Watcher performance

pcr start is designed to run in the background indefinitely without you noticing it. On a typical multi-project setup it uses a few percent of one CPU core in steady state and writes a few KB / second to the local SQLite store.

Polling intervals

Bounded work

• –Project list is cached per-process and only re-read when ~/.pcr-dev/projects.json's mtime changes — turns ~30 disk reads/min into one metadata() syscall per call
• –Dirty file fingerprints are bounded — files larger than 50 MB are tracked by (size, mtime) only; everything else SHA-256s the first 256 KB plus the file size. A stray .log file in the working tree can't peg a CPU core.
• –Diff-tracker state file is only re-written to disk when at least one project's hashes actually moved — idle projects produce zero disk writes
• –Cursor session JSON (often 50–500 KB) is parsed once per session per minute and reused across save_completed_turn calls
• –TUI rendering clears only what it touches and pre-sanitizes prompt / response text before wrap so long lines never overflow the pane

What you can do if it ever feels heavy

• –Run pcr gc --drafts-older-than 30d to keep the local store small
• –Use pcr start --plain if you don't want the full-screen TUI in the watcher terminal
• –If a specific repo is producing huge dirty files (build artifacts, logs), add them to .gitignore so git status stops tracking them