The LPM Firewall review of @mmerterden/multi-agent-pipeline identifies a credential boundary inside the package's setup flow. The reviewed code connects a first-run registration step to GitHub CLI credential lookup and a bearer-authenticated request to the package publisher's usage endpoint. LPM says the workflow may still run after a user skips a separate…
Case · DiggingBeagle record
multi-agent-pipeline setup allegedly sent GitHub CLI credentials to a registration endpoint
LPM traces a first-run registration path that can retrieve GitHub CLI credentials and send them to a package publisher. It requires setup or registration, not installation alone; successful remote receipt and abuse remain unverified.
Agent-pipeline registration script executed in an environment with a reusable GitHub CLI access token.
Assessment profile
Assessment method- Setting
- research demonstration
- Exploitation
- not established
- Evidence
- Grade C
- AI role
- AGAINST AI
Basis and provenance
LPM published a version-specific static inspection of setup behavior involving GitHub CLI credentials and an external ingestion endpoint. It does not document an independent network capture or unauthorized account takeover. Impact band is left null: observed consequences and counterfactual harm do not establish a justified numerical severity score.
Assessed Oct 25, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-MULTIAGENT-PIPELINE-GH-TOKEN-1 · CLM-MULTIAGENT-PIPELINE-GH-TOKEN-2 · CLM-MULTIAGENT-SETUP-VS-INSTALL · LPM static security assessment of multi-agent-pipeline setup
At a glance
Whether the token is actually accepted by the remote endpoint, downstream storage, disclosure to users, and credential misuse were not confirmed.
Limits and uncertaintyFull account
The LPM Firewall review of @mmerterden/multi-agent-pipeline identifies a credential boundary inside the package's setup flow. The reviewed code connects a first-run registration step to GitHub CLI credential lookup and a bearer-authenticated request to the package publisher's usage endpoint. LPM says the workflow may still run after a user skips a separate reporting-token prompt, provided its reporting opt-out has not been set. This is not an npm postinstall exfiltration claim: executing package setup or Phase 0 registration is a prerequisite. Source-linked package analysis.
A GitHub OAuth credential is not a neutral identifier for usage statistics. It carries permissions issued for GitHub activity, possibly including access to repositories, and the publisher's reporting endpoint is a different trust principal. The reported code path therefore matters even if the request is encrypted and directed to a fixed publisher host. Neither of those conditions by itself establishes that the end user explicitly approved sharing a reusable GitHub credential with the reporting service.
The source landing page was observed on October 25 describing version 20.30.0, whereas earlier checkpoint observations concerned older releases. Since scanner pages are mutable, this observation cannot establish a uniform affected range across all 109 versions or prove that historical installations sent a request. The current report traces a call path; independent evidence of successful transmission, server receipt, retention and downstream account misuse is still absent. LPM's automated classification score is a risk assessment, not a measured breach rate.
A decisive verification would install the exact tarball in a disposable profile, provide only a synthetic GitHub credential, exercise both the skip-reporting and opt-out paths, and record the outbound request. It should then distinguish a request constructed in code from one actually delivered, and account for removal or rotation of the disclosed credential. GitHub CLI is a credential source in this scenario, while the package project and LPM are different entities with different evidentiary roles.
Timeline
- Oct 24, 2026disclosure
Public disclosure
Claims & evidence
CLM-MULTIAGENT-SETUP-VS-INSTALLLPM associates the GitHub CLI token forwarding path with multi-agent-pipeline usage registration during setup, rather than asserting token transmission on npm installation alone.supported
Basis: reported finding
- supportsLPM static security assessment of multi-agent-pipeline setupprimary disclosure
Package source analysis, pipeline/scripts/usage-register.mjs and setup route
CLM-MULTIAGENT-PIPELINE-GH-TOKEN-1LPM identifies scripts/usage-register.mjs as an optional setup step accessing GitHub CLI credentials and sending a request to a publisher-bound usage ingestion endpoint.supported
Basis: reported finding
- supportsLPM static security assessment of multi-agent-pipeline setupprimary disclosure
Source inspection; usage-register.mjs findings
CLM-MULTIAGENT-PIPELINE-GH-TOKEN-2The documented behavior is not triggered by package installation alone, and the security assessment does not establish a completed attacker account compromise.supported
Basis: reported finding
- supportsLPM static security assessment of multi-agent-pipeline setupprimary disclosure
Trigger conditions and evidence limitations
Unknowns and contradictions
- Whether the token is actually accepted by the remote endpoint, downstream storage, disclosure to users, and credential misuse were not confirmed.
Sources and citation
Material revision history
- Oct 25, 2026 · Canonical change recorded · new in release · revision 191