- Home
- Skills
- Code Quality & Review
- Grill With Docs: Architecture Pressure Test
Grill With Docs: Architecture Pressure Test
Find conflicting assumptions and unresolved edge cases in architecture plans using your repository's docs and code.
Free
Works with the AI Agents you already use
Grill With Docs: Architecture Pressure Test
Example session with this skill installed
Pressure-test our cancellation plan. Docs say order-level; code cancels pending items but leaves shipped items unchanged. We want to label the whole order cancelled. Keep docs unchanged.
- Read your context and instructions
- Compiled the grill with docs
I pressure-tested the proposal against your supplied description. No docs were changed; the implementation details remain unverified without repository access.
Cancellation semantics
| Concern | Evidence supplied | Assessment |
|---|---|---|
| Order status may overstate the result | Shipped items remain unchanged, but the order would read “cancelled.” | A likely ambiguity for customers and support. |
| Docs and code may disagree | Docs describe order-level cancellation; code acts on pending items. | Possible conflict, not confirmed. Order-level may describe the request scope rather than its effect. |
Scenario: An order has one shipped item and one pending item. Cancellation stops the pending item. The shipped item still exists, so “cancelled” needs a definition that accounts for it.
Key question: Does cancelling an order mean stopping all remaining fulfilment, or undoing the entire order?
Recommendation: Define it as stopping remaining fulfilment, with item outcomes visible. Use “partially cancelled” for mixed outcomes unless an existing status convention expresses this clearly.
Alternative: Reject cancellation once any item ships. This is simpler, but prevents stopping the remaining pending items.
Next steps
- Choose the intended meaning of order cancellation.
- Share the status definitions and cancellation handler to check this concern against code.
Connects securely to your tools. The creator never sees your data.
About this skill
An architecture plan can sound clear while leaving retries, ownership, and failure states unresolved. Get an interactive challenge session that checks your assumptions against repository docs and code before you commit to the design.
What it does
- Finds conflicting meanings: Highlights terms that mean different things in your proposal, documentation, and implementation.
- Checks implementation claims: Traces relevant code to distinguish intended behavior from what actually happens.
- Challenges edge cases: Examines duplicate requests, late success, concurrent changes, and partial failures.
- Returns evidence-backed findings: Explains concerns, confidence, alternatives, and decisions that still need your input.
- Preserves agreed conclusions: Can update repository documentation when authorized.
How it works
- Share the plan or domain question and provide access to the relevant repository.
- Discuss focused questions supported by documentation and implementation evidence.
- Refine the proposal as objections are tested and uncertainties become clearer.
- Request documentation updates if you want settled conclusions recorded.
Works with
Repository source code, Markdown documentation, glossaries, architecture decision records, tests, and operational notes. It follows the repository's existing documentation conventions.
Why use this skill
Catch disagreements about behavior before they become implementation work. Evidence checks help separate real design gaps from concerns that existing code or documentation already answers.
Use cases
- Clarify cancellation rules across orders and items.
- Challenge retry behavior before changing a workflow.
- Resolve disagreements between architecture docs and code.
Known limitations
Needs relevant repository evidence. Designed for iterative discussion, not routine bug fixes or a one-shot architecture review.
Needs relevant repository docs and code for evidence-backed conclusions. Built for iterative pressure-testing, not routine fixes or one-shot reviews.
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
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Permissions
Reads repository docs, code, and tests and may use read-only terminal inspection. Write access applies only in separately authorized persistence mode; default critique mode does not modify files.