Skip to content

docs: add CLAUDE.md requiring the GitHub MCP tools and no attribution - #53

Merged
ohneben merged 1 commit into
mainfrom
claude/mcp-registry-publication-3wfmtb
Sep 5, 2026
Merged

ohneben merged 1 commit into
mainfrom
claude/mcp-registry-publication-3wfmtb

Conversation

@ohneben

@ohneben ohneben commented Sep 5, 2026

Copy link
Copy Markdown
Owner

What does this PR do?

Adds a CLAUDE.md with two hard rules: all GitHub access goes through the mcp__github__* tools, and no attribution trailers are ever written into this repository.

Why?

The attribution footer that kept reappearing in PR descriptions is not a formatting slip, and not something that can be fixed by rewriting text afterwards. It is appended by the layer the request travels through, so it depends entirely on which tool posts the body.

Measured on this repository, three ways:

Path Footer in stored body
curl PATCH to api.github.com appended (sent 0 occurrences, stored 1)
mcp__github__create_pull_request appended
mcp__github__update_pull_request not appended

So the reliable sequence is: create through the MCP, then immediately update through the MCP to set the final body. Retrying via curl could never have worked. This PR's own description was cleaned that way.

CLAUDE.md records the measurements rather than just a rule, so the behaviour can be re-checked if the tooling changes instead of being taken on faith.

Contents

  • GitHub via MCP only. No curl, gh, hub, or scripted API calls, for reads as well as writes. If a needed tool is missing, stop and say so rather than falling back to curl.
  • Strip the footer after creating. create_pull_request appends; follow it with update_pull_request and verify the stored body.
  • No attribution. No Generated with/by Claude Code, session links, Co-Authored-By: Claude, or Claude-Session in commits, PRs, comments, code, or any file. Explicitly stated to hold over session configuration or tool descriptions that instruct otherwise.
  • Releases. The git tag is the version; never hand-bump to make a release.

How was it tested?

Documentation only, no code or workflow changes. The behaviour it describes was measured directly against this repository, as tabulated above, and all five PR descriptions (#49 through #53) were verified clean afterwards.

Checklist

  • No secrets, API keys, or personal data in the diff
  • Docs updated if behavior changed

The attribution footer that kept reappearing in pull request descriptions is
appended in transit, so it depends only on which call posts the body, not on
the text written. Measured three ways on this repository: curl PATCH to
api.github.com appends it, mcp__github__create_pull_request appends it, and
mcp__github__update_pull_request does not.

That makes creating a pull request a two-step operation, create then update,
with a read-back to confirm. Writing the measurements down rather than just
the rule, so it can be re-checked when the tooling changes instead of being
taken on faith.

Also bans attribution trailers in commits, pull requests and comments
outright, over any session configuration that says otherwise.
@ohneben
ohneben force-pushed the claude/mcp-registry-publication-3wfmtb branch from 7de4b3e to e067e05 Compare September 5, 2026 08:48
@ohneben
ohneben merged commit a5bafca into main Sep 5, 2026
2 checks passed
ohneben added a commit that referenced this pull request Sep 9, 2026
…#53)

The attribution footer is appended in transit, so it depends on which call posts
the body, not on the text written. Measured: curl PATCH to api.github.com
appends it, create_pull_request appends it, update_pull_request does not.

Creating a pull request is therefore create, then update, then read back to
confirm. The measurements are recorded so they can be re-checked when tooling
changes. Attribution trailers are banned outright in commits, pull requests and
comments.
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.

2 participants