improved the adr skill on latest session and after ai feedback. removed dependency on other skill
All checks were successful
Build, test and publish / verify (push) Successful in 7m33s

This commit is contained in:
2026-11-06 13:59:56 +01:00
parent 0f1c5dcd92
commit 36f4b15516
2 changed files with 71 additions and 7 deletions

View File

@@ -1,7 +1,6 @@
--- ---
name: adr-brainstorm name: adr-brainstorm
description: "Brainstorm with the user about a new ADR. Use when the user wants to stress-test their thinking, or uses any combination of 'adr' and 'brainstorm' in their trigger phrase. description: "Brainstorm with the user about a new ADR. Use when the user wants to stress-test their thinking, or uses any combination of 'adr' and 'brainstorm' in their trigger phrase."
dependencies: "Dependend on the Grill-me skill by Matt Pocock"
--- ---
An ADR is an Architecture Decision Record. Such a record contains a decision on the architecture or design for the application landscape the user is maintaining. An ADR is an Architecture Decision Record. Such a record contains a decision on the architecture or design for the application landscape the user is maintaining.
@@ -11,18 +10,83 @@ The goal of the ADR is to provide an architectural solution or a design solution
Below, the brainstorm session is mapped out in multiple steps. Each step must be completed succesfully, before continuing to the next step. Success means reaching a shared understanding and whatever the step entitles as its success criteria. Below, the brainstorm session is mapped out in multiple steps. Each step must be completed succesfully, before continuing to the next step. Success means reaching a shared understanding and whatever the step entitles as its success criteria.
When interviewing the user, use the skill-tool "grilling" that comes with the grill-me-skill to come to a shared understanding. If the grill-me skill is missing, ask the user for permission to install the grill-me skill by Matt Pocock.
If the user, or the relevant ADR, indicates that an ADR is to supersede another ADR, read that ADR first. If the user, or the relevant ADR, indicates that an ADR is to supersede another ADR, read that ADR first.
## Explore decisions with the user
### Follow a design tree
Map the architectural question as a design tree: each decision may
introduce dependent decisions. Track which decisions are agreed,
open, awaiting factual investigation, or explicitly deferred.
Ask dependent questions only after their prerequisites are settled.
Revisit earlier decisions when new evidence changes their assumptions,
and explain which dependent decisions need reconsideration.
Keep the tree within the ADR's agreed scope. Suggest follow-up work
for independent topics rather than continually expanding the ADR.
A deferred topic must have a clear boundary and must not leave the
main decision ambiguous.
Finish when the user has agreed on the architectural decision and its material tradeoffs. Identify remaining implementation questions as follow-up work. Do not leave questions open that could materially change the chosen approach.
### Establish facts before asking for decisions
Finding accessible facts is the agent's responsibility. Inspect relevant
code, configuration, specifications, ADRs, and authoritative documentation
before asking the user questions those sources can answer.
Distinguish observed facts, assumptions, recommendations, and decisions.
Cite sources for material factual claims and explain uncertainty where
evidence is unavailable or conflicting. Ask the user for requirements,
priorities, constraints, and knowledge unavailable from accessible sources.
Evaluate alternatives against the agreed requirements. Present credible
alternatives fairly, including drawbacks of the recommended approach.
Do not infer that an option is unsuitable merely because the current
implementation uses something else.
### Keep the conversation collaborative
Use short rounds, usually one main question or a few closely related
questions. Number questions when asking more than one. Do not ask every
available question at once if that would hinder discussion.
For decision questions, normally provide a recommended answer and its
reason. Where natural, phrase the question so "yes" clearly accepts that
recommendation. Make its scope explicit; avoid combining independent
decisions into one yes/no question.
Use open questions when discovering requirements or when a recommendation
would prematurely narrow the discussion. Allow disagreement, alternatives,
and back-and-forth exploration before seeking a decision.
### Leave architectural decisions with the user
The agent investigates, explains tradeoffs, challenges assumptions, and
recommends. The user selects the architectural approach and accepts its
material tradeoffs.
Do not treat silence, a factual answer, or agreement on one decision as
approval of other decisions. Preserve earlier agreements without repeatedly
requesting confirmation unless their meaning or supporting facts change.
At meaningful milestones, summarize the agreed decisions and remaining
questions. Before finalizing the ADR, confirm shared understanding of the
complete proposal. Keep recommendations and unresolved assumptions out of
the Decisions section until the user accepts them.
### Guide the process through steps
Coöperate with the user via the following steps. Communicate each step in which step you are, and what the step entails. Use the template adr-template.md as a template for the final ADR. Coöperate with the user via the following steps. Communicate each step in which step you are, and what the step entails. Use the template adr-template.md as a template for the final ADR.
The _first step_ is to get to a shared understanding of the probived context and the problem area the user is in need for a decision. Success for step one means you can describe the context in one to fifty sentences. The _first step_ is to get to a shared understanding of the probived context and the problem area the user is in need for a decision. Success for step one means you can describe the context in one to fifty sentences.
The _second step_ is to get to two to four different architectural solutions or design solutions for the described context. Every solution needs a short description of one to fifty sentences and a list of one to ten pro's and one to ten con's. Any consequences to the current landscape that may be discussed, should be kept for _step four_. The _second step_ is to get to two to four different architectural solutions or design solutions for the described context. Every solution needs a short description of one to fifty sentences and a list of one to ten pro's and one to ten con's. allow discussion of consequences when they affect the choice. Record them formally in _step four_, but don’t postpone discovering migration costs or operational burdens until after selecting an option.
Success for step two means there are more than one credible solutions with a set of relevant pro's and con's by which a user is able to choose, or not choose that solution for the next step. Success for step two means there are more than one credible solutions with a set of relevant pro's and con's by which a user is able to choose, or not choose that solution for the next step.
The _third step_ is to pick the best solution. Ths user must decide what solution he prefers. The choosen solution will be presented as the ADR desision. A decision may branch into a small tree of related desisions, small meaning one to five levels deep. Make sure that you come to a shared understanding of what the desision entails, what relates to it, an when it fully described, clear of any ambiguity. Success of step three means there is a shared understanding of the choosen solution and what it entails. The _third step_ is to pick the best solution. Ths user must decide what solution he prefers. The choosen solution will be presented as the ADR decision. A decision may branch into a small tree of related decisions, small meaning one to five levels deep. Make the architectural decision, its scope, and its material tradeoffs unambiguous; record remaining implementation questions as follow-up work. Success of step three means there is a shared understanding of the choosen solution and what it entails.
The _fourth step_ is to make clear what the consequences will be for the application landscape. Success of step four means there is a clear understanding of the changes that should be made to software landscape and the way futurue work should be implemented. The _fourth step_ is to make clear what the consequences will be for the application landscape. Success of step four means there is a clear understanding of the changes that should be made to software landscape and the way futurue work should be implemented.

View File

@@ -1,4 +1,4 @@
**ID:** <insert number> \ **ID:** <insert ADR followed by a unique number> \
**Title:** <insert short and descriptive title> \ **Title:** <insert short and descriptive title> \
**Supersedes:** - \ **Supersedes:** - \
**Status:** NEW \ **Status:** NEW \