address and version) with a sketch record. Add an OTel-parallel figure and stop highlighting LoreKit in the format map. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Keep 'Agent memory needs its OpenTelemetry moment' (~8 min) focused on the formats, the existing proposals and what a shared spec could look like. Move LoreKit's record, its seven gaps, the sketch record and the CI test into 'Seven gaps in LoreKit's memory format' (~7 min), dated 2026-11-14 so it stays a draft in production until then. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
The 'Seven gaps' post now describes the stable id, attributes, content_type and supersedes as shipped (#736), adds a What changed section with the new CLI flags, and keeps anchors, an explicit expiry rule and the address grammar as the open gaps. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Removes the post, its registry entry and its two figures, and rewords the three sentences in the OpenTelemetry post that pointed to it. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
…r-own angle Cuts about a third: folds the four-questions section into the spec section, shortens the tour, drops the weakest objection and the map figure that repeated the table. Fixes the shape counts, the OpenTelemetry figure caption, the four-layers mismatch, an overstated MCP claim, and an undefined use of 'anchors'. Adds an early paragraph on how many people are building their own memory layer, and why that needs an agreement rather than another tool. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
The blog's example record is a markdown fence, and MDX_SHIKI_LANGS only loads the grammars content declares, so rehype-shiki.spec failed. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
A reader pointed out that traces are append-only but memories get corrected, so a shared record needs a version rule or every tool invents its own. The post now asks a fifth question (which version is current), adds version and supersedes to the example record and the conformance tests, and answers the objection in the skeptic section. Claude-Session: https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa
Contributor
|
The dashboard preview for this PR — redeployed on each push that changes the web app.
Preview always points at this PR's latest commit; Deployment is the immutable build of |
Owner
Author
✅ No issues foundAdds the version rule the prior run asked for: a fix to the same claim bumps <a href=" Review details
Nothing to report — standards (skipped), measurability (skipped), severity, reach (impact graph found no consumer or importer), 0 files skipped.
Generated by Claude Code |
mthines
marked this pull request as ready for review
November 9, 2026 08:49
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A reader pointed out a gap in "Agent memory needs its OpenTelemetry moment": traces are append-only, but memories get corrected. A runbook from January is fixed in March, and without a version rule both copies look right. If the spec doesn't define that rule, every tool invents its own and the lock-in stays.
The post now:
versionandsupersedesto the example record, and explains them under itversion, and a different claim that replaces it is a new record, starting again at version 1, thatsupersedesthe old oneThe post makes no new claim about LoreKit. The matching LoreKit change is in #745.
Verified:
npx vitest run src/lib/blog src/lib/mdxpasses, and the post compiles through the real MDX pipeline. The blog isn't part ofllms.txt, so no other docs needed changes.https://claude.ai/code/session_012YX3RPBKN75heHAMkL2Toa