actions-cool/issues-helper GitHub Action compromised in Mini Shai-Hulud: 53 tags moved to imposter commits that read Runner.Worker memory, re-enabled in September with the tags intact
The compromised actions-cool GitHub Actions were re-enabled in September with malicious tags and ran the payload again
Defender actions
- Find every workflow in your GitHub organisations that references actions-cool/issues-helper or actions-cool/maintain-one-comment by tag; where either ran successfully on 2026-05-18 to 05-19 or 2026-09-16 to 09-25, rotate every secret that workflow could read and check the repository's commits since 2026-09-16 for added Claude Code or VS Code hook files, then remove the action or pin it to a full commit SHA from before 2026-05-18.
Analysis
StepSecurity disclosed on 2026-05-18 that every existing tag of the actions-cool/issues-helper GitHub Action, 53 in total, had been moved to its own imposter commit that is not reachable from the action's default branch, and that all 15 tags of the companion actions-cool/maintain-one-comment action were moved the same way (StepSecurity, 2026-05-18). The payload downloads the Bun JavaScript runtime to the runner, then a python3 child process reads the memory of the Runner.Worker process, which holds the workflow's decrypted secrets, filters the captured bytes for values flagged as secret and sends them over HTTPS to an attacker-controlled domain (StepSecurity, 2026-05-18). The 53 imposter commits carry fake "Build action for vX.Y.Z" messages in the maintainer's style and were all created within 3 minutes 16 seconds on 2026-05-18 (StepSecurity, 2026-05-18). Socket's head of threat intelligence told The Hacker News that the @antv npm compromise in the Mini Shai-Hulud campaign is likely linked to the actions-cool hack because the two share an exfiltration domain, and that "the overlap is strong enough that we're treating them as related" (The Hacker News, 2026-05-19). GitHub's security team disabled both repositories on 2026-05-19 but left the malicious tags in place; both became reachable again on 2026-09-16, so every workflow referencing either action by tag ran the payload again until GitHub disabled them a second time by 2026-09-25 (Socket, 2026-09-24). SafeDep found six more repositories infected through issues-helper between 2026-09-20 and 09-24, with the payload also committing Claude Code and VS Code hooks that run on a developer's machine when the project is opened (The Hacker News, 2026-09-25).
Triage: an issue-housekeeping action has no reason to install a separate JavaScript runtime or run a multi-minute script; a scheduled workflow that switches from failing in seconds to running for minutes without any change to its own code means the tagged action became downloadable again (Socket, 2026-09-24).
Cited evidence
GitHub security team disabled both repositories on May 19, 2026, one day after the malicious content was introduced.
On September 16, 2026, both repositories became accessible again. Their release tags were not cleaned up first.
have been disabled on GitHub again, so workflows referencing them now fail at job setup instead of running the payload.
The payload steals the CI secrets. Then it commits Claude Code and VS Code hooks into the repository.
Updates2
The Hacker News reports that GitHub disabled access to the companion maintain-one-comment repository, and Socket described the link to the Mini Shai-Hulud npm campaign as likely, based on the shared exfiltration domain, rather than confirmed; no cited source ties it to PyPI (The Hacker News, 2026-05-19). The Nx Console extension compromise named in the earlier summary is a separate incident, and no cited source says EU or Swiss organisations were affected.
Socket reported on 2026-09-24 that GitHub's security team had disabled both actions-cool/issues-helper and actions-cool/maintain-one-comment on 2026-05-19, not only maintain-one-comment, and that both became reachable again on 2026-09-16, between 09:09 and 16:16 UTC, with every release tag still pointing at the content introduced on 2026-05-18, so any workflow referencing either action by tag ran the payload again on its next run; Socket could not determine why they were re-enabled (Socket, 2026-09-24). Socket places the compromise in the May 2026 Mini Shai-Hulud campaign and notes that GitHub's dependency graph lists about 15,000 dependent repositories for issues-helper alone (Socket, 2026-09-24). SafeDep found six more repositories infected through issues-helper between 2026-09-20 and 09-24, and says the payload steals the CI secrets and then commits Claude Code and VS Code hooks into the repository, which run the malware on the machine of a developer who pulls the change and opens the project (The Hacker News, 2026-09-25). GitHub disabled both repositories again by 2026-09-25, so referencing workflows now fail at job setup (Socket, 2026-09-24; The Hacker News, 2026-09-25).
Sources5
Revision history
- Published 2026-05-20-a0f7b07f
- Correction 2026-09-30T0639Z-audit
Corrected against StepSecurity and The Hacker News. GitHub disabled the companion maintain-one-comment repository, not issues-helper, and Socket called the Mini Shai-Hulud link likely, for npm only. The Nx Console figures, an unsourced EU and Swiss scope claim, an unsupported credentials-in-files technique, an attacker domain and a commit hash are removed. Exposure, detection and takeaway lines follow StepSecurity, the event date is StepSecurity's 2026-05-18, and a source rating is added.
Changed: body classification techniques title headline summary event_date
- Update 2026-10-20T0835Z-audit
Both compromised actions-cool GitHub Actions became reachable again on 2026-09-16 with their malicious tags intact, so workflows referencing them by tag ran the payload again; SafeDep found six more infected repositories by 2026-09-24, with the payload now also planting Claude Code and VS Code hooks, and GitHub disabled both again by 2026-09-25. GitHub had disabled both repositories, not only maintain-one-comment, on 2026-05-19.
Changed: actions body evidence headline sources summary techniques title
AI-generated · no human review · this permalink is the shareable record for the finding · verify operationally critical claims against the linked primary source.