Welcome. You’re early. This document tells you what we’re testing, what’s known broken, and how to send feedback so we can iterate fast.
This is not the end-user walkthrough — that lives at docs/USING_YAKCC.md. Read that first; this file adds the alpha-specific context.
v0.5.0-alpha.0 is the first version of yakcc anyone outside the core team has used. The flywheel works — every Claude Code / Cursor session you run can hit registry atoms (registry-hit), pass through your own emission (passthrough), or atomize your novel code into the local registry for future sessions (atomized). The corpus grows as you use it.
What we’re testing:
docs/USING_YAKCC.md end-to-end should produce the experience it describes, without you having to read source or dig in Discord. If it doesn’t, that’s the most important bug class.yakcc query. If the round-trip is silent or surfaces wrong atoms, we need to know.yakcc_resolve returns a high-confidence match), hooked Claude/Cursor produces fewer output tokens than unhooked, at equal-or-better quality — B4-v5 measured 91% oracle pass under auto_accept. In raw aggregate the win is coverage-gated (auto_accept coverage 56–72% against the 6-atom corpus today); the gap closes as your corpus grows. Telemetry captures the data; we’ll ask for snapshots.yakcc seed --yakcc), early sessions on your code may see lots of synthesis-required. That’s expected; what matters is whether it improves week-over-week as your personal corpus grows.What we are not testing in this alpha:
yakcc.dev peer) does not exist yet. If you set up a team peer, that’s beta-territory work.yakcc bootstrap has a known 30+ minute regression on schema v9 (tracked at #377, fix imminent). Daily incremental shave caching is on the roadmap at #363.Primary install: clone from monorepo. This is the only path that gives every platform a
working yakcc today.
git clone https://github.com/cneckar/yakcc.git ~/.yakcc-cli
cd ~/.yakcc-cli
git checkout v0.5.0-alpha.0 # pin to the alpha tag
pnpm install --frozen-lockfile
pnpm -r build
export PATH="$HOME/.yakcc-cli/packages/cli/dist:$PATH"
yakcc --version # confirm
Optional secondary: build a standalone binary locally (Linux x64 only today — macOS and Windows arrive with #361 Slice 2). This is the Slice-1-of-#361 output; a downloadable GitHub Release artifact is not yet available.
# inside ~/.yakcc-cli (already cloned above)
pnpm --filter @yakcc/cli build:binary
# produces packages/cli/dist/yakcc-bin
cp packages/cli/dist/yakcc-bin ~/bin/yakcc # or any dir on your PATH
yakcc --version # confirm
Coming soon: downloadable binaries via GitHub Releases and multi-arch support land with #361 Slices 2-3.
Then in your own project:
cd ~/my-project
yakcc init # creates .yakcc/, .yakccrc.json, auto-detects + wires IDE hooks, seeds the bootstrap corpus
yakcc init is a single command (per DEC-CLI-INIT-002): it detects your installed IDEs (Claude Code, Cursor, Cline, Continue.dev), wires each hook, and seeds the yakcc bootstrap corpus (~4k atoms) in one step. No separate yakcc seed --yakcc call is needed.
Open Claude Code (or Cursor) in the project. Ask it to do real work. Watch the hook fire.
yakcc init Windows behavior is being verified at #385; if you hit a no-op yakcc init, that’s the bug — file with platform info.Things we know about, in priority order. You will hit some of these. That’s the point of the alpha.
| What | Status | Issue |
|---|---|---|
| Substitution-integration: 4 specific test cases fail | Under triage; classifying which (if any) hit real-world usage | #365 |
| Bootstrap perf: schema v9 write path 30+min vs prior 5min | Fix in flight, lands within ~3 hours of investigation | #377 |
Windows yakcc init may no-op | Under investigation; #274 fix may be incomplete | #385 |
| Standalone binary Slice 1 shipped (#361, local Linux x64 build only); multi-arch download via GitHub Releases pending Slices 2-3 | Slices 2-3 in flight | #361 |
No global registry peer (yakcc.dev) yet | Post-v0 by design; alpha = your local corpus + yakcc seed | #371 |
| Codex CLI hook not wired | Deferred to v0.5+ on demand | #220 (closed not-planned) |
| Recursive self-hosting proof not yet green | Surgical fix in flight; load-bearing for v2 narrative not v0 use | #355 |
Shave cache (--verify flag for byte-identical) not yet built | Daily-UX improvement; v0.5+ | #363 |
If you hit something not on this list, that’s the bug we want to hear about.
GitHub issues are the official channel. Use the alpha-feedback issue template:
https://github.com/cneckar/yakcc/issues/new?template=alpha-feedback.md
The template asks for:
yakcc --version or the git SHA you cloned)yakcc telemetry --tail 10 to get the last 10 events from your latest session)The telemetry sample is the most useful diagnostic we have. Please include it. It’s local-only, never leaves your machine until you paste it into a GitHub issue.
Note: telemetry lives in
~/.yakcc/telemetry/(your home dir), not in<project>/.yakcc/. Runyakcc telemetry --pathto confirm the location. Seedocs/USING_YAKCC.md §4dfor the full two-directory layout explanation.
We’re not promising hotfixes for cosmetic issues. We are promising to read every report.
v0.5.0-alpha.0 ships:
settings.json hook injection; Cline and Continue.dev marker surfaces installed (per DEC-CLI-HOOKS-CLINE-INSTALL-001, DEC-CLI-HOOKS-CONTINUE-INSTALL-001); intercepts Edit|Write|MultiEdit; routes through local registry first.docs/USING_YAKCC.md.v0.5.0-alpha.1: substitution-integration fixes (#365), bootstrap perf (#377), Windows bin fix (#385), starter-corpus polish.v0.5.0-beta.0: binary distribution multi-arch + downloadable (#361 Slices 2-4), shave cache (#363), recursive self-hosting proof closure (#355).v0.5.0: global registry peer (#371), OSS-libs shave campaign begins, B4 token-savings benchmark published.We’ll cut a new alpha tag for each substantive fix wave. Pin to a specific tag in your install instructions; we’ll announce when to move forward.
The most impactful thing you can do as an alpha tester, beyond filing bugs, is run the B3 cache-hit sprint — a structured 3+3+3 day experiment that measures the real-world registry-hit rate on your own code. The protocol is self-contained:
yakcc init followed by one IDE
session produces no telemetry, that’s a regression; file with your platform info and the output
of tail -20 ~/.yakcc/telemetry/*.jsonl.You’re testing software whose primary value-proposition is “every session grows the corpus, and the corpus saves you tokens on the auto_accept path.” That value emerges over time, not on day 1. Your feedback during the awkward early period is what lets us reshape the rough edges before public launch.
We owe you working software in return. If something is broken in a way that wastes your time, tell us — that’s the most useful signal we can get.