output-style-governance

Purpose

Validate output-style manifests so that preferred output styles selected via /output-style remain consistent with rules/98-output-styles.md and rules/98-communication-protocol.md. Ensure each style manifest has correct frontmatter, clearly documented behavior, and does not weaken protocol invariants around safety, correctness, or structure.

IO Semantics

Input: Output-style manifest files under output-styles/*.md in the user workspace. Output: Governance findings and suggested fixes for style manifests (for example, missing fields, protocol violations, or unclear behavior descriptions). Side Effects: None directly; orchestration commands such as /llm-governance handle backups and writeback when applying suggested fixes.

Deterministic Steps

1. Target Discovery

• Locate output-style manifests in:
  • output-styles/*.md when running inside the user workspace (~/.claude)

2. Frontmatter Validation

• Check that each manifest contains:
  • name: non-empty, lowercase identifier without spaces.
  • description: concise human-readable description of the style.
  • keep-coding-instructions: boolean flag.
• Flag manifests that:
  • Omit required fields.
  • Use identifiers that conflict with existing core styles or are ambiguous.

3. Protocol and Rule Compliance

• Scan manifest bodies for instructions that conflict with:
  • Absolute prohibitions in rules/98-communication-protocol.md (for example, emotional language or small talk if disallowed).
  • Invariants and constraints in rules/98-output-styles.md (for example, attempts to weaken safety or encourage vague answers).
• Verify that manifests explicitly treat style instructions as additive to protocol invariants rather than replacements.
• Flag any instructions that attempt to override or ignore protocol prohibitions.

4. Style Behavior Clarity

• Check that each style manifest:
  • States whether it builds on a core style (for example, default, explanatory, learning) or defines a new core style.
  • Describes typical interaction patterns (for example, concise coding, explanatory insights, teaching interactions).
  • Avoids unnecessary narrative unrelated to style behavior.
• Recommend improvements where style behavior is ambiguous or underspecified.

5. Integration with Governance Workflows

• When invoked by agent:llm-governance or /llm-governance:
  • Provide a structured list of findings per manifest (missing fields, violations, ambiguities).
  • Suggest minimal, deterministic edits that bring manifests into compliance with rules/98-output-styles.md and protocol rules.

Validation Criteria

• All governed manifests include valid name, description, and keep-coding-instructions fields.
• No manifest contains instructions that contradict rules/98-communication-protocol.md or rules/98-output-styles.md.
• Core styles (default, explanatory, learning) are present and consistent with their rule definitions when used.
• Extended styles document their relationship to core styles and keep technical behavior aligned with protocol invariants.