@markjaquith/agency@3.4.2
Manage agentic work across repositories with durable workbases
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThe package has an automatic postinstall hook that modifies a global AI-agent extension path and injects workbase context at runtime. The behavior is scoped and source-visible, with no confirmed theft, remote execution, or broad foreign-control-surface mutation.
Decision evidence
public snapshot- The automatic postinstall hook installs a Pi agent extension.
- The installer creates and writes to the user-wide Pi extensions directory.
- The installed extension injects workbase instructions into Pi's system prompt and exposes skills from several agent-tool directories.
- The lifecycle installer only copies the package's static extension and removes that same path on uninstall.
- The extension activates only after it finds an Agency workbase and validates Agency context.
- No source evidence of network exfiltration, self-dependency, payload download, or credential harvesting was found.
Source & flagged code
10 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgPackage source references dynamic require/import behavior.
src/utils/interactive-loader.tsView on unpkg · L3Findings
Affected versions and remediation
This report applies to @markjaquith/agency@3.4.2.
See all versions of @markjaquith/agency for other recorded verdicts.
- Review the evidence and your use of @markjaquith/agency@3.4.2 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.