WIP: spec driven workflow

[Nikita Vlaev (nikitavlaevjb)]

This is a prototype of a spec-to-code approach setup.
It should work something like this:

1. User calls /specify *vague unstructured description of what needs to be done*
2. Command tries to fill out the spec template; leaves out places for clarifications and questions.
3. User refines spec manually or with AI; discussions happen.
4. Spec is finalized on the PR. <-- Documentation.

---

5. User calls /plan to create an implementation plan.
6. Command tries to fill out the plan template; leaves out places for clarifications and questions.
7. User refines plan manually or with AI; discussions happen.
8. Plan is finalized on the PR.

---

9. User calls /implement; agent implements according to plan.
10. User adjusts implementation details and accesses the result.
11. Iterates.
12. Code goes on PR; reviewer might do a full review, quick review, or skip it at all <-- Potential win; bottleneck eased.

---

Spec prompt:

 /specify We need to implement OSS Health metric for project. Here is RFC,  
  that contains a section describing this metric:                            
  /Users/nikita.vlaev/Downloads/KTL-4246 Exploratory research for            
  author-faced insights.md. Do a thorough research on the mentioned papers   
  and align the metric to the klibs.io usecase.  Archived / disabled GitHub  
  repo -- skip. GitHub returns 202 for /stats/participation or               
  /stats/contributors -- use reliable graphql apis to compute that instead   
  of relying on GH. as for the quota usage: spread it                        
  across time, don't do heavy job all at once. do one repo per period of     
  time. we also expose ossHealth in the search results DTO, yes. Prefer JPA  
  and HQL for validation.

Spec total cost: $4.32

---

TODO:

1. Explain /spike and /spec-from-spike
2. Iterate on the "plan" command and template.
3. Add an agentic review in the pipeline