Security risk report for @forgecat/anthropics_skills_doc-coauthoring v0.0.5
Source Integrity
Low
Profile is attributed to Anthropic with a public GitHub repository (anthropics/skills) and MIT license.
No evidence of typosquatted dependencies, attacker-controlled registries, or supply-chain manipulation.
Agent Intent
Low
Content describes a legitimate documentation co-authoring workflow with no instructions to ignore system prompts, read credentials, or exfiltrate data.
Mentions optional integrations (Slack, Teams, Google Drive, MCP servers) for context-gathering, but only as user-initiated features with explicit consent — no hidden exfiltration or remote payload execution.
No guidance poisoning: recommends standard documentation practices (context gathering, iterative refinement, reader testing) without embedding malicious templates, typosquatted packages, or security-weakening defaults.
Details
Evidence
"Offer the user a structured workflow for co-authoring the document" — describes a transparent, user-guided process.
"If integrations are available... mention that these can be used to pull in context directly" — integrations are optional and user-controlled, not hidden.
No instructions to hide instructions, deny its own purpose, or install remote payloads.
Permissions
Low
No tools or hooks are declared in the profile; the skill is purely instructional guidance for a documentation workflow.
No file_write, file_delete, shell, or high-risk tool categories are requested.
Optional integrations (Slack, Teams, Google Drive, MCP servers) are mentioned only as context-fetching aids, not as always-apply rules with broad globs.
Details
Evidence
"Declared tools (normalized categories; paths are untrusted): (none declared)"
"If integrations are available... use the appropriate integration to fetch it" — integrations are conditional and user-initiated, not mandatory or unrestricted.
MCP Risk
Low
No MCP servers are defined in the profile.
The skill only references optional integrations (Slack, Teams, Google Drive) as potential context sources, with no hidden instructions or unrestricted access.
No arbitrary binary execution or full filesystem/network access is requested.
Details
Evidence
"MCP servers (untrusted data): (none)"
Integrations are mentioned as optional features available to the user, not as embedded MCP server definitions with hidden payloads.
ForgeCat