What problem does it solve? After profiling a Lens Studio project, developers still face the slow, error-prone work of applying each performance recommendation by hand, verifying it did not break visuals, and measuring whether it actually helped. This Skill automates that loop: it consumes the ranked output of /specs-lens-perf-attribution and iteratively applies, validates, and commits or reverts each fix. ## Core Features & Use Cases - Iterative optimization loop: Dispatches a worker per ranked recommendation, applies the fix, checks visual parity (SSIM + histogram cosine) and experiential markers, re-measures with Perfetto, then commits kept fixes or auto-reverts regressions. - Safe git workflow: Always works on a new branch forked from the current HEAD, one commit per kept fix, never pushes, and hard-stops on a dirty tree, MCP failure, or three consecutive reverts. - Configurable stop conditions: Supports max-iterations, target ms/frame, cumulative improvement targets, wall-clock budgets, and minimum predicted-delta filtering. - Use Case: After running /specs-lens-perf-attribution on a Lens that misses its frame budget, invoke this Skill with the attribution directory to automatically halve VFX spawn rates, throttle update loops, and gate tracking — ending with a branch of measured commits and a report tying each commit to its ms/frame gain and parity score. ## Quick Start Run /specs-lens-perf-optimize with attribution-dir pointing at the output directory from /specs-lens-perf-attribution to start the iterative optimization loop.
