meta-retrospective

To detect anti-patterns in prompt engineering iteration history, analyze multi-iteration prompt backups and evaluation reports to identify degradation directions and systemic issues like symptom-driven fixes. The skill parses sequential diffs and execution metrics to pinpoint exactly when performance regresses.

Prompt performance degrades over multiple iterations due to drifting rules, conflicting edits, and testset overfitting. Analyzing prompt backups and per-iteration evaluation reports identifies these degradation directions and marks the exact turning points where optimization cycles collapse.

To run a retrospective analysis on prompt backups and evaluation logs, provide the current prompt path, a backup directory with at least two .bak files, and the evaluation reports directory. The analysis outputs structured JSON and Markdown reports detailing added, removed, and modified prompt sections.

Yes, you can quantify execution metrics from per-iteration evaluation reports by ingesting the reports and optional run logs to extract average, minimum, and maximum scores, low-scoring cases, tool call patterns, and error rates for comprehensive iteration analysis.

The minimum requirement for iteration analysis is at least two prompt backups stored in a .bak directory. Running a full retrospective after three or more iteration backups is recommended to effectively stop change-and-degrade cycles and identify systemic anti-patterns.

To stop simultaneous multi-variable changes from breaking your prompt workflow, use rule-based heuristics to detect this and 12 other built-in anti-patterns in your iteration history. The analysis generates a forced new directions file to break harmful optimization cycles and guide the next engineering loop.