aiwg
Version:
Cognitive architecture for AI-augmented software development with structured memory, ensemble validation, and closed-loop correction. FAIR-aligned artifacts, 84% cost reduction via human-in-the-loop, standards adopted by 100+ organizations.
165 lines (114 loc) • 5.88 kB
Markdown
# AIWG Rules
Deployable enforcement policies that define default behaviors AIWG users can count on.
## What Are Rules?
Rules are structured enforcement documents that configure AI assistant behavior. Unlike guidelines or suggestions, rules are **non-negotiable defaults** that AIWG deploys to every supported platform.
When a user runs `aiwg use sdlc` or `aiwg new`, rules are copied or injected into the platform's native rule location. The AI assistant then follows them automatically.
## Tiers
| Tier | Deployed When | Purpose |
|------|--------------|---------|
| **core** | Always (`aiwg use`, `aiwg new`) | Universal behaviors every user can count on |
| **sdlc** | With SDLC framework (`aiwg use sdlc`) | Workflow-specific enforcement |
| **research** | When research features active | Research corpus management |
### Core Rules (Always Deployed)
These define what AIWG **is** as a tool:
| Rule | Enforcement | What It Does |
|------|-------------|-------------|
| `no-attribution` | CRITICAL | Zero AI branding in commits, PRs, docs, code |
| `token-security` | CRITICAL | Never hardcode, log, or expose tokens |
| `versioning` | CRITICAL | CalVer format, no leading zeros |
| `citation-policy` | CRITICAL | Never fabricate citations or sources |
| `anti-laziness` | HIGH | Never delete tests or remove features to pass |
| `executable-feedback` | HIGH | Execute tests before returning code |
| `failure-mitigation` | HIGH | Mitigate known LLM failure archetypes |
These 7 rules ship with every AIWG installation. They're the behaviors that make AIWG a trustworthy tool rather than an unpredictable assistant.
## Deployment Strategy: Consolidated Index
Instead of deploying all 31 individual rule files (~9,321 lines), AIWG deploys a single **`RULES-INDEX.md`** file (~200 lines) containing:
- 2-3 sentence summaries of each rule (enough to determine relevance)
- @-links to full rule files for on-demand loading
- Organization by tier and enforcement level
- Quick Reference table mapping task types to relevant rules
This achieves **~95% context reduction** while preserving discoverability. Agents scan the summaries and load full rules via @-links only when needed.
### Before/After
```
# Before: 31 individual files deployed
.claude/rules/
├── no-attribution.md
├── token-security.md
├── versioning.md
├── ... (28 more files)
# After: Single consolidated index
.claude/rules/
└── RULES-INDEX.md
```
### Cleanup Behavior
When redeploying, old individually-deployed `.md` files in the target rules directory are automatically cleaned up. Non-`.md` files (e.g., Cursor's `.mdc` files) are preserved.
### Fallback
If `RULES-INDEX.md` is not found in the source, providers fall back to deploying individual rule files.
## Platform Deployment
Rules deploy to platform-native locations:
| Platform | Target | Mechanism |
|----------|--------|-----------|
| Claude Code | `.claude/rules/RULES-INDEX.md` | File copy |
| Cursor | `.cursor/rules/RULES-INDEX.md` | File copy |
| Copilot | `.github/copilot-instructions.md` | Content injection |
| Codex | `.codex/rules/RULES-INDEX.md` | File copy |
| Warp | `.warp/rules/RULES-INDEX.md` | File copy |
| Factory AI | `.factory/rules/RULES-INDEX.md` | File copy |
| OpenCode | `.opencode/rule/RULES-INDEX.md` | File copy |
| Windsurf | `.windsurf/rules/RULES-INDEX.md` | File copy |
**File copy platforms** deploy the consolidated `RULES-INDEX.md` with summaries and @-links.
**Content injection platforms** embed the index content into the platform's context file.
## CLI Integration
```bash
# Deploy all rules for a framework
aiwg use sdlc # Deploys core + sdlc tier rules
# Deploy to specific platform
aiwg use sdlc --provider cursor # Deploys to .cursor/rules/
# Scaffold new project with rules
aiwg new my-project # Includes core tier rules
# Regenerate rules for a platform
aiwg regenerate claude # Syncs rules to .claude/rules/
# List deployed rules
aiwg rules list
# Check rule compliance
aiwg rules check
```
## Rule Structure
Each rule file follows this format:
```markdown
# Rule Name
**Enforcement Level**: CRITICAL | HIGH | MEDIUM
**Scope**: What this rule applies to
**Issue**: #NNN (Gitea tracking issue)
## Principle
Why this rule exists.
## Mandatory Rules
### Rule 1: ...
**FORBIDDEN**: (what not to do)
**REQUIRED**: (what to do instead)
## References
Links to related rules, schemas, and research.
```
## Manifest
`manifest.json` (v2.0.0) registers all rules with metadata for the CLI:
- **name**: Rule identifier
- **file**: Filename in this directory
- **enforcement**: critical | high | medium
- **tier**: core | sdlc | research
- **description**: One-line purpose
- **issue**: Gitea issue reference
- **consolidation**: Strategy metadata (`index-with-links`, deploy single index file)
- **deployment.platforms**: Per-platform target paths for `RULES-INDEX.md`
## Adding New Rules
1. Create the rule file in this directory following the standard structure
2. Add an entry to `manifest.json`
3. Choose the appropriate tier (core = universal, sdlc = workflow, research = corpus)
4. Set enforcement level based on impact of violations
5. Run `aiwg rules check` to validate
## Relationship to .claude/rules/
`.claude/rules/` is the **deployment target** for Claude Code. This directory (`agentic/code/frameworks/sdlc-complete/rules/`) is the **canonical source**.
- Source of truth: `agentic/code/frameworks/sdlc-complete/rules/`
- Claude Code deployment: `.claude/rules/`
- Other platforms: Platform-specific locations per deployment map
The `aiwg use` and `aiwg regenerate` commands sync from source to targets.
For this repository (which dogfoods AIWG), `.claude/rules/` may also contain project-specific rules that aren't part of the framework. Only rules listed in `manifest.json` are deployed to other projects.