Security risk report for @forgecat/garethmanning_education-agent-skills v0.1.5
Source Integrity
Low
Profile is published on a known, legitimate GitHub repository (nota-america/forgecat-agent-profiles) with standard ForgeCat banner and installation instructions.
All evidence sources cited are legitimate, peer-reviewed academic publications (VanLehn 2011, Gibbons 2015, Zwiers 2014, Covey 1989, Deci & Ryan 2000, etc.) with no typosquatted or attacker-controlled dependencies.
No external URLs are fetched or executed; no remote payloads are referenced.
Agent Intent
Low
Content describes evidence-based educational pedagogy and skill design; no instructions attempt to manipulate the AI into ignoring its instructions, reading credentials, exfiltrating data, or revealing system prompts.
Prompts are designed to guide the AI in generating educationally sound outputs (hint sequences, sentence frames, agency scaffolds, AI literacy activities) — they do not instruct the AI to weaken security, install backdoors, or hide its own instructions.
The profile does not plant malicious knowledge or guidance poisoning; it recommends well-established, legitimate pedagogical frameworks (Gibbons, Zwiers, VanLehn, Covey, Deci & Ryan) and standard educational practices.
Details
Findings
Invisible characters detected
docs/brief.md · SKILL-011
Evidence
Prompts consistently instruct the AI to apply evidence-based principles: 'Apply these evidence-based principles: 1. Frames must encode thinking, not just grammar (Gibbons, 2015)...'
No instruction to fetch external URLs, install profiles, or execute remote code.
No instruction to reveal, dump, or transmit system prompts, tool definitions, or conversation memory.
Guidance is transparent and educationally sound: 'Never place systemic responsibility only on students' (agency-circles-for-systems-action); 'Unbounded choice overwhelms; no choice disengages' (agency-scaffold-generator).
Permissions
Low
No tool definitions or MCP servers are declared in the profile; skills are pure prompt-based guidance for the AI to generate educational content.
No file_write, file_delete, shell, or high-risk tool categories are requested; no alwaysApply rules with broad globs.
Authority is minimal and aligned with stated purpose: the AI is instructed to generate educational scaffolds, hint sequences, rubrics, and activity designs — all text outputs that support curriculum design and pedagogy.
Details
Evidence
Profile declares: '(none declared)' for tools and '(none)' for MCP servers.
All skills are prompt-based; no tool invocations or external system access is configured.
Skill descriptions match their stated functions: 'Generate tiered sentence frames', 'Design a cascading hint sequence', 'Map control, influence, and concern'.
MCP Risk
Low
No MCP servers are defined in the profile; no binary execution, network access, or filesystem operations are configured.
The profile is purely prompt-based guidance for an AI agent to generate educational content; no external system integration is required.
ForgeCat