Skip to main content
  • Status: accepted · Date: 2026-07-17 · Deciders: Davor Runje

​
Context

A repo must be set up to use honest-scholar — both fresh repos and existing ones with prior research artifacts (mononet has benchmarks, reference PDFs, informal hypotheses). This is the payoff for the “benchmarks folder out of control” problem that motivated the effort.

​
Decision drivers

  • Both greenfield and brownfield onboarding are needed.
  • They share the same end state (the consumer layout).
  • Backfill is inventory-and-map on top of scaffolding.

​
Considered options

  1. One research-init skill with init (greenfield) and adopt (backfill) modes.
  2. Two separate skills.

​
Decision

Option 1. adopt inventories an existing repo and proposes mappings (PDFs/digests → bib+triage; benchmark data → datasets.yml; past results/specs → retroactive hypotheses; orchestration → experiment backend), materialized with human confirmation. mononet is the first adopt. Folded into the meta-spec (not its own sub-spec).

​
Consequences

  • One onboarding entry point; brownfield is init + a map phase.
  • Delegates per-item registration to the capability skills’ verbs.

​
Rejected alternatives

  • Two skills — they drive to the same end state; splitting duplicates the scaffolding logic.
meta-spec §6.