Thank you to the six people who closed the consolidation problem

[OthmanAdi]

Seven months, four attempts, six people. v3.11.0 finally closes it, and I want to say who got it there, including the person whose pull request I turned down.

The problem

The Claude Code plugin registered six skills: the canonical one plus five translations. Every session paid for all six skill descriptions in its system prompt, whether or not the user had any use for Arabic or Traditional Chinese. Roughly half the plugin's always-on cost was languages nobody had picked.

The people

@tiptinker (#47, January) and @back1ply (#53, January) filed the earliest two reports, about the same duplication from the other direction: too many copies of one skill across too many folders. scripts/sync-ide-folders.py exists because of them.

@sean3808 (#130, April) wrote the report that framed everything after it. The scripts and templates were identical, the prose was the only real difference, and five descriptions were loading into every session for languages most users would never invoke. Every later attempt is a variation on that issue. It stayed open for four months, which was not neglect.

@oaabahussain (#151, May) asked for a single canonical source with a CI parity gate. The gate shipped in v2.37.0. Five releases later it is what made a 120-file move verifiable instead of hopeful.

@SomSamantray (#216, August) built the first working answer and I did not merge it. That deserves saying plainly. The audit inside that PR was the most useful thing anyone had sent this repo in a while: it proved the drift #130 predicted, named the twelve missing scripts and the missing Windows UTF-8 fix exactly, and showed the sync tooling covered only three dispatch targets. All of that shipped as v3.10.0. The approach was deletion, and deletion was the one thing this repo could not do. The work was right; the mechanism was not available to me.

@dylanpulver (#226, August) found the mechanism that was. Move the five variants one directory deeper instead of deleting them. Claude Code scans skills/*/SKILL.md at a single level and does not recurse, so the plugin stops registering them. npx skills add resolves --skill by skill name across a recursive scan, so every install command keeps working and every translation stays individually installable. He checked the install route instead of assuming it, carried the literal English status-token warning into all five language commands where the earlier attempt had not, and named the one cost he could not remove. He also diagnosed an unrelated macOS bug while rebasing, which is fixed in the same release.

Why it took seven months

The obvious fix was to delete the five directories. It was proposed more than once and it was always the wrong tool.

Skill listings on a public directory cannot be retracted once published. Deleting the folders would not have removed the entries; it would have left five permanent listings whose install command had started to fail. That is worse for the people using them than either keeping them or removing them properly. And the translations have real users who chose them on purpose. Breaking a working install is not a reasonable price to pay for a packaging problem.

So the requirement was a route that changes nothing anyone depends on. That took until #226 to find. The delay was never doubt about whether the problem was real. It was refusing to fix it by breaking installs.

What shipped

Measured against the real Claude Code loader, not inferred from the docs:

	before	after
Registered plugin skills	6	1
Components	19	14
Always-on tokens per session	~2,254	~1,042

All thirteen slash commands still work, including /plan-ar through /plan-zht. Every translation is still installable by name. Nothing was deleted and nothing was renamed.

The translations also finally have a documentation page of their own, docs/languages.md. There was a setup guide for every supported IDE and nothing at all for languages.

Thank you

To the six of you: this is what a maintainer wants an issue tracker to look like. Reports with evidence, pull requests with tests, and someone eventually finding the door the rest of us walked past. Two of these attempts did not merge and both of them still shipped, in the form of things this release depended on.

Release: https://github.com/OthmanAdi/planning-with-files/releases/tag/v3.11.0