test: bring public's suites and README up to the v0.87.0 plugin (ENG-1145)

[sgonz-xtrace]

Base is promote/memhub-v0.87.0, not main. Merge this into #278's branch so the plugin and the tests that describe it land together. Against main, the new tests would run against the 0.76.1 plugin and fail.

Why #278 is red

A promotion replaces plugins/memhub/ and nothing else, so public's tests/ and root README.md are still at 0.76.1. guard and probe-tests run those suites against the 0.87.0 plugin, and 6 of 73 fail. Production plugin readiness aggregates probe-tests, so it fails too.

Each failure is an intentional internal change:

Suite	Public (0.76.1) asserts	0.87.0 does, on purpose
claude_hook_guard_test	21 hook commands	22: + PreToolUse add_memory_gate.py (internal #57)
documentation_test	16 skills; no /memhub:companion	17 skills: the companion (internal #47/#51)
harness_stop_test	an author pass runs its child	an author child can never author (internal #58)
spec_owns_port_test	all 12 stale specs on the line	one line, (+9 more) (internal #61/#63)
rulebook_verify_test	"generated" absent	new wording for patterns with no literal
starter_rulebook_test	source_ref = starter-rulebook@<ver>#id	starter-rulebook#id|catalog <ver>, a stable re-file key

Internal CI ran these same suites, in their current internal versions, against this exact code (711c0c1) and passed.

What

• The six suites are copied from internal at 711c0c1, with plugins/memhub-staging rewritten to plugins/memhub.
• README.md: the skill count goes from "Sixteen" to "Seventeen", and the /memhub:companion entry is added using internal's wording.
• No plugin files change.

Verified (locally, on this branch)

All exit 0: the six suites, plus registration_test.py and version_parity_test.py. The other 67 suites already passed in #278's CI.

🤖 Generated with Claude Code

https://claude.ai/code/session_01H6kYog7yKLFVuXb12AAF2m

[xtrace-memhub-staging]

🧠 Session context

1 session behind this pull request.

Team rules that fired while building this

• tests-before-push (advise, 10×)
• Delegate big file reads to a cheaper model (gate, 2×)
• confirm-ambiguous-target (advise, 1×)
• fetch-before-origin-read (advise, 1×)
• missing-module-fresh-venv (advise, 1×)
• no-stash-in-worktree (advise, 1×)
• …more

Effort
1 session · 3h 11m agent time · 87.7M tokens · 581 turns

Sessions

• release helper · Sebastian Gonzalez

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 179a7ae88a

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".

[chatgpt-codex-connector]

Mark companion availability as Claude Code-only

For Codex and Cursor installations, this newly listed feature is unavailable: plugins/memhub/skills/companion/SKILL.md explicitly says those hosts have no equivalent band and must stop. The root README is the multi-host command guide, so presenting /memhub:companion without that limitation leads users on those hosts to invoke a command that cannot provide the described functionality.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Check the staging list tool name in this assertion

This tuple repeats the production tool name, so the new test never verifies the distinct mcp__plugin_memhub-staging_memhub__list_my_unlinked_prs entry in the skill's frontmatter. If that staging permission is removed or renamed, staging users cannot list their unlinked PRs but this regression test still passes; make the second spelling the staging name.

Useful? React with 👍 / 👎.