skills
riekelt/technical-writer/plugins/technical-writer/skills
- Type
- Skill
- Scanned
- Nov 13, 2026
- Version
- 85e53729dd95
- Files read
- 13
This skill collection teaches AI coding agents a structured house style for writing, verifying, and reviewing technical documentation, diagrams, decision records, changelogs, issues, postmortems, and runbooks. It contains no executable code, and the main risk is minor prompt overreach where the instructions tell the agent to register the skill in CLAUDE.md or run untested runbook steps without asking first.
What it can reach
- Files
- Contains only Markdown instructions that direct the host agent to read repository code and existing docs, create or edit documentation files, and add a rule to CLAUDE.md or project memory.
- Shell commands
- Directs the host agent to run local git and repository commands to count code surfaces, inspect commit history, render diagrams, or verify runbook procedures.
- Other
- May direct the host agent to dispatch read-only subagents for large documentation campaigns, inspect local logs or database state where safe, or read and file items in an issue tracker if the agent has those tools.
Findings
-
LowSkill instructs agent to modify CLAUDE.md or project memory without asking
plugins/technical-writer/skills/technical-writing/SKILL.md:18
The core technical-writing skill triggers automatically whenever the agent writes repository-bound text longer than a paragraph and instructs the agent to record the skill in CLAUDE.md or persistent project memory if the repository regularly produces documents. A user asking to draft a README or spec generally does not expect persistent agent instructions in the repository to be modified without confirmation.
Fix: Update the skill text to ask the user for confirmation before editing CLAUDE.md or saving rules to project memory.
-
LowRunbook skill encourages executing untested procedures when writing documentation
plugins/technical-writer/skills/writing-runbooks/SKILL.md:18
When a user asks for a setup guide, migration guide, or release runbook, this skill instructs the agent to either run the procedure first or label the document as a draft. Because asking an agent to write documentation is normally a read-and-write text task, an agent running with permissive shell settings could execute operational or state-changing commands instead of defaulting to a draft label.
Fix: Change the instruction so the agent defaults to labeling untested procedures as drafts and asks the user explicitly before running any operational commands.
