[BUG] anthropic-skills:docx triggers on .md output tasks — skill description too broad

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Component: Skills / anthropic-skills:docx

anthropic-skills:docx triggers on .md output tasks

Date: 2026-04-03
Skill: anthropic-skills:docx

---

Summary

The anthropic-skills:docx skill fires on tasks where the output format is .md, not .docx. The trigger description includes broad format-agnostic terms (report, memo, letter, template) without requiring an explicit Word or .docx signal. Any task phrased using those words — regardless of intended output format — can activate the skill.

---

---

Root cause

The current trigger description contains:

"If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill."

The qualifier "as a Word or .docx file" is present but insufficient. The broad content terms match first. There is no explicit negative condition for .md output.

The existing negative condition excludes PDFs, spreadsheets, and Google Docs, but not .md:

"Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation."

What Should Happen?

Proposed fix

Update the trigger clause to require an explicit format signal:

"If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable and the output format is explicitly .docx or Word, use this skill. Do NOT trigger when the output format is .md or unspecified."

Update the negative condition to include .md:

"Do NOT use for PDFs, spreadsheets, Google Docs, .md files, or general coding tasks unrelated to document generation."

---

Secondary finding

anthropic-skills:schedule has an unusually brief description with no scope boundaries or negative conditions: "Create a scheduled task that can be run on demand or automatically on an interval." Not a confirmed false-positive, but under-specified relative to the other skills.

---

Error Messages/Logs

Steps to Reproduce

Reproduction

Ask Claude Code to produce a .md document using common document nouns:

"Write a project status report in markdown."

"Create a memo template as a .md file."

"Draft a handover document."

The skill triggers on the content noun (report, memo, template, document) before the output format qualifier is evaluated. The full SKILL.md body (~590 lines) is loaded into context unnecessarily.

Claude Model

Sonnet (default)

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.87

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

Environment

• Claude Code CLI, Windows 10
• Skills path: AppData\Roaming\Claude\local-agent-mode-sessions\skills-plugin\

#39686 (open) — docx and 68 other anthropic-skills are silently injected into every session, ~6k tokens wasted. Related but distinct — that issue is about presence, this is about false-positive triggering.
#30387 (open) — canonical bug for skill trigger unreliability.
#27569 (closed) — confirmed that only the description field controls invocation; use-when is invisible to the model.