Datadog analyses TeamPCP's open-sourced Shai-Hulud framework: credential and runner-memory theft, IDE and Claude Code hook persistence, OIDC npm publishing and a wipe-on-revocation daemon
TeamPCP open-sourced its Shai-Hulud framework; Datadog maps its theft, persistence and npm-poisoning code
Analysis
On 2026-05-12 a GitHub repository appeared with what Datadog Security Labs assesses as the complete source code of the Shai-Hulud offensive framework attributed to TeamPCP; its README reads "Love - TeamPCP" and "Change keys and C2 as needed", its commits are authored as TeamPCP_OSS, and GitHub removed it shortly after publication, though not before it was forked several times (Datadog Security Labs, 2026-05-13). Datadog maps the source directly to compiled payloads seen in the wild, including those of the TanStack and UiPath compromises, and finds 19 of the 22 TeamPCP techniques described in earlier reporting implemented in it (Datadog Security Labs, 2026-05-13).
The modular TypeScript and Bun toolkit exits on systems with a Russian locale, then reads OS-specific lists of more than 100 sensitive file paths, runs gh auth token and captures the process environment, and in GitHub Actions on Linux feeds a script to sudo python3 that reads the Runner.Worker process memory for values marked secret; further providers walk the AWS credential chain including instance and container metadata, read every Kubernetes secret in every namespace and enumerate Vault (Datadog Security Labs, 2026-05-13). Data leaves AES-256-GCM encrypted over HTTPS to a campaign domain, with a fallback domain found through signed GitHub commits and, failing that, new public GitHub repositories used as dead-drops (Datadog Security Labs, 2026-05-13). Stolen GitHub tokens are used to commit VS Code folder-open tasks and Claude Code SessionStart hooks to every branch of the victim's repositories, stolen npm tokens to republish the victim's packages with a preinstall loader, and, in a release workflow it was hardcoded to target, the workflow's OIDC token to publish a poisoned package with genuine Sigstore provenance; where it attaches a stolen token it installs a LaunchAgent or systemd user service that checks the token every minute and runs rm -rf ~/ once it is revoked (Datadog Security Labs, 2026-05-13).
Cited evidence
The repository's README explicitly states: "Love - TeamPCP" and "Change keys and C2 as needed."
This appears to have been removed by GitHub shortly after publishing, though not in time to prevent multiple forks.
the framework installs a daemon (gh-token-monitor) that runs rm -rf ~/ if your token is revoked. Disable it first, then rotate credentials.
Updates1
The published repository's README reads "Love - TeamPCP" and "Change keys and C2 as needed", its commits are authored as TeamPCP_OSS, and GitHub removed the repository shortly after publication, though not before it was forked, so the code remains available to other actors (Datadog Security Labs, 2026-05-13). Datadog's analysis describes no pull-request exploitation, pull_request_target targeting or expression-injection capability in the code. The framework harvests credentials where it already runs, reads Runner.Worker memory, poisons repository branches and npm packages with stolen tokens, and uses a workflow's own OIDC token only inside a release workflow and for a package it was hardcoded to target, so that path was manually targeted while the token-driven spread is opportunistic (Datadog Security Labs, 2026-05-13).
The analysis also documents what defenders need for a compromise check: an exit on Russian-locale systems, a daemon that deletes the home directory when a stolen GitHub token is revoked and must be disabled before rotation, and the hook files, commits and npm releases it leaves behind (Datadog Security Labs, 2026-05-13).
Sources1
Revision history
- Published 2026-05-15-58b94fbd
- Correction 2026-10-21T1148Z-audit
The repository that published the Shai-Hulud framework source is signed "Love - TeamPCP", and GitHub removed it after it had been forked. Datadog's analysis finds no pull_request_target or expression-injection exploitation in the code: the framework harvests credentials where it already runs, reads Runner.Worker memory, poisons repositories and npm packages with stolen tokens, and uses a workflow's OIDC token only in a release workflow it was hardcoded to target. Datadog's compromise checks include disabling the wipe-on-revocation daemon before rotating tokens.
Changed: affected_products body classification entities event_date evidence headline sources sourcing_note summary tags 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.