← ClaudeAtlas

openreflexlisted

Use OpenReflex's local execution memory for this project - check prior experience and suggested strategies before substantial coding tasks, declare a different approach, record verified outcomes, or inspect what has been learned. Use when an [OpenReflex] context or alert appears, or when the user asks what worked before, why a path was suggested, or how tasks have gone.
vishnu-77/openreflex · ★ 4 · AI & Automation · score 70
Install: claude install-skill vishnu-77/openreflex
# OpenReflex OpenReflex captures coding-agent work automatically through lifecycle hooks and learns from it. Everything stays on this machine. You do not need to do anything for capture to work; the tools below make its advice more accurate. ## When a task starts - If the conversation already contains an `[OpenReflex]` block, use it as advice: it lists similar past tasks, a suggested strategy with alternatives (strategies that are worse on every measure are marked dominated), an execution budget, likely relevant files, and lessons. Start by checking the listed files before broad searching, and aim to finish within the budget. - If there is no such block and the task is substantial (a bug fix, feature, refactor, or migration), call `get_execution_context` with a one-sentence description of the task. If the user gave limits ("keep it quick", "no more than 20 minutes"), pass them as `max_minutes`, `max_tool_calls` or `max_context_tokens`. - If you deliberately take a different approach than the suggested one, call `choose_path` with the strategy name (or your own name plus 2-4 short steps). Skip this when you follow the suggestion. ## During the task An `OpenReflex:` alert means repeated failures, repeated identical calls, stalled progress, runaway context growth, or work past the budget were detected. Each alert ends with a recommendation: - **continue**: the path is still worth it; fix the approach within it (re-read the error, question the current assumpti