Skip to content

chore(release): memhub v0.151.0 - #296

Merged
sgonz-xtrace merged 3 commits into
mainfrom
promote/memhub-v0.151.0
Nov 12, 2026
Merged

sgonz-xtrace merged 3 commits into
mainfrom
promote/memhub-v0.151.0

Conversation

@sgonz-xtrace

@sgonz-xtrace sgonz-xtrace commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor

MemHub plugin v0.151.0, built from internal main.

Source-RevId: e4fcf6d235d4249c96ff51840fbcb1abfa95c9cc
SHA256SUMS: d5230f230ce9a1cc41988219d6163cc6cbe60a518b53d74cffbf40c9deb50ad6

Built by promote run 42953286520 at XTraceAI/agent-plugins-internal@e4fcf6d235d4249c96ff51840fbcb1abfa95c9cc: the plugin suites pass at that commit, and the
export is verified — named memhub, one version across all four manifests, the production
backend in .mcp.json, and nothing shaped like a credential. Every file's digest was
re-checked against SHA256SUMS in the clone this branch was pushed from, so what is here is
what that run produced.

What's in this release

Everything merged on internal main since the last release, v0.133.0 (5403278ed): 33 change(s), 30 of them in the plugin (full range).

Security

Features

  • Codex sessions now capture reasoning tokens, request size, service tier and the account rate-limit snapshot for usage and cost analytics. (#220)
  • Claude Code sub-agents spawned by dynamic workflows are captured with their status, details and workflow runs, and forks no longer double-count the parent's response. (#221)
  • Team rules can now run a command before another is allowed, react to session conditions and final replies, and quarantine gated file edits made by shell scripts. (#231)
  • Codex sessions now get team rules drafted automatically in a background fork, filed as proposed rules for your review. (#235)
  • Cursor sub-agents, from both cursor-agent and the IDE, are now captured as conversations linked to their parent session. (#227)
  • Codex sub-agents are captured as linked conversations, and compressed rollouts and a custom Codex home directory are now read. (#229)
  • Cursor now enforces team edit rules, rules on command output, and run-X-before-Y ordering rules. (#236)
  • A session that fixes a bug now reports which pull request introduced it, and Codex and Cursor sessions record their git branch. (#238)
  • Hooks write a local trace of which ran in Cowork cloud sessions, with no payloads or keys, to help diagnose coverage. (#243)
  • Cursor sessions now start with the Rulebook and repo brain brief, and get PR-link prompts and spec reminders after tool calls. (#248)
  • Adds opt-in local rewind for Codex, replaying an earlier turn in an isolated checkout to compare models or instructions. (#249)
  • Rewind can now be started locally with a new skill and commands, and keeps explicitly granted extra writable directories. (#250)
  • Rewind now works with the unmodified Codex plugin, running custom hooks and local MCP servers through Codex's own sandbox. (#252)
  • Messages from a teammate's agent now appear above the prompt with Run and Dismiss, and only one open session runs each. (#255)
  • Local session rewind now supports Claude Code, replaying an earlier turn in a confined fork of the session. (#258)

Bug fixes

  • Artifacts saved from your session are now linked to it, so they appear in that session's contents in MemHub Studio. (#222)
  • Rule enforcement no longer falls back to the slower command hooks after a Mac sleeps overnight with no network. (#223)
  • Team rules on Codex now clear after a passing command, ignore failed commands, run on prompts, and the hook bridge refreshes itself on upgrade. (#232)
  • Cursor CLI and IDE sessions now record real timestamps, so agent time no longer shows as zero. (#230)
  • Cursor IDE token counts no longer include cache reads twice, so input usage is reported correctly. (#228)
  • Rewind now supports normal desktop sessions, compatible Codex versions and turns with clarification replies. (#253)
  • Concurrent rewind runs no longer fail on checkpoint locks, and completed forks no longer count inherited usage. (#254)
  • Codex rule drafting now runs quietly in the background with the parent's model, and a stale hook bridge no longer bypasses it. (#257)
  • Rule drafting in Claude Code sessions hosted by Agent SDK apps such as Zed no longer dumps a long tool log after the answer. (#260)

Changes

  • Agent capture now runs through a shared layer for all hosts, and agent type, description and result preview are redacted before upload. (#224)
  • The installer now lets rule-drafting forks file proposed rules under Claude Code auto mode. (#247)
  • In interactive Claude Code sessions, rule drafting now runs in an unseen fork and only shows a line when a rule is filed. (#256)
  • Only blocking rules now show on screen; advisory rules fire silently and the agent mentions them only when one changed its action. (#261)

Tests, CI and tooling

  • Fixes the three recurring failures in the nightly harness run. (#237)
Merged on internal but not in the plugin (3)
  • The CI Codex lane now runs Codex 0.163 with the user-level hooks that onboarding installs. (#233)
  • The Codex harness installs user hooks once and fails early with a readable reason when they cannot work. (#234)
  • Harness gains infrastructure for agent-capture parity checks on Codex and Cursor, including updated host version pins. (#226)
This PR replaces `plugins/memhub/` — and the root `install.sh`, when the export carries one — and nothing else.

Proved on staging, per surface (scripts/surface_evidence.py, source commit and v0.151.0):

Surface Proved by Result
Claude Code CLI harness release tier ⚠️ waived — nightly harness run 42976527333 on this commit (e4fcf6d2): 123/130 pass (auth 9/9, capture 24/26, rulebook 54/55, skills 36/40); failures in rewind probes and skill wording; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no release-tier harness run on this commit)
Codex CLI (codex exec) harness release tier ⚠️ waived — nightly harness run 42976527333 on this commit (e4fcf6d2): 24/27 pass; failures are the 2 rewind probes and a known double gate fire; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no release-tier harness run on this commit)
Codex app engine (codex app-server, the app's bundled version) harness release tier ⚠️ waived — nightly harness run 42976527333 on this commit (e4fcf6d2): 23/27 pass; failures are the 2 rewind probes, a known double gate fire and a flaky persisted-stdout case; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no release-tier harness run on this commit)
Cursor CLI (cursor-agent) harness release tier ⚠️ waived — nightly harness run 42976527333 on this commit (e4fcf6d2): smoke 3/3 pass (session, shell, edit); shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no release-tier harness run on this commit)
Claude Desktop, Code tab audit of real use ⚠️ waived — no real-use audit on this version; shares the Claude Code CLI engine covered by ci:claude; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no audit report)
Codex app (ChatGPT.app) audit of real use ⚠️ waived — no real-use audit on this version; the app engine is covered by ci:codex-desktop; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no audit report)
Cursor IDE audit of real use ⚠️ waived — no real-use audit on this version; Cursor CLI smoke passes; shipped ahead of the release-tier gate on the release owner's call; failing cases audited after release (no audit report)

Before this can merge: main requires the Real agent evidence check, which is a commit
status, not a check run — it exists only once someone dispatches it on this PR's exact head:

gh workflow run real-agent-evidence.yml -R XTraceAI/agent-plugins --ref promote/memhub-v0.151.0

Push again and the new head needs a new run.

After merge, in this order (RELEASING.md):

  1. Tag the merge commit memhub--v0.151.0 and push the tag first — the Claude pin
    cannot resolve a ref that is not there yet.
  2. Move ref and sha in .claude-plugin/marketplace.json to that tag, by PR.
  3. Smoke-test Claude, Codex and Cursor.

Codex and Cursor ship the moment this merges: their catalogs carry no pin and cache by
version string.

Promoted from XTraceAI/agent-plugins-internal by scripts/promote_ship.py.

Source-RevId: e4fcf6d235d4249c96ff51840fbcb1abfa95c9cc
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 11, 2026 19:12 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 11, 2026 19:12 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 11, 2026 19:12 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 11, 2026 19:12 — with GitHub Actions Active
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 11, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-12T02:04:12.882453Z 60e4a90 New commits
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ec0ad4b4a4

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

Comment thread plugins/memhub/scripts/redact.py Resolved
Comment thread plugins/memhub/board/register.ts Resolved
Comment thread plugins/memhub/board/register.ts Resolved
Comment thread plugins/memhub/scripts/rewind/runner.py Resolved
Comment thread plugins/memhub/scripts/rewind/runner.py Resolved
A promotion replaces plugins/memhub/ and install.sh only, so public's tests/
stayed at 0.133.0 while the plugin moved to 0.151.0. Seventeen suites failed
guard and probe-tests on #296, each on an intentional internal change.

The seventeen suites, tests/agent_fixture.py, the cursor/rewind fixtures, the
readers-cli codex golden, scripts/gen_codex_hooks.py and
docs/specs/pr-linking-plugin-spec.md are internal's at e4fcf6d2 (the promoted
commit), with memhub-staging rewritten to memhub. codex_correctness_test keeps
its staging fixture name so it still proves the prod install wins. The root
README gains /memhub:rewind and counts seventeen skills. No file under
plugins/ changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
test: bring public's suites up to the v0.151.0 plugin
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 12, 2026 01:48 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 12, 2026 01:57 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 12, 2026 01:57 — with GitHub Actions Active
@sgonz-xtrace
sgonz-xtrace deployed to production-plugin-release November 12, 2026 01:57 — with GitHub Actions Active

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 60e4a906b9

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

Comment thread plugins/memhub/scripts/rewind/runner.py Resolved
@sgonz-xtrace
sgonz-xtrace merged commit 91448b7 into main Nov 12, 2026
12 checks passed

This branch was successfully deployed

1 active deployment
production-plugin-release — 60e4a906 Deployed Nov 12, 2026 by sgonz-xtrace via Real agent session (claude) #99
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