build-perf-baseline

# Build Performance Baseline & Optimization Agent establishes repeatable measurements before optimization work starts. ## When to Use Agent uses this skill when: - Build slowness is real but root cause is not proven - User wants before and after evidence for performance work - No-op or warm builds look too expensive - Agent needs the correct entry point before deeper diagnostics ## When Not to Use Agent does not use this skill when: - Build system is not MSBuild based - Agent already knows the failing target, task, or incremental rule - User needs a narrow fix for one custom target ## Required Inputs | Input | Required | Notes | |---|---|---| | Build scope | Yes | Solution, project, or filtered subset | | Environment | Yes | Local machine, CI runner, clean image, or warmed cache | | Scenario set | Yes | Cold, warm, no-op, or all three | | Guardrails | No | Limits on server, graph, layout, or caching changes | ## Diagnostic Workflow | Step | Agent action | Test | Pass | |---|---|---|---| | 1. Define scope | Agent fixes one build command, one configuration, and one environment | Agent writes the exact command set | Same command set stays stable across all runs | | 2. Capture baseline | Agent runs cold, warm, and no-op builds with binlogs | Agent preserves timings and binlogs for each run | Agent has one comparable result per scenario | | 3. Record findings | Agent records total duration, restore time, rebuild breadth, and no-op cost | Agent compares scenarios side by side | Agent identifies one or more evidence-backed red flags | | 4. Apply first-line changes | Agent applies only changes supported by baseline evidence | Agent reruns the same scenario set | Agent sees measured improvement or rejects the change | | 5. Escalate deliberately | Agent routes unresolved problems to the next specialist skill | Agent maps each symptom to one follow-up skill | Agent avoids speculative tuning | ## Baseline Matrix | Scenario | Agent measures | Red flag | |---|---|---| | Cold build | Full restore, full compile, total wall clock | Restore or evaluation dominates unexpectedly | | Warm build | Small-change rebuild scope and duration | Too many projects or targets rerun | | No-op build | Immediate rebuild with zero source changes | Wall clock stays noticeably high or custom targets rerun | ## First-Line Optimization Map | Evidence | Agent uses | Expected outcome | |---|---|---| | Repeated CLI startup overhead | `msbuild-server` | Lower warm and no-op startup time | | Idle CPU during multi-project builds | `build-parallelism` guidance | Lower wall clock with the same graph | | Output collisions or cache confusion | Artifacts layout or `check-bin-obj-clash` guidance | Cleaner bin and obj isolation | | Warm builds touching too much of the graph | Dependency trimming or graph cleanup | Smaller rebuild blast radius | | No-op builds still expensive | `incremental-build` guidance | More skipped targets on second build | ## Validation | Check | Test | Pass | |---|---|---| | Scenario consistency | Agent reruns the same command in the same environment | Inputs remain unchanged across before and after runs | | Evidence preservation | Agent confirms binlogs exist for comparison | Each run has a retained binlog or equivalent timing record | | Improvement proof | Agent compares before and after timings | At least one target scenario improves measurably or the change is rejected | | Correct escalation | Agent maps unresolved issues to a follow-up skill | Next step matches the observed symptom | ## Common Pitfalls | Pitfall | Agent correction | |---|---| | Agent optimizes before measuring | Agent captures baseline data first | | Agent compares different scopes or machines | Agent keeps scope, configuration, and environment fixed | | Agent treats one fast cold build as enough evidence | Agent measures warm and no-op separately | | Agent jumps into deep diagnostics too early | Agent escalates only after baseline evidence exists |