Compose an Instruction

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

Step 1 - Install the Plugin

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:compose-instruction skill is available in your Claude Code session. Confirm it loaded:

/writing-style-catalog:compose-instruction --help

Step 2 - Run compose-instruction

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

Basic example - voice and format only

/writing-style-catalog:compose-instruction voice=pragmatic-architect format=adr

Full four-axis example

/writing-style-catalog:compose-instruction voice=pragmatic-architect tone=candid style=problem-solution format=adr

Using a different combination

/writing-style-catalog:compose-instruction voice=product-thinker tone=optimistic-realist style=layered-disclosure format=daily-standup

Step 3 - Read the Composed Output

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- Status: [Proposed | Accepted | Deprecated | Superseded]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.

Step 4 - Use the Composed Instruction

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

Option A - Paste into system prompt

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

Option B - Prepend to your user message

[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

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.

Listing Available Entries

To see all available entries for a given axis:

/writing-style-catalog:compose-instruction --list voices
/writing-style-catalog:compose-instruction --list tones
/writing-style-catalog:compose-instruction --list styles
/writing-style-catalog:compose-instruction --list formats

Common Combinations

Use Case	Suggested Combination
Technical decision record	`voice=pragmatic-architect tone=candid style=decision-log format=adr`
Daily async standup	`voice=ops-realist tone=candid style=step-by-step format=daily-standup`
User-facing feature announcement	`voice=product-thinker tone=energizing style=narrative-arc format=bullet-brief`
Incident post-mortem	`voice=ops-realist tone=cautious style=problem-solution format=technical-rfc`
Mentoring code review	`voice=technical-educator tone=mentoring style=layered-disclosure format=bullet-brief`

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