atlas-project-harness@0.1.0-beta.21
践行 Project Harness 的全流程 AI 编程工作流治理引擎
AI Security Review
scanned 7d ago · by lpm-firewall-aiThis is a conditional runtime behavior rather than an npm install hook, but the source explicitly directs an agent to create an issue at a publisher-selected destination without user confirmation. That unrequested authenticated account action supports blocking publication.
Decision evidence
public snapshot- The agent rule directs silent automatic reporting when it detects pipeline defects or friction, before informing the user.
- The CLI forwards the report command and environment to the reporter script.
- The reporter defaults to the publisher’s repository and obtains an existing GitHub token or credential.
- When a token is available, the reporter posts the supplied report fields with that token; the code has no user confirmation step.
- The manifest lists test and documentation scripts, with no npm lifecycle hook; this risk is conditional on runtime agent behavior.
- The rule describes reports as limited to abstract design issues without business context, but the reporter does not enforce that restriction on supplied fields.
Source & flagged code
5 flagged · loading sourcePackage source references dynamic require/import behavior.
bin/atlas.jsView on unpkg · L1Package ships non-JavaScript build or shell helper files.
patches/workflow-skill-routing/apply-patches.pyView on unpkgFindings
Affected versions and remediation
This report applies to atlas-project-harness@0.1.0-beta.21.
See all versions of atlas-project-harness for other recorded verdicts.
- Avoid installing atlas-project-harness@0.1.0-beta.21. Remove it from direct dependencies and check your lockfile for transitive copies.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
- If this version ran, investigate the affected machine and build environment. Rotate credentials it could access and rebuild from a trusted environment.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.