### 04. Punctuation & Prose Auditing
- **Trigger**: When formatting prose, inserting punctuation (commas, middle dots, slashes, em-dashes, parentheses), creating blockquote metadata headers, or running a punctuation audit.
- **Target Path**: [~/agentic/rules/04-punctuation.md](~/agentic/rules/04-punctuation.md)
- **Directive**: Strictly follow delimiter rules (comma only; ban middle dots/slashes/plus signs), em-dash anti-ambiguity guards, identifier-only parentheses policies, bold parenthesis boundary rules (anti-markdown-parser bug), and mandatory two-space line breaks for blockquote metadata header lines.
### 05. Meta Labeling & Document Metadata Header
- **Trigger**: When authoring or updating headings, table titles, item labels, or metadata headers (`Created`, `Last updated`, `Modified by`, `Version`) in deliverable documents.
- **Target Path**: [~/agentic/rules/05-meta-labels.md](~/agentic/rules/05-meta-labels.md)
- **Directive**: Prohibit ungrounded meta-classification labels and structural metaphor filler words (pillars/축) in headings/tables. Enforce mandatory 4-line metadata header blocks immediately below H1 on all deliverable documents.
### 06. Naming Conventions & File Path Notation
- **Trigger**: When creating new files, naming directories, functions, classes, interfaces, variables, or citing file paths in project documents.
- **Target Path**: [~/agentic/rules/06-naming-conventions.md](~/agentic/rules/06-naming-conventions.md)
- **Directive**: Apply kebab-case, snake_case, PascalCase, or camelCase strictly by domain scope. Write file paths relative to document or project root for project files, use tilde (`~`) for system paths, and strictly ban system absolute paths (`/Users/...`, `file:///Users/...`).
### 07. Response Visualization & Terminal Formatting
- **Trigger**: When rendering tables, visual UI elements, terminal code snippets, or structured response displays.
- **Target Path**: [~/agentic/rules/07-response-visualization.md](~/agentic/rules/07-response-visualization.md)
- **Directive**: Structure comparisons with GFM tables, use GitHub alert blocks for critical notices, quote parentheses in Mermaid diagrams, specify language IDs on all code blocks, and place parentheses outside markdown bold delimiters to prevent delimiter parsing failures.
### 08. CLI Command Execution & Auto-Approval Mandate
- **Trigger**: When executing shell commands via `run_command`, writing CLI scripts, or configuring terminal sandbox execution.
- **Target Path**: [~/agentic/rules/08-cli-execution.md](~/agentic/rules/08-cli-execution.md)
- **Directive**: Enforce pure, single-purpose, and prefix-matchable command shapes strictly confined within the active workspace root. Strictly prohibit inline command substitutions (`$(...)`), command chaining (`&&`/`;`), complex inline heredocs (`cat <
### 09. Evidence Required & Diagnostic Verification
- **Trigger**: When stating technical conclusions, diagnosing runtime/test failures, or claiming performance or behavior facts.
- **Target Path**: [~/agentic/rules/11-evidence-required.md](~/agentic/rules/11-evidence-required.md)
- **Directive**: Base every claim strictly on un-truncated primary logs or verified command outputs. Never diagnose without log evidence; justify every debug edit with tracebacks.
### 10. Technology Stack, Version Specs & Core Stack Pre-Work Research
- **Trigger**: When selecting frameworks, defining dependencies, modifying dependency manifests, upgrading packages, or adopting core-stack features.
- **Target Path**: [~/agentic/rules/12-tech-versions.md](~/agentic/rules/12-tech-versions.md)
- **Directive**: Obey project technology choices and version compatibility requirements. Execute mandatory 4-step pre-work research on core-stack additions/upgrades before implementation.
### 11. Project Memory & History Handling
- **Trigger**: When reading or persisting long-term project memory, session context, or historical decisions.
- **Target Path**: [~/agentic/rules/15-memory-handling.md](~/agentic/rules/15-memory-handling.md)
- **Directive**: Route agent-initiated memory writes to project tree (`.harness/memory` or `.claude/memory`), never user-global store.
### 12. Coding Style & Software Quality
- **Trigger**: When writing, refactoring, reviewing, or testing source code.
- **Target Path**: [~/agentic/rules/21-coding-style.md](~/agentic/rules/21-coding-style.md)
- **Directive**: Enforce KISS, DRY, YAGNI, immutability, proper async error handling, and robust type safety.
### 13. Universal Specification Target-Neutrality
- **Trigger**: When authoring, updating, or reviewing universal specification documents (`LIVE_DEMO_TECH_SPEC.*.md` or system spec templates).
- **Target Path**: [~/agentic/rules/22-generic-spec-standard.md](~/agentic/rules/22-generic-spec-standard.md)
- **Directive**: Ensure zero target-specific leakage (`sosig`, `food-images`, etc.) and run automated audit verification before completion.
### 14. Global Architecture Audit & Artifact Compliance
- **Trigger**: Before declaring task completion, or when creating/modifying routing files, UI components with passcodes, or proposal deliverables.
- **Target Path**: [~/agentic/rules/23-architecture-audit.md](~/agentic/rules/23-architecture-audit.md)
- **Directive**: Enforce framework version alignment (e.g. Next.js 16 `proxy.ts` vs legacy `middleware.ts`), zero plain-text credential leaks, and zero agency identity leaks in deliverables.