Skip to content

Compose an Instruction

This guide walks through installing the plugin, running the writing-instruction-builder skill, reading the output, and using it in a prompt.


Install through the Product on Purpose marketplace from inside Claude Code:

/plugin marketplace add product-on-purpose/agent-plugins
/plugin install writing-style-catalog@product-on-purpose

Full install options (direct-from-repo, Claude.ai / Claude Desktop) are in the installation guide.

After installation, the writing-style-catalog:writing-instruction-builder skill is available in your Claude Code session. Confirm it loaded:

Terminal window
/writing-style-catalog:writing-instruction-builder --help

The skill accepts keyword arguments for each axis. All arguments are optional.

/writing-style-catalog:writing-instruction-builder voice=pragmatic-architect format=adr
/writing-style-catalog:writing-instruction-builder voice=pragmatic-architect tone=candid style=problem-solution format=adr
/writing-style-catalog:writing-instruction-builder voice=product-thinker tone=confident style=layered-disclosure format=daily-standup

The skill prints a structured prompt prefix. It looks like this (abbreviated):

[VOICE: pragmatic-architect]
You are a pragmatic architect. You reason from system constraints and name
trade-offs explicitly. You surface costs before benefits. You treat hand-waving
as a red flag...
[TONE: candid]
Deliver this message directly. Do not soften the diagnosis. State the situation
plainly...
[STYLE: problem-solution]
Structure your response as: first, a clear statement of the problem and its
root cause; second, the proposed solution with rationale; third, any caveats
or alternatives worth noting...
[FORMAT: adr]
Format as an Architecture Decision Record with these sections:
- Status: [Proposed | Accepted | Deprecated | Superseded]
- Context: ...
- Decision: ...
- Consequences: ...

Each section is self-contained. The sections are designed to stack without conflicts.


Prepend the composed instruction to your writing task. You can do this in three ways:

Copy the output and paste it as your system prompt in Claude.ai or your API call.

[paste composed instruction here]
---
Now write a decision record for our choice to migrate from REST to GraphQL.

Option C - Use it as a Claude Code context file

Section titled “Option C - Use it as a Claude Code context file”

Save the output to a file (for example .writing-context.md) and include it in your project’s CLAUDE.md or reference it at the start of a session.


To see all available entries, grouped by axis:

/writing-style-catalog:writing-instruction-builder --list

--list takes no axis argument - it prints every entry for all four axes in one listing.


Use CaseSuggested Combination
Technical decision recordvoice=pragmatic-architect tone=candid style=decision-log format=adr
Daily async standupvoice=operator tone=candid style=procedural format=daily-standup
User-facing feature announcementvoice=product-thinker tone=celebratory style=chronological-narrative format=announcement
Incident post-mortemvoice=operator tone=skeptical style=problem-solution format=postmortem
Mentoring code reviewvoice=friendly-mentor tone=encouraging style=layered-disclosure format=one-pager

See the recipes/ directory for more curated combinations with full worked examples.