1.6 KiB
1.6 KiB
Changelog Generation
After the session completes (pre-submit review and adversarial review both pass), generate .codeflash/changelog.md from the experiment history. This file can be used directly as a PR description body.
Input sources
.codeflash/results.tsv— every experiment with status, metrics, and pattern.git log <base>..<branch> --oneline— commit messages for kept optimizations..codeflash/HANDOFF.md— key discoveries and session context.
Structure
Write .codeflash/changelog.md:
## Summary
<1-3 sentences: what was optimized and why, derived from the original user request>
## Optimizations
| # | Target | Pattern | Before | After | Improvement |
|---|--------|---------|--------|-------|-------------|
| 1 | function_name | antipattern-name | 2.3s | 0.8s | 65% faster |
| 2 | function_name | antipattern-name | 450 MiB | 280 MiB | 38% less memory |
**Commits:**
- `abc1234` — Replace list.pop(0) with deque in score_records
- `def5678` — Use __slots__ on SensorReading dataclass
## Key Discoveries
<Non-obvious findings from HANDOFF.md that reviewers should know>
## Test Plan
- [x] All existing tests pass after each optimization
- [x] No performance regressions in non-targeted benchmarks
## Session Stats
- **Experiments**: <total> (<keeps> kept, <discards> discarded)
- **Domains**: <list of domains optimized>
Use appropriate units per domain: CPU (seconds, speedup %), Memory (MiB, reduction %), Async (latency ms + throughput req/s), Structure (import time seconds).
After writing, print: [changelog] Written to .codeflash/changelog.md — <N> optimizations across <M> domain(s)