Output Style Builder

[H1][OUTPUT-STYLE-BUILDER]

Dictum: Parsing clarity requires structured serialization.

Build agent response formats. Configure scope-based style.

[DELEGATE] Voice, formatting, constraint rules → style-standards skill.

Tasks:

1. Select domain — FORMATS (structured data) or CONFIGURATION (response style scope)
2. (formats) Read formats.md — Selection metrics, embedding, validation scoring
3. (configuration) Read configuration.md — Scope hierarchy, embedding patterns
4. (schema) Read schema.md — Delimiter syntax, canonical examples
5. (structure) Read structure.md — Ordering patterns, composition
6. (prose) Load style-standards skill — Voice, formatting, constraints
7. Execute per domain — Apply Guidance and Best-Practices
8. Validate — Quality gate; see §VALIDATION

Scope:

• Formats: Structured data serialization (JSON, YAML, Markdown-KV, XML) for agent output.
• Configuration: Response style scope hierarchy (global → project → skill → command).

Domain Navigation:

• [FORMATS] — Structured data output. Load for: agent schemas, API responses, validation scoring.
• [CONFIGURATION] — Scope hierarchy. Load for: CLAUDE.md output sections, precedence rules.
• [SCHEMA] — Delimiters, examples. Load for: syntax reference, canonical patterns.
• [STRUCTURE] — Ordering, composition. Load for: section sequencing, chaining.

References:

[1][FORMATS]

Dictum: Format optimization requires accuracy-token tradeoff analysis.

[IMPORTANT] Format choice impacts accuracy (16pp variance) and tokens (6.75x variance).

Guidance:

• Selection — Markdown-KV: 60.7% accuracy, 2.7x tokens. JSON: 52.3% accuracy, 0.85x tokens.
• Embedding — Inline for single use. Reference (@.claude/styles/) for 3+ consumers.
• Validation — 100-point scoring. Deployment requires score >= 80.

Best-Practices:

• Single format per output type.
• Max 5 variables per format; every optional has default.
• Constrained decoding guarantees 97-100% schema compliance.

References:

• →formats.md: Selection, embedding, validation.
• →format.template.md: Format scaffold.

[2][CONFIGURATION]

Dictum: Narrow context requires specialized output rules.

[IMPORTANT] Higher precedence overrides lower (command level 5 > global level 1).

Guidance:

• Global — CLAUDE.md [OUTPUT] section. 50-100 LOC optimal.
• Project — PROJECT.md overrides. Document divergence reason.
• Skill/Agent — Inline or reference pattern. Match reuse requirements.
• Command — Narrowest scope. Specialized output for single invocation.

Best-Practices:

• Global rules in CLAUDE.md; overrides minimal (max 30 LOC).
• Reference style-standards for voice/formatting rules—no duplication.
• Place weight-10 constraints first in all scopes.

References:

• →configuration.md: Scope hierarchy, embedding.
• →style.template.md: Style scaffold.
• [DELEGATE] style-standards skill: Voice, formatting, constraints.

[3][SCHEMA]

Dictum: Performance consistency requires delimiter standardization.

[IMPORTANT] Maintain delimiter consistency throughout output.

Guidance:

• Delimiters — Code fence (```), separator (---), soft break (<br>).
• Consistency — Prohibit mixed delimiter styles within single output.

References:

• →schema.md: Syntax reference.

[4][STRUCTURE]

Dictum: Attention decay requires priority-first placement.

[IMPORTANT] Primacy effect assigns 5.79x weight to early items.

Guidance:

• Ordering — Action-first, Priority-first, or Context-first patterns.
• Hierarchy — Maximum 3 levels. 2-7 items per container.
• Composition — Base-override inheritance. Shallow or deep merge.

References:

• →structure.md: Sequencing, composability.

[5][TEMPLATES]

Dictum: Structural consistency requires scaffold reuse.

FORMATS Domain:

• →format.template.md: Structured data format scaffold.

CONFIGURATION Domain:

• →style.template.md: Response style scaffold.

[6][VALIDATION]

Dictum: Gates prevent incomplete artifacts.

[VERIFY] Completion:

• Domain: Selected FORMATS or CONFIGURATION scope.
• References: Required domain files loaded per Tasks.
• Style: style-standards delegation applied—zero duplication.
• Format: Validation score >= 80 (formats domain).
• Quality: LOC within limits, weight-10 constraints first.

[REFERENCE] Operational checklist: →validation.md