[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:
- Read index.md — Reference file listing for navigation
- Select domain — FORMATS (structured data) or CONFIGURATION (response style scope)
- (formats) Read formats.md — Selection metrics, embedding, validation scoring
- (configuration) Read configuration.md — Scope hierarchy, embedding patterns
- (schema) Read schema.md — Delimiter syntax, canonical examples
- (structure) Read structure.md — Ordering patterns, composition
- (prose) Load
style-standardsskill — Voice, formatting, constraints - Execute per domain — Apply Guidance and Best-Practices
- 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.
[REFERENCE]: index.md — Complete reference file listing
[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-standardsfor 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-standardsskill: 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-standardsdelegation applied—zero duplication. - Format: Validation score >= 80 (formats domain).
- Quality: LOC within limits, weight-10 constraints first.
[REFERENCE] Operational checklist: →validation.md