docs: fix an inaccurate install section, and separate the two installs - #28
Open
duncancrawford wants to merge 1 commit into
Open
docs: fix an inaccurate install section, and separate the two installs#28duncancrawford wants to merge 1 commit into
duncancrawford wants to merge 1 commit into
Conversation
The install section was wrong in ways that would stop someone installing at all, and one that would silently break their estate. **The feed is not private.** It said releases were reachable only 'Inside HMCTS', told everyone else to install from source, and required keyring/artifacts-keyring plus a device-flow sign-in. Verified otherwise: an anonymous pip download with a clean HOME and no credentials resolves and fetches the 0.5.1 wheel. The feed's root URL does prompt for sign-in, which is what makes it look private, but pip requests the per-package path, which serves the listing and artefacts unauthenticated. Credentials are only needed to publish. **The Quickstart taught the trap the library warns about.** Its example filter put a trailing comment on a team rule. A rule's value is the whole rest of the line, so the slug became 'my-github-team # or: everything a GitHub team owns' and matched nothing -- the same silent failure that dropped three repositories from the crime estate. The comment moves to its own line, the rule is stated, and the warning that catches it is pointed out. **Two installs, not one.** A pip install from an Azure feed and a plugin install from a GitHub clone were documented 200 lines apart with nothing saying they are unrelated, so they read as steps in one flow. There is now a table naming both, what each is for, and who needs which -- plus the asymmetry that surprises people: a release does not ship a skill change, and a skill change needs no release. Also drops two counts that had already gone stale (estate size, test count), and the pinned example version that rots with every release. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
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.



Left behind when #27 merged — pushed while that merge was in flight, so
maintook the plugin-manifest fix without these README corrections. Same
stranded-commit pattern
CLAUDE.mdrecords; this is the fourth time.The feed is not private
The section said releases were reachable only "Inside HMCTS", told everyone else
to install from source, and required
keyring/artifacts-keyringplus adevice-flow sign-in. None of that is true.
Verified:
pip downloadwith a cleanHOME, no~/.netrcand no keyringresolves and fetches the 0.5.1 wheel (77,012 bytes, valid zip). The feed root
does return 401, which is what makes it look private, but
piprequests theper-package path — that serves the full distribution listing and the artefacts
unauthenticated. Credentials are needed only to publish, which is the release
workflow's job.
The practical cost of the old text: anyone outside HMCTS was told to install
unreleased
mainfrom source instead of a pinned release, and anyone inside wassent to set up credentials they never needed.
The Quickstart taught the trap this library warns about
Its example filter put a trailing comment on a
teamrule. A rule's value isthe whole rest of the line, so it parsed as:
— a slug matching no team, contributing nothing, silently. That is the same
failure that dropped three repositories from the crime estate and became issue
#20. The comment moves to its own line, the rule is stated, and the warning that
now catches it (
discover, and--strictfor CI) is pointed out.Two installs, not one
A
pipinstall from an Azure feed and a plugin install from a GitHub clone weredocumented 200 lines apart with nothing saying they are unrelated, so they read
as steps in one flow. There is now a table naming both — what each gives you,
where it comes from, who needs it — with the two
## Install:headings renamedinto a matching pair and cross-linked.
It also states the asymmetry that surprises people: a release does not ship a
skill change, and a skill change needs no release, so a store can pin library
v0.5.1while running skills from a latermain.Stale numbers dropped
The estate size (
93 repositories, 745k nodes), the test count (110, actually232) and the pinned example version (
==0.5.0, released 0.5.1) — all of whichrot on their own schedule.
Note on ordering
hmcts/cp-knowledge-store#30 deep-links to
#install-the-claude-code-plugin,which exists only once this merges.
🤖 Generated with Claude Code