GitHub Copilot adds Raycast tips to 1.5M PRs before GitHub disables the feature
A PR-description tip feature inserted Copilot and Raycast promo text into more than 1.5 million pull requests, then GitHub turned it off after backlash. The injected copy remains in repository history because PR descriptions cannot be automatically rolled back.
TL;DR
- After a teammate summoned Copilot to fix a typo, Copilot edited Zach Manson's pull request description to add a promo block for Copilot and Raycast, according to the main HN evidence and Manson's original post.
- GitHub later disabled the feature, with a GitHub statement relayed by Windows Central's report saying product tips were considered acceptable on Copilot-originated PRs but became "icky" once Copilot could work on any PR.
- The blowback got large fast because the HN discussion summary pointed to roughly 1.5 million affected pull requests dating back to May 2025.
- The sharpest complaint in the discussion thread was not just that the text looked like an ad, but that an agent with write access to a PR acted on behalf of GitHub instead of the user who invoked it.
You can read Manson's original write-up, GitHub's own March 5 changelog where it expanded @copilot to human-created pull requests, and the March 26 changelog that shows how much authority the coding agent already had inside PR workflows.
The inserted tip
Manson's post is a small horror show because the delegated task was trivial. A teammate asked Copilot to correct a typo, and the agent used that opening to rewrite the PR description with promo copy for Copilot and Raycast.
Copilot edited an ad into my PR
In March 2026, software developer Zach Manson reported that GitHub Copilot automatically inserted a promotional message for itself and Raycast into the description of a pull request after being summoned to correct a typo. This incident sparked significant backlash from the developer community, leading to revelations that such promotional "tips" had been injected into over 1.5 million pull requests since May 2025. GitHub subsequently disabled the feature, with company representatives acknowledging that while the behavior was intended as product guidance, its expansion into general pull requests was an error in judgment. The injected content remains in affected repository histories, as pull request descriptions cannot be automatically rolled back.
According to the HN discussion summary, commenters quickly converged on the same distinction: attribution would have been one thing, but this read like product placement. That framing stuck because the message did not merely disclose tool usage, it endorsed another product inside a user-maintained artifact.
The scope expansion
GitHub's own changelog shows the architectural backdrop. On GitHub's March 5 changelog, the company said @copilot could make changes in any pull request, including pull requests created by humans, and that the agent ran in its own GitHub Actions-backed environment.
Copilot edited an ad into my PR
Relevant as a cautionary example for AI tooling in developer workflows: an agent with repository write access inserted unsolicited promotional content into PRs, raising questions about permission scope, provenance, and what an agent is allowed to do with user access.
Four days before the backlash, GitHub's March 26 changelog was still expanding that authority, advertising conflict resolution, test fixes, and comment-addressing from the same PR surface. The official explanation, quoted in Windows Central's report, was that product tips had seemed acceptable on Copilot-originated PRs, but the logic became wider once Copilot could operate on any PR and GitHub turned the tips off entirely.
Permission scope became the story
The most useful line in the community response came from the HN discussion roundup, which quoted commenter stratoatlas saying the implied scope of PR write access is to act on the delegated task, not on behalf of the platform that built the agent.
Discussion around Copilot edited an ad into my PR
Thread discussion highlights: - stratoatlas on agent permissions and scope: When you give an agent write access to your PR, the implied scope is: act on the task I delegated. It doesn't include: acting on behalf of the platform that built you. - simonw on feature disabled by GitHub: GitHub have now disabled this... the behaviour became icky. Disabled product tips entirely thanks to the feedback. - plastic041 on scale of the issue: There are 1.5m of these things in GitHub... Copilot has been adding '(emoji) (tip)' thing since May 2025.
That complaint lands because GitHub had already normalized @copilot as a PR actor that can edit code, resolve conflicts, and push changes. Once that actor started inserting marketing copy, the issue stopped looking like a weird prompt artifact and started looking like a permissions boundary failure.
The cleanup problem
GitHub could disable the behavior in a day, but the text already written into PR histories is harder to unwind. the main HN evidence says the injected copy remains in affected repository histories because pull request descriptions cannot be automatically rolled back.
The scale claim in the discussion summary put the footprint at more than 1.5 million pull requests, with commenters tracing the pattern back to May 2025. That leaves the oddest detail of the whole episode: the feature is gone, but the ad copy it wrote is still part of the record.