docs: the upstream kit is raw material, not text to post - #108
Merged
Conversation
Supersedes docs/kit-add-symptom, which added the missing symptom block but left the document's central instruction intact -- and that instruction was the actual problem. The kit offered paste-ready PR prose and issue comments. ggml-org/llama.cpp prohibits exactly that: "It is strictly prohibited to use AI to write your posts for you (bug reports, feature requests, pull request descriptions, Github discussions, responding to humans, ...)", with undisclosed AI usage risking a permanent ban. AI-generated CODE is allowed with disclosure; AI-generated PROSE is not. We found out by doing it. PR ggml-org/llama.cpp#27044 was filed with an AI-written description and commit message, and ggml-gh-bot flagged it within hours on two counts: template not filled in, and AI-generated description and commit message. Both had to be rewritten by hand, the commit amended and force-pushed, and a disclosure added. The one-line code change was never the problem. Merging the old branch would have left guidance on main that reads as battle-tested while pointing the next person at an account ban. So the document is reframed rather than patched: facts, measurements and file:line references to write from, with the policy and the ban risk stated up front, and what actually happened recorded so the failure mode is not rediscovered. Renamed to upstream-mmq-submission-material.md, since "PR description and comments" was the misleading part of the old name. Keeps everything that took work to produce -- the defect analysis, the verbatim symptom, the observed fault, the measurements, the reviewer caveats, the related issues, and the fork mechanics that cost a 403 to learn. Also keeps the symptom-first note from the superseded branch, which was independently right: our first draft led with the code argument and omitted the error text, making the report unfindable by anyone searching the crash they were hitting. 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
1 participant
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.
Supersedes
docs/kit-add-symptom(unmerged), which fixed the missing symptom block but left the document's central instruction intact — and that instruction was the real problem.The kit offered paste-ready PR prose and issue comments. ggml-org/llama.cpp prohibits precisely that:
AI-generated code is allowed with disclosure. AI-generated prose is not.
We found out by doing it. ggml-org/llama.cpp#27044 was filed with an AI-written description and commit message;
ggml-gh-botflagged it within hours on both the missing template and the AI-generated text. Body and commit message were rewritten by hand, the commit amended and force-pushed, and a disclosure added. The one-line code change was never the issue.Merging the old branch would have left guidance on
mainthat reads as battle-tested while pointing the next person at an account ban. So this reframes the document rather than patching it: facts, measurements andfile:linereferences to write from, with the policy and ban risk up front, and the filing incident recorded.Renamed to
upstream-mmq-submission-material.md— "PR description and comments" was the misleading part of the old name.Nothing of value is lost: the defect analysis, verbatim symptom, observed fault, measurements, reviewer caveats, related-issue hypotheses and the fork mechanics (which cost a 403 to learn) all stay. The symptom-first note from the superseded branch is kept — it was independently right.
Docs only.
🤖 Generated with Claude Code