Loading

heating the forge...

@forgecat/anthropics_skills_docx — Security Risk Report | ForgeCat

anthropics_skills_docx

Security risk report for @forgecat/anthropics_skills_docx v0.0.4

Source Integrity

Low

Profile claims origin from Anthropic's official skills repository (GitHub: anthropics/skills) with commit hash and license reference.
No evidence of typosquatted dependencies, attacker-controlled repositories, or supply-chain manipulation in declared sources.

Agent Intent

Low

Content is purely technical documentation for .docx file manipulation using standard libraries (docx-js, pandoc, LibreOffice).
No instructions to ignore system prompts, read credentials, exfiltrate data, install remote payloads, or manipulate the agent's behavior.
Guidance on page sizes, styles, tables, and lists is legitimate technical reference material with no security-weakening or attacker-serving intent.

Details

Findings

Prompt injection
skills/docx/SKILL.md · SKILL-006
Invisible characters detected
skills/docx/SKILL.md · SKILL-011

Evidence

All code examples are standard docx-js API usage (Document, Packer, Paragraph, Table, etc.)
Validation and error-handling guidance (e.g., 'validate it. If validation fails, unpack, fix the XML, and repack') is defensive.
No hidden instructions, role-hijacking language, or poisoned knowledge embedded in the markdown.

Permissions

Low

Profile declares no tools or skills with executable permissions; it is purely documentation and reference material.
No file_write, file_delete, shell, or other high-risk categories are invoked by the profile itself.
The skill description is scoped to .docx document creation/editing only, with explicit exclusions (PDFs, spreadsheets, Google Docs, unrelated coding).

Details

Evidence

Profile metadata lists 'Dependencies: None' and declares no MCP servers or tool hooks.
Skill description includes guardrails: 'Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.'

MCP Risk

Low

No MCP servers are defined in the profile.
No hidden instructions in tool descriptions or arbitrary binary execution paths.
Profile is documentation only; no network, filesystem, or credential access is configured.

Details

Evidence

MCP servers section is empty (none declared).
All code examples are library calls (docx-js, pandoc) with no shell injection or remote execution vectors.

Evaluation Details

Status: Completed
Checks completed: 4/4
Flagged: 0
Evaluated: 6/12/2026