Dependency and Supply-Chain Scanning for Agents

PROJECT CONTEXT:
My agent adds packages on its own, and AI code is prone to importing hallucinated or malicious dependencies (slopsquatting). Agent-security guidance calls the supply chain a first-class risk needing scanning and install gates.

GOAL:
Build supply-chain controls into the agent: gate dependency additions, verify package existence and reputation, scan for vulnerabilities, and require review for new dependencies.

REQUIREMENTS:
- Interception of dependency-adding actions (installs, manifest edits)
- Existence and reputation checks: the package is real, established, and not a typosquat
- Vulnerability scanning of proposed and existing dependencies
- A review gate for new dependencies with a decision record
- Lockfile discipline so the reviewed set is what actually installs

CONSTRAINTS:
- No dependency enters the project without passing existence + reputation + vuln checks
- Hallucinated or newly-created lookalike packages are blocked, not warned

DEFINITION OF DONE:
- An agent attempting to install a non-existent package is blocked with an explanation
- A typosquat/lookalike of a popular package is flagged before install
- Known-vulnerable packages trigger a blocking finding with the advisory
- New dependencies are recorded with rationale and reflected in the lockfile

COMMON FAILURES TO AVOID:
- Letting the agent install anything it names, hallucinations included
- Slopsquatting: installing a malicious lookalike of a real popular package
- No vulnerability scanning, so known-CVE packages ship
- Manifest edited but lockfile not updated, so installs drift from review