- Home
- Skills
- Agents & Orchestration
- AI Collab Rules
AI Collab Rules
Enforce architectural integrity and human comprehension through ADR gates, explain-back checks, and quality audits.
Free
Works with the AI Agents you already use
AI Collab Rules
Example session with this skill installed
I want to change the Redis implementation to include key expiration. /one-way-door
- Read your context and instructions
- Compiled the ai collab rules
I have identified this as a high-stakes decision regarding data lifecycle and storage strategy. Please review the options below.
Architectural Decision Gate: Key Expiration Strategy
| Option | Pros | Cons | Reversal Cost | Impact Area |
|---|---|---|---|---|
| Delete on Access | Simple, no extra threads | Expired keys occupy memory until read | Low | Read paths |
| Periodic Sampling | Better memory recovery | Requires timer/concurrency | Medium | Event loop, Store |
| Hybrid (Redis style) | Balanced performance | Double implementation effort | Medium | Both of above |
Recommendation: Start with "Delete on Access." For current project scales, memory pressure is negligible, and sampling can be added later without refactoring existing access logic.
PM Perspective: This approach has the lowest testing overhead and minimizes schedule risk.
Next steps
- Select an option to generate the ADR draft in
docs/adr/. - I will then initialize the implementation and mark it for a future
explain-backcheck. - Update
ARCHITECTURE.mdif this establishes a new pattern for resource cleanup.
Connects securely to your tools. The creator never sees your data.
About this skill
AI-generated code often introduces technical debt through hidden complexity, copy-pasted patterns, and decisions the developer doesn't fully grasp. The ai-collab-rules package enforces human accountability and structural integrity by requiring manual verification of high-stakes changes and monitoring codebase health.
What it does
- Architectural Guardrails identifies one-way-door decisions like schema changes or new abstractions and forces an ADR draft before implementation.
- Explain-Back Validation prompts developers to explain the logic and side effects of recent changes in two lines to ensure actual understanding.
- Quality Auditing measures code verbosity and erosion trends weekly using cyclomatic complexity and duplication metrics.
- Log Management maintains
docs/explain-log.mdanddocs/adr/to track decision history and comprehension check pass rates.
How it works
- Decision Gate: The agent detects high-cost changes (e.g., new API shapes) and presents a trade-off table for user selection.
- Verification: After implementation, the agent selects a commit and asks the user to explain "Why this way?" and "What breaks if changed?".
- Trend Analysis: Run
/weekly-auditto generate a report on LOC, verbosity, and complexity erosion compared to previous weeks. - Simplification: The agent identifies the top three functions for refactoring based on mass and CC scores.
Frameworks & tools
Uses lizard for cyclomatic complexity and jscpd for duplication detection. Documentation is managed via Markdown files in docs/adr/ and docs/explain-log.md.
Why this beats prompting it yourself
Standard prompting lacks the persistence to track quality trends over weeks or the strictness to stop implementation until a decision is recorded. This skill formalizes a collaborative workflow where the human stays in control of the architecture while the agent handles the auditing and documentation grunt work.
Use cases
- Preventing "ghost" technical debt by forcing explanations for complex logic.
- Documenting critical architectural pivots via automated ADR scaffolding.
- Tracking codebase erosion in projects with heavy AI-assisted coding.
- Enforcing consistency between
ARCHITECTURE.mdrules and actual implementation.
Known limitations
Requires lizard and jscpd installed in the environment for full quality metrics. Relies on the user to honestly fill in the "Decision Reason" in ADRs to be effective. Skill and rule files are written in Korean. Agents follow them without changes, but you will need Korean (or a translation) to read or customize them.
Requires lizard and jscpd for metrics. User must manually provide ADR reasons and explain-back logic to ensure accountability.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 10 days ago
- Free with an Agensi account
Requires