You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Tracking issue for the codebase-wide simplification + architecture refactor. Full context lives in docs/refactoring/: roadmap · architecture report (6 judge-endorsed structural proposals) · tactical report (~118 verified line-level items, ~5,000–5,500 LOC net reduction).
Sequencing principle: most tactical batches are the cheap first steps of the architecture moves. Deletions and dedups land before each structural move so there is less code to relocate, fewer symbols to export, and smaller test migrations. Every finding was adversarially verified against the code (166 raw → 151 confirmed → deduplicated).
Sequence
Phases are ordered; each unlocks or shrinks the ones after it. The parallel track has no ordering constraints.
Move-only PRs stay separate from behavior-change PRs. Three deliberate behavior changes each ship alone: closing the claude_cli auth-validation gap, shortSHA 7→12 in agents output, non-atomic→atomic JSON writes.
Run the architecture harness (internal/architecture, per-pkg imports_test.go) and full test suite per step; add boundary entries as new packages appear.
Phases 4 and 5 touch the two hottest files in the repo — land those steps quickly or during a lull.
Tracking issue for the codebase-wide simplification + architecture refactor. Full context lives in
docs/refactoring/: roadmap · architecture report (6 judge-endorsed structural proposals) · tactical report (~118 verified line-level items, ~5,000–5,500 LOC net reduction).Sequencing principle: most tactical batches are the cheap first steps of the architecture moves. Deletions and dedups land before each structural move so there is less code to relocate, fewer symbols to export, and smaller test migrations. Every finding was adversarially verified against the code (166 raw → 151 confirmed → deduplicated).
Sequence
Phases are ordered; each unlocks or shrinks the ones after it. The parallel track has no ordering constraints.
internal/fsatomic, cobra Args constructors,configPath/JSON-render/progress-logger dedup,OpenLedger. The vocabulary later moves are written in.internal/appcomposition root. Finishes Architecture refactor 02: extract command-independent runtime assembly #422/Architecture refactor 03: move reusable app contracts out of cmdruntime #423 knownDebt. Keystone: unblocks Phases 3 and 6.internal/app(needs Phase 2). Includes llm subprocess/stream dedup riders.execute()decomposition → file split → extractinternal/workbench, absorb intollmlifecycle/runartifact, extractinternal/dossier.internal/cmd/initcmd. After Phase 0 (~1,700 LOC lighter). Editor-family consolidation → tactical batch → mechanical move → prompter collapse.--cr-bin.Ground rules
claude_cliauth-validation gap,shortSHA7→12 in agents output, non-atomic→atomic JSON writes.internal/architecture, per-pkgimports_test.go) and full test suite per step; add boundary entries as new packages appear.Suggested first three PRs
internal/fsatomic+ migrate the three JSON writers (Refactor Phase 1: shared leaf packages + mechanical boilerplate (fsatomic, cobra Args, view/progress helpers) #448)SelectionRuntimeintointernal/app, repoint benchmarkcmd, delete the Architecture refactor 03: move reusable app contracts out of cmdruntime #423 knownDebt entry (Refactor Phase 2: extract internal/app composition root (finishes #422/#423) #449)