// the find
rohitg00/pro-workflow
Claude Code learns from your corrections: self-correcting memory that compounds over 50+ sessions. Context engineering, parallel worktrees, agent teams, and 17 battle-tested skills.
A Claude Code workflow plugin that puts a SQLite store underneath AI coding sessions so corrections you give the agent become FTS5-searchable rules that auto-load next session. On top of that: persistent research wikis, a budget-capped auto-research loop, 41 skills, 8 agents, 23 commands, and 37 hook scripts across 24 events. The core idea is solving a real problem; the execution is buried under a lot of marketing.
The self-correction loop is the right design — corrections land in SQLite with FTS5 indexing, auto-injected at SessionStart. You can audit the schema directly, nothing is magic. The budget-capped BFS research loop with a `touch ~/.pro-workflow/STOP` kill switch is the correct approach for autonomous research: runaway API costs are a genuine risk and this addresses it explicitly. The PreCompact/PostCompact hooks for state preservation through context compaction is a detail that actually matters in long sessions and most similar tools skip it entirely. LLM council uses Promise.allSettled so a single provider failure doesn't abort the deliberation — that's defensive programming done right.
The comparison table lists every competitor as 'No' for every feature, which is a marketing table not a technical one — the readme itself links to everything-claude-code which has 140+ skills, clearly overlapping. The primary install path includes `/plugin marketplace add rohitg00/pro-workflow`, which is not a real Claude Code command; there is no plugin marketplace with that syntax, so the install instructions are partly aspirational. 41 skills plus 37 hook scripts is an enormous automation surface area running on every session event, with one visible test file in `src/optimizer/__tests__/` — adopting this means trusting a large mostly-untested layer that intercepts tool calls, commits, and context compaction. Cross-agent install depends on skills.sh, a third-party commercial service; if that goes away, the install story for non-Claude agents breaks with it.