Dead-code pruning report — comprehensive analysis of ~7,800 removable lines

[akashgit]

Summary

Comprehensive dead-code analysis across runners, workflow modes, tests, and documentation. Report-only — no code changes.

Key Findings

• 2,944 lines HIGH-confidence dead code: 3 dead runners (codex/bob/opencode), 1 dead agent prompt (evolver.md), ~1,960 dead test lines
• ~4,378 lines MEDIUM-confidence: 9 uncertain workflow modes (founder, study, refine, parallel-improve, 3 frontend-design variants, build/improve registry entries)
• 4,201 lines LOW-confidence: 10 contributed benchmark workflows (recommend keeping as plugins)
• Corrected false positives: refactory.py and cycle_analyzer.py flagged as dead by research — verification proved both actively used
• Critical dependency: build_workflow is base for design_workflow, improve_workflow is base for research_workflow — functions must be retained even if direct invocation is removed

Deliverable

docs/pruning-report.md — 10-section report with verification evidence, dead test inventory, 9-phase deletion order, and risk assessment.

Next Steps

Phases 1-6 (runner deletion) can proceed immediately. Phase 8 (mode consolidation) requires user confirmation.

[lambdabaa]

Let's split this up into sub-issues focused on individual areas called out in the pruning report. Let's make sure that we have appropriate test coverage (integration / e2e level) for each change to prevent regressions.

• Move all modes other than design and create into a community workflows repo
• CLI should be aware of community repo and be able to pull in contributed workflows automatically
• Transition path could be to keep community workflows in another repo