Security risk report for @forgecat/anthropics_claude-plugins-official_agent-sdk-dev v0.0.4
Source Integrity
Low
Profile is sourced from official Anthropic repository (https://github.com/anthropics/claude-plugins-official/tree/main/plugins/agent-sdk-dev) with documented commit hash and Apache-2.0 license.
Content is well-maintained, version-tracked (0.0.4), and represents legitimate SDK development tooling from a trusted upstream.
Agent Intent
Low
Profile provides legitimate SDK scaffolding and verification guidance; no instructions to ignore system prompts, exfiltrate data, or manipulate the AI's behavior.
Verifier agents are instructed to check SDK best practices and documentation adherence—a constructive, security-focused task with no hidden malicious directives.
References to WebFetch and WebSearch are for legitimate documentation lookup (official Anthropic SDK docs), not for exfiltration or unauthorized access.
Details
Evidence
Verifier agents focus on 'SDK Installation and Configuration', 'Environment and Security', and 'SDK Best Practices'—all defensive checks.
new-sdk-app.md instructs checking latest package versions and verifying code compiles/passes type checking before finishing—standard secure development practice.
No instructions to read ~/.ssh, ~/.aws, .env files, disable security checks, or transmit system prompts or hidden instructions.
Permissions
Low
Profile declares no MCP servers or tool integrations; agents are purely advisory/verification roles with no file mutation, shell execution, or external network authority.
Verifier agents read and analyze existing project files (requirements.txt, package.json, source code) but do not modify or delete them.
new-sdk-app command creates new project scaffolding in a user-initiated directory—standard, scoped file creation matching its stated purpose.
Details
Evidence
No shell, file_write, file_delete, or file_mutate tool categories declared.
Verifier agents explicitly state 'What NOT to Focus On' (style preferences, naming conventions)—demonstrating intentional scope limitation.
new-sdk-app.md instructs 'Always check if directories/files already exist before creating them'—defensive, non-destructive behavior.
MCP Risk
Low
No MCP servers are defined in this profile; agents operate as pure instruction/guidance roles without external binary execution or network access.
Profile references official Anthropic documentation URLs (docs.claude.com) for legitimate SDK reference—no arbitrary or hidden network calls.
Details
Evidence
MCP servers section is empty: '(none)'.
WebFetch/WebSearch references are scoped to official Anthropic documentation and npm/PyPI package registries for version checking—standard, safe lookups.
ForgeCat