@trypattern/sdk@0.1.16
AI Security Review
scanned 5d ago · by lpm-firewall-aiThe package does not automatically publish during installation or pulling, but its first-pull instructions explicitly direct an agent to publish a change without owner approval. That conditional agent capability is a concrete residual risk, while the source does not establish an autonomous attack.
Decision evidence
public snapshot- package.json exposes `dist/cli/app.js` as a command and lists development scripts, with no install lifecycle hook shown.
- The `.claude` references in `dist/cli/app.js` and `dist/cli/publishing.js` are formatter input and an empty-folder exclusion, not agent configuration writes.
- After an explicit `app pull`, first-time or starter pulls return onboarding instructions through `firstChange`.
- Those instructions direct Claude to change the site and publish the first change “without asking.”
- Publishing sends a signed-in POST to the app’s publishes endpoint; the default dashboard is `https://onpattern.app`.
- The CLI’s publish command is described as building and putting the app live.
- The CLI does not invoke publishing as part of the pull flow; the publish instruction is printed as a next step and requires a separate command.
- Pull and publish operate on the selected Pattern site, and publishing uses the Pattern dashboard.
Source & flagged code
9 flagged · loading sourceThis package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
dist/cli/publishing.jsView on unpkgPackage source references child process execution.
dist/cli/publishing.jsView on unpkg · L1Findings
Affected versions and remediation
This report applies to @trypattern/sdk@0.1.16.
See all versions of @trypattern/sdk for other recorded verdicts.
- Review the evidence and your use of @trypattern/sdk@0.1.16 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.