Output Format Selection for Claude Prompts

Calibration: Tier 1, Opus-primary. See repository README for model compatibility.

Tested output format specifications that control what Claude's deliverable looks like — structure, sections, length, and format. Without an explicit output format, Claude defaults to its own formatting preferences, which often means overlong prose with excessive bullet points.

When to Use This Skill

Use this Skill when you need to:

• Choose an output format for a Claude prompt you are building
• Structure a deliverable (brief, memo, plan, analysis, etc.)
• Design a custom output format for a deliverable type not covered here
• Fix output quality problems caused by missing or vague format instructions

This Skill provides the format specification (Layer 5 of a well-structured prompt). It does not cover identity, reasoning methodology, or quality control — for those, see rootnode-block-selection or rootnode-prompt-compilation if available.

Quick-Reference: Deliverable Type → Format

Selection principle: Match the format to the audience and their primary question. Executives ask "what should we do?" → Executive Brief or Strategic Memo. Engineers ask "how does this work?" → Technical Design Document. Operators ask "what do I do next?" → Process Documentation or Implementation Plan. Analysts ask "what did we find?" → Research Summary.

How to Use a Format Specification

Each format in the reference files includes:

1. Use-when guidance — the scenario and audience it fits
2. Complete XML specification — copy directly into your prompt's output layer
3. Per-section length guidance — word counts, sentence counts, or item counts for each section
4. Watch-for notes — Claude's common failure modes with that format and specific countermeasure language to add

To apply a format: copy the <output_format> XML block into your prompt. Customize section names and lengths to your situation. Remove sections that are not relevant. The specifications are tested starting points — adjust to match what your audience actually needs.

Critical: Controlling Claude's Formatting Defaults

Claude has strong formatting tendencies that output specifications must actively manage:

Bullet point overuse. Claude defaults to bullet points for almost everything. If you want prose, say "write in prose" explicitly in the relevant section. If you want a table, say "present as a table with columns: X, Y, Z."

Length inflation. Claude tends to write longer than necessary. Per-section length guidance (e.g., "2-3 sentences" or "1 paragraph") is more effective than total word counts alone. Specify both when possible.

Burying the lead. Claude often builds up to its conclusion rather than leading with it. Executive-facing formats (Executive Brief, Strategic Memo, Stakeholder Update) all specify lead-with-the-answer structure — enforce this explicitly if Claude reverts.

Hedging on assessments. In formats requiring judgment (Competitive Analysis, Decision Matrix, Post-Mortem), Claude may soften findings. The watch-for notes in each format specification include countermeasure language to add when this happens.

Building Custom Output Formats

When no existing format fits, use this template:

<output_format>
Structure your response as follows:

[SECTION 1 — name and what it contains]: (length guidance)
[SECTION 2 — name and what it contains]: (length guidance)
[SECTION 3 — name and what it contains]: (length guidance)

Constraints:
- Total length: [word count or page count]
- Tone: [formal / direct / conversational / technical]
- Format: [prose / table / numbered list — specify per section if mixed]
- Audience: [who will read this and what they need from it]
</output_format>

Design Principles for Custom Formats

Lead with what the reader wants most. Executives want the recommendation first. Engineers want the architecture first. Researchers want the findings first. Put the highest-value section at the top.

Specify length per section, not just total. "500 words" tells Claude the total budget but not how to allocate it. "Bottom Line: 2-3 sentences. Analysis: 3 paragraphs. Next Steps: 3-5 items." tells Claude how much attention each section deserves.

Constrain the format where it matters. If you want prose, say "write in prose" — otherwise Claude defaults to bullet points. If you want a table, say "present as a table with columns: X, Y, Z" — otherwise Claude may use bullets where a table would be clearer.

Name sections descriptively. "Analysis" is vague. "Competitive Assessment" or "Root Cause Analysis" tells Claude what kind of analysis belongs there. Section names are implicit instructions.

Reference Files

This Skill includes four reference files with complete format specifications. Read the relevant file when you need the full XML specification, section details, and watch-for guidance for a specific format:

• references/executive-formats.md — Executive Brief, Strategic Memo, Stakeholder Update. Read when the deliverable targets senior leadership, decision-makers, or project stakeholders.
• references/technical-formats.md — Technical Design Document, Process Documentation. Read when the deliverable targets engineers, technical reviewers, or process executors.
• references/analytical-formats.md — Research Summary, Decision Matrix, Competitive Analysis. Read when the deliverable involves evidence synthesis, structured comparison, or market assessment.
• references/operational-formats.md — Implementation Plan, Post-Mortem / Retrospective. Read when the deliverable involves project execution planning or incident/project analysis.

Troubleshooting

Output ignores the format specification. The format block may be positioned too late in the prompt. Place the <output_format> block after the identity and reasoning instructions but before any input data. Claude processes instructions with a primacy-recency bias — format instructions buried after long input sections lose influence.

Sections are the right structure but wrong length. Add explicit per-section length constraints if you only specified a total word count. "Total length: 800 words" gives Claude a budget; "Key Findings: 3-5 items, each 2-3 sentences. Analysis: 2-3 paragraphs per theme." tells Claude how to allocate it.

Claude adds sections not in the specification. Add a constraint: "Use only the sections specified above. Do not add additional sections, appendices, or disclaimers."

Format works for simple inputs but breaks on complex ones. The format may need more sections or different length guidance for complex inputs. Consider whether the task actually ne