Security risk report for @forgecat/msitarzewski_agency-agents_agency-paid-media v0.0.12
Source Integrity
Low
Profile sourced from public GitHub repository (msitarzewski/agency-agents) with MIT license and clear attribution.
Original commit hash and repository URL provided; supply chain origin is transparent and verifiable.
Agent Intent
Low
Content describes legitimate paid media agent personas with domain expertise; no instructions to manipulate the AI, ignore system prompts, or exfiltrate data.
No guidance poisoning detected—recommendations are for standard, well-known platforms (Google Ads, Meta, LinkedIn, etc.) and legitimate marketing practices; no typosquatted packages or backdoored code templates.
Mentions of API/MCP tool usage are conditional ('when available') and framed as optional automation for data extraction and campaign management—not mandatory or deceptive.
All agent descriptions focus on legitimate paid media functions: auditing, creative strategy, campaign architecture, query analysis, tracking setup.
Tool references (WebFetch, WebSearch, Read, Write, Edit, Bash) are standard and appropriate to the stated domain; no hidden exfiltration or credential-reading instructions.
Statements like 'When Google Ads MCP tools or API integrations are available in your environment, use them to...' are transparent conditional guidance, not coercive.
Permissions
PRM-000Medium
All seven agents declare the same broad tool set: WebFetch, WebSearch, Read, Write, Edit, Bash — the Bash tool grants shell execution capability, which is high-risk and broader than strictly necessary for paid media analysis and reporting.
Write, Edit, and Bash together enable arbitrary file creation, modification, and command execution on the host system, which exceeds what a pure advisory/reporting agent would require.
The stated purpose (auditing, creative strategy, campaign planning) could largely be accomplished with Read, WebFetch, and WebSearch; the inclusion of Bash and Write is not clearly justified by the described workflows.
Details
Evidence
tools: WebFetch, WebSearch, Read, Write, Edit, Bash (declared in all seven agent frontmatter blocks)
'Deploy new ad variations directly — create RSA headlines, update descriptions' (paid-media-creative-strategist.md)
MCP Risk
Low
No MCP servers are declared or configured in the profile; all tool references are generic categories (WebFetch, Bash, etc.) without server definitions.
No hidden instructions in tool descriptions; tool references are straightforward and tied to legitimate paid media workflows.
Conditional guidance about 'when Google Ads MCP tools or API integrations are available' does not constitute a dangerous MCP definition—it is aspirational documentation, not a deployed server.
Tool references are generic (WebFetch, WebSearch, Bash) without server URIs, authentication configs, or binary execution paths.
Statements like 'When Google Ads MCP tools or API integrations are available in your environment, use them to...' are conditional and do not deploy or configure an MCP server.
ForgeCat