crapkit

crapkit scores every function in your repo on complexity times uncovered risk, ranks the worst ones by how often the file changes, and blocks commits that add more. It reads Python, TypeScript, TSX, JavaScript, Swift, Go, Rust, shell, PowerShell, C and C++, Objective-C, Vue, Java and Zig through lizard, and joins per-function branch coverage from the istanbul or coverage.py artifact your own test command already writes. Every read command speaks sorted-keys JSON on a pinned schema, because half the callers are coding agents.

CRAP = ccn^2 * (1 - cov)^3 + ccn

The name is not ours: C.R.A.P. (Change Risk Anti-Patterns) was coined for crap4j by Alberto Savoia and Bob Evans in 2007.

ccn is the smaller of standard and modified cyclomatic complexity, both read off one lizard pass. cov is branch coverage inside the function's span; with no branches it falls back to statement coverage, and with no statements to invoked-or-not, so a half-executed straight-line function never reads as fully covered.

Above the ceiling, coverage cannot save you. Decompose. At the default target of 6, a function at ccn 7 with 100% coverage still scores 7 and still fails the gate. The only move that clears it is splitting the function.

crapkit scores git-tracked files only. Source you have not git added is invisible to it.

The 60-second start

pip install crapkit
cd your-repo
crapkit init        # crapkit.toml and .gitignore lines, plus a live coverage lane when it
                    # recognizes the runner and a scope speaks its language: pyproject.toml,
                    # pytest.ini or setup.cfg for pytest; a test script or vitest/jest in
                    # package.json for the JS side
                    # without one: the lane comes commented out, init says to declare one,
                    # and docs/lanes.md is how to fill it in
crapkit coverage    # runs the lane, joins coverage, stores a scored run
crapkit worklist    # the ranked risk map
crapkit ratchet seed && git add crapkit.toml crapkit-ratchet.tsv .gitignore

coverage scores, worklist ranks:

$ crapkit coverage
run 1 @ fae4db93108: 2 functions scored — 2 measured / 0 untested / 0 no-lane / 0 cc-only, 1 over target 6, CRAP load 41.0, grade F

$ crapkit worklist
worklist @ fae4db93108 (run 1, floor ccn>=5, churn 12mo) — 1 active, 0 dormant
  risk      0.0  ccn  14 ( 14 std)    1c/1a w   0.00  calc/grade.py:7  classify( score , attempts , late , bonus )

risk 0.0 is what a one-commit repo scores, because churn needs a spread of commits to rank and ccn order stands in until then (Risk).

ratchet seed signs today's debt at today's score. From then on marks only ever fall, so the repo can get better and never worse while you burn it down.

One thing stops most first runs: the coverage plugin. init writes a lane that shells out to your own test runner, and the runner needs its coverage package installed: pytest-cov for pytest, @vitest/coverage-v8 (pinned to your vitest major) for vitest. Without it the lane produces no artifact and coverage exits 5 quoting the runner's own error. For pytest, init probes the python its lane will run and prints the install command when pytest_cov is missing; pip install "crapkit[py]" pulls the plugin alongside crapkit when the two share a venv. On a Windows PATH holding only the py launcher it writes py, not a python3 the lane could never run, and when cmd.exe cannot start the interpreter at all (exit 9009, the Store alias) it names that instead of guessing at pytest-cov. A repo that pins no lockfile and carries its own .venv gets that venv's interpreter in the lane, when that interpreter can import pytest, rather than whichever python the shell answers with. The two quickstarts below walk a real repo end to end.

On Windows a lane command is read by cmd.exe, the shell that will run it, not by sh. Double quotes are the portable quoting. A single-quoted value is refused at config load with exit 3, because cmd.exe would hand pytest five words and the lane would write no artifact:

# the lane in crapkit.toml
command = "python -m pytest -m 'not live and not perf' --cov=calc --cov-branch --cov-report=json:.crapkit/cov/py.json"

$ crapkit doctor
crapkit: lane 'py': positional argument 'live' narrows a full-suite coverage run; drop it, attach it to the flag it belongs to (-n8, --numprocesses=8), or set full_suite = false deliberately (cmd.exe does not treat ' as a quote: write the value in double quotes); a suite whose testpaths cannot be collected in one process needs one lane per testpath, each with full_suite = false and its own artifact

Write it -m "not live and not perf". Carets, && and | segments, redirections and empty quoted arguments all read the way the shell reads them, so a chained lane (cd tests && python -m pytest --cov ...) is checked one segment at a time. doctor reads a lane the same way, and FAILs one whose runner will not start.

Related MCP server: PhpCodeArcheology

Install

pip install crapkit

That is the release on PyPI. For the unreleased tip of main, or from a local clone (run at the clone root):

pip install git+https://github.com/JeanFrancoisGagne/crapkit.git
pip install .

Every route pulls one dependency, lizard>=1.24.0, a normal PyPI wheel, so an offline mirror installs fine. Requires Python 3.11 or newer. The pip install -e ".[dev]" under Development is a different thing: it adds the test extra, for people changing crapkit.

Scoring runs your own test command on your own machine and reads the artifact it writes. There is no network call anywhere in crapkit, so no source, no score and no telemetry leaves the box (SECURITY.md).

$ crapkit --version
crapkit 0.4.15

python -m crapkit works identically to the console script and is what to use from a source checkout. Every subcommand accepts --repo PATH (default: the current directory), so you never have to cd into the repo you are scoring; Subcommands shows where the flag goes.

Upgrading from 0.4.4

Run crapkit ratchet seed first. Shell cognitive complexity now nests, which is analysis version 8, and marks measured under version 7 are not comparable. Until you re-seed, verify refuses at exit 3:

$ crapkit verify
crapkit: ratchet marks were recorded under [crapkit-analysis=7 lizard=1.24.0] but this run measures [crapkit-analysis=8 lizard=1.24.0] — CRAP scores are not comparable across metric versions; re-baseline with `crapkit ratchet seed`

Only shell and PowerShell cognitive numbers move. ccn does not, so a re-seed re-stamps the file and leaves the marks where they were.

Five more things change under you. Three of them need nothing from you:

• New cache files. .crapkit/coupling-cache-v1.json joins churn-cache-v2.json and churn-log-v2.z. A warm 0.4.4 churn cache is adopted once and its file removed, and .crapkit/ is already gitignored, so nothing new reaches your index.

• trend and report write. Both read a per-run rollup table, filled once per run and pruned with its run, instead of rescanning every scored row. A read-only .crapkit/ costs the speedup, never the command.

• Nested scopes may move files. One predicate decides scope ownership now, and the deepest declared path wins, so a repo whose [[scope]] paths nest inside each other can see files change scope, rollup and ceiling on the next scan. Scopes that do not nest see no change.

The other two put something in front of you:

• mutate keeps a worktree pool. With mutation_workers > 1 the worker worktrees now live under .crapkit/mutate-pool/ between runs and are re-prepared each run, which is the setup cost gone (30.6 s to build four on a 31,459-file repo, 0.46 s to re-prepare them). The pool is not size-bounded and nothing sweeps it: crapkit mutate --drop-pool removes it and exits. Single-worker runs are untouched.

• doctor WARNs on a lane with no results_artifact. Every coveragepy or istanbul lane written before 0.4.5 gets one, with the two lines that fix it. Coverage is unaffected. What the lane cannot feed without a results file is the crashed-worker check and the no-new-failures check (exit 8).

The exe lock on Windows

uv tool upgrade crapkit, and pip install -U into a tool venv, fail with os error 32 ("The process cannot access the file because it is being used by another process") while a crapkit MCP server is live: an agent session spawns crapkit.exe mcp, which holds the launcher, and Windows will not overwrite a running executable. The venv upgrades before that copy fails, so crapkit --version already reports the new version and only the launcher is stale. Quit the agent session and rerun the upgrade, or rename the locked exe aside (Windows allows renaming a running one) and copy the new one in. Two lines in cmd.exe, where both % variables expand:

move %USERPROFILE%\.local\bin\crapkit.exe %USERPROFILE%\.local\bin\crapkit.exe.old
copy %APPDATA%\uv\tools\crapkit\Scripts\crapkit.exe %USERPROFILE%\.local\bin\crapkit.exe

Git Bash has no move and passes %APPDATA% through as literal text, so that block fails there on its first line. Its form is mv and cp over "$USERPROFILE" and "$APPDATA", which Git Bash sets to the same two directories.

The Claude Code plugin

claude plugin marketplace add JeanFrancoisGagne/crapkit
claude plugin install crapkit@crapkit

Two commands, installed once per user, and every repo on the machine gets it. The plugin ships three skills, the read-only MCP server, and one advisory PostToolUse hook that names any function an edit pushed over its ceiling. Claude reaches two of the skills by itself, crapkit and crapkit-recover; the third you type, as /crapkit:crapkit-onboard, because wiring a repo up happens once and its description has no business in every turn's window. It adds no files to your repo, and it needs the crapkit CLI on PATH.

A repo with no crapkit.toml costs a silent sub-50 ms no-op per edit. Other agent runtimes have no marketplace: copy plugin/skills/* into their skills directory instead.

The hook registers on Edit|Write, which is every write that names a file. An agent that writes its source through a shell heredoc names none, so a Bash event is judged off the working tree instead. That half is yours to register, because it costs two git spawns per shell call. Add a second PostToolUse entry to your own settings, same command, matcher Bash:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          { "type": "command", "command": "crapkit claude-hook --protocol 1", "timeout": 20 }
        ]
      }
    ]
  }
}

The cost is one git rev-parse --show-toplevel and one git status --porcelain -z -uall per shell call in any git repo, whether or not crapkit measures it: about 30 ms together on crapkit's own checkout, and more on a bigger tree. What comes back is the dirty or untracked *.py files written in the last 12 seconds, 25 at most, each judged the way an edit is. Python only, so a TypeScript or Go repo pays the two spawns and hears nothing.

Languages

14 languages, two coverage parsers. Coverage joins where a parser exists; everything else scores on complexity alone.

A cc-only scope declares coverage_optional = true, scores crap = ccn, and needs no lane. Nothing about it is provisional: the ceiling still binds and the gate still refuses a function over it. Add a coverage lane the day a parser exists and the same scope starts joining coverage.

crapkit init writes that key itself, on every scope whose languages all lack a parser, and leaves it off any scope a lane could still measure. So the 60-second start above runs unchanged on a Go, Rust or shell repo: crapkit coverage scores it with no lane at all, and that run is the baseline worklist, next-item, ratchet seed and verify read.

Three readers are crapkit's own. lizard ships none for shell or PowerShell, so crapkit counts their functions itself. Its Rust reader scores a 7-arm match as ccn 2 (filed as lizard #494), so crapkit counts each non-wildcard arm like a C case and retires the override the day upstream fixes it. The cognitive column charges that same block once, the way Sonar charges a switch.

The gate

Four surfaces ask the same question, ccn against the scope's ceiling, with four different powers:

Both hooks exempt a function the committed ratchet already carries a mark for, so touching signed debt never refuses a commit. verify is what fails a mark that rises. Since 0.4.5 its gate exempts a touched function whose fresh CRAP sits at or under its mark, the rule rescore --gate already applied; push it past the mark and the gate fires again. The pre-commit hook still exempts on the mark's existence alone, on purpose: a staged blob has no coverage, so there is no fresh CRAP to compare against. It reports each exemption count on stderr (staged function(s) carry a ratchet mark and were not gated), and says the same about a staged file no [[scope]] claims, so a new top-level directory cannot go ungated in silence.

The crapkit root does not have to be the git top. Since 0.4.5 every git spawn runs with diff.relative=true and core.quotePath=false, so a crapkit.toml in packages/api gates that package's own staged files and names them app/m.py, not packages/api/app/m.py, and a dirty non-ASCII path is a real row rather than an invisible one. Before that a nested root matched staged paths against no scope, and a function at twice the ceiling committed with a warning.

Git runs hooks outside your shell's activated venv. Bare python must resolve to an interpreter that has crapkit installed, or spell it out (exec /path/to/venv/Scripts/python -m crapkit hook-precommit).

Route 1: .git/hooks/pre-commit (local, not committed)

cat > .git/hooks/pre-commit <<'EOF'
#!/bin/sh
exec python -m crapkit hook-precommit
EOF
chmod +x .git/hooks/pre-commit

Route 2: a committed hooks directory

The whole route, from a repo that has no githooks/ yet:

mkdir -p githooks
cat > githooks/pre-commit <<'EOF'
#!/bin/sh
exec python -m crapkit hook-precommit
EOF
chmod +x githooks/pre-commit
printf 'githooks/pre-commit text eol=lf\n' >> .gitattributes
git add .gitattributes githooks/pre-commit
git update-index --chmod=+x githooks/pre-commit
git commit -m "add crapkit gate hook"
git config core.hooksPath githooks

The --chmod goes between the add and the commit. It writes the executable bit to the index, so a commit that already happened does not carry it: run it after and git ls-tree HEAD still says 100644, which is a hook Unix checkouts silently skip. The .gitattributes line is the harder half of the same failure: under Windows' default core.autocrlf the hook checks out CRLF and #!/bin/sh\r dies on Linux and macOS with a bad-interpreter error. crapkit doctor warns when a file under core.hooksPath is not 100755 in the index and prints the update-index line for it.

Git will not read a hooks path out of a committed file, so that git config line belongs in your CONTRIBUTING setup steps. Every clone arms the gate with it.

Route 3: the pre-commit framework

crapkit ships a .pre-commit-hooks.yaml declaring id: crapkit-gate. In your .pre-commit-config.yaml:

repos:
  - repo: https://github.com/JeanFrancoisGagne/crapkit
    # crapkit's release step rewrites this line to the tag it just cut
    rev: v0.4.15
    hooks:
      - id: crapkit-gate

That file arms nothing on its own. The framework writes .git/hooks/pre-commit when you tell it to, and until then git commit runs no gate and says nothing:

pip install pre-commit
pre-commit install

pre-commit install is the line every clone needs, the way Route 2 needs its git config core.hooksPath line.

rev is a git ref pre-commit resolves against that remote. Pin a release tag, not a branch: pre-commit autoupdate only moves between tags, and a moving main would change your gate under you.

Route 4: CI

A CI job runs on a fresh clone, which has no .crapkit/ store, so bare crapkit verify exits 1. Running coverage first would make the PR's own tree the baseline, a gate that can never fail. The portable baseline is the mechanism:

# on the default branch, after a passing verify: commit this file
crapkit verify --emit-baseline crapkit-baseline.tsv

# in the PR job, against the committed baseline
crapkit verify --baseline-tsv crapkit-baseline.tsv --github

--github emits ::error file=... annotations that land on the PR diff; --sarif PATH writes SARIF 2.1.0 for code-scanning upload. Refresh the committed baseline whenever the default branch's verify passes.

Two things the job has to do before those lines run. Install crapkit, pip install crapkit, and pin the version the way Route 3 pins rev: an unpinned install moves your gate on whatever day a release lands. Fetch the whole history. actions/checkout clones one commit by default, verify reads the diff against the baseline's commit out of git, and a shallow clone does not have that commit:

$ crapkit verify --baseline-tsv crapkit-baseline.tsv
crapkit: baseline commit a74260f321f is not an ancestor of HEAD (rebase or amend rewrote history) — run `crapkit coverage` for a fresh baseline

That is exit 4 on a git clone --depth 1 of a repo whose baseline verifies at full depth. Set fetch-depth: 0 on the checkout step, which is what crapkit's own .github/workflows/ci.yml does.

The whole PR job, on GitHub Actions:

on: pull_request
jobs:
  crapkit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0        # verify needs the baseline's commit
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - run: pip install crapkit
      - run: pip install -e ".[dev]"   # your own test dependencies
      - run: crapkit verify --baseline-tsv crapkit-baseline.tsv --github

The second install is the one people leave out. verify reruns your lanes, so the job needs whatever your test command needs: the coverage plugin, npm ci, a database, all of it. Without them the lane writes no artifact and verify exits 5 quoting the runner's own error, which is a broken job and not a verdict.

What a refusal looks like

$ git commit -m "add route"
crapkit gate: 1 staged function(s) exceed the complexity ceiling of 6:
  ccn   7  app/m.py:9  route( a , b , c , d )
decompose before committing (coverage cannot save a function above the target).

That commit exited 1, not 6. Git collapses any failed hook to 1, so 6 is a code you only ever see by running the hook yourself: crapkit hook-precommit exits 6 on a violation and 0 otherwise. The stderr block above is the same either way.

CRAPKIT_OVERRIDE_REASON is not a bypass. Setting it routes the commit through the full three-record audit: an alert line through alert_command, a ratchet entry staged into the commit, and a row in the override log. All three land or nothing does, and an unset alert_command refuses the override outright. See docs/ratchet.md.

The GitHub Action

action.yml at this repository's root is a composite action, so a reviewer sees crapkit's numbers on the pull request without installing anything. Four lines add it to a workflow, and every input has a default:

      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: JeanFrancoisGagne/crapkit@v0.4.15

The whole job those four lines sit in:

on: pull_request
jobs:
  crapkit:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write             # the comment, and nothing else
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0               # the diff, and verify's baseline commit
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"       # the interpreter the install below lands in
      - run: pip install -e ".[dev]"   # whatever your lanes need to run
      - uses: JeanFrancoisGagne/crapkit@v0.4.15
        with:
          gate: "false"

That pip install step is the one people leave out, and it is the same one Route 4 above names: the action installs crapkit and nothing else, so your lanes still need whatever your test command needs. Without it the lane writes no artifact and the comment says so.

fetch-depth: 0 is the other one. actions/checkout clones a single commit; the action reads the pull request's changed files out of git and verify reads the diff against the baseline's commit. With a shallow clone the file list comes back empty and the comment ranks the whole repository instead of the diff.

The action installs crapkit from $GITHUB_ACTION_PATH, which is its own checkout of the ref you pinned in uses:. So a pin left at last month's tag scores your tree with last month's crapkit rather than with whatever released since, and pinning a tag is the whole version policy; the snippets above name the current release.

What the comment looks like

One comment per pull request, edited in place on every push. A hidden <!-- crapkit-action --> line is how the next run finds it, so a fifteen-push branch carries one comment and not fifteen. On a push event there is no pull request to carry it, and the same text goes to the job log instead.

Rendered against this repository's own store, with a base commit fifteen back standing in for a pull request:

<!-- crapkit-action -->

## crapkit

1333 functions in 61 files, 0 over target, CRAP load 3713.89, grade A+.

**verify passed.** Run 2 against baseline 1, 7 changed files.

### Worklist: 39 changed files

| File | Function | ccn | risk | remedy |
|---|---|---:|---:|---|
| `src/crapkit/cli/admin.py:1026` | `_recorded_roots( recorded )` | 6 | 10.9908 | ok |
| `src/crapkit/lanes.py:646` | `build_retest_command( template : str , tests : set [ str ] )` | 6 | 10.8234 | ok |
| `src/crapkit/cli/admin.py:80` | `_next_step( scopes : dict , lanes : tuple )` | 5 | 9.159 | ok |
| `src/crapkit/cli/admin.py:239` | `_pytest_cov_probe( command : str )` | 5 | 9.159 | ok |
| `src/crapkit/cli/admin.py:505` | `_segment_problems( name : str , cwd : Path , tokens : list [ str ] )` | 5 | 9.159 | ok |

The rows are the ranked worklist for the files the pull request changed, worst first, top of them. risk is ccn times churn weight, the number crapkit worklist ranks on, and remedy is the run's own verdict for that function: decompose, add-tests or ok. A pull request that touches no ranked function gets the heading and no table.

The two file counts describe the same diff, counted twice. 39 changed files is git diff --name-only base.sha...HEAD, the branch's own commits, and it is what the table is filtered to. The count on the verdict line is what verify measured from the same fork point. With delta: "false" the second one is 0, because there is nothing behind the checkout to measure from.

The inputs

gate: "false" is the default on purpose. A team adopts the action before it has decided which findings should stop a merge, and a check that fails on day one gets turned off on day two.

What the verdict line covers

On a pull request, the commits the pull request adds. The action scores the fork point first, then the checkout, then runs crapkit verify --base <fork>, which measures the diff from there and takes the fork point's run as its baseline. So the gate judges the functions in the diff a reviewer is reading, and a repository that was already over its ceiling before the branch started does not fail every pull request that touches it.

The fork point is git merge-base of base.sha and HEAD, not base.sha itself. base.sha is the base branch's tip when the event fired, so a base branch that moved after the branch forked carries commits HEAD never saw, and a run there would be neither the baseline verify wants nor a diff anyone is reviewing.

The base run happens in a detached worktree under RUNNER_TEMP, and its store is copied over the checkout's so both runs sit in one place. The cost is two lane runs on a pull request: your suite runs once at the fork point and once on the checkout. Set delta: "false" to skip the base run, and the verdict falls back to the checkout against its own run, which reports the tree's own health and judges no changed function.

Three things leave the base run unmade, and none of them fails the job: a shallow clone that does not hold the fork point, a fork point older than your crapkit.toml, and a lane that will not run against that tree. The step logs crapkit base scoring exited N and the verdict falls back the same way delta: "false" does. A push event never makes one, because there is no base commit and no pull request to comment on.

One requirement the base run adds: the lane has to measure the tree it runs in. A lane that reaches an installed copy of your package instead of the checkout will measure the pull request's code while standing on the base commit, and the two runs then describe the same tree. crapkit verify refuses a run whose artifact names files outside the tree (exit 5), which catches the loud version of this; a lane pinned to a path outside the worktree is the quiet one. Point the lane at the tree, or set delta: "false".

--reuse-artifacts is what keeps each of those runs to one pass of your suite. coverage ran the lanes moments earlier on that tree, and verify parses those artifacts rather than running the whole suite a second time for the same numbers.

The other gate that judges a delta is the portable baseline in Route 4: commit crapkit-baseline.tsv on the default branch and run crapkit verify --baseline-tsv crapkit-baseline.tsv in a step of your own. It needs no second lane run, and it needs someone to keep that file current.

The comment is posted with gh api and the job's own GITHUB_TOKEN, which needs pull-requests: write. Two things it cannot do: a pull request from a fork gets a read-only token, so the POST is a 403 there, and a self-hosted runner without the gh CLI on PATH fails that step. Both leave the rendered text in the job log.

Subcommands

Every subcommand takes --repo PATH (default .), and the flag goes after the subcommand. claude-hook is the one exception: it has no --repo, because it takes its root from the file named in the hook payload it reads.

$ crapkit worklist --repo /path/to/repo --scope util --top 1
worklist @ a7c5c85ac37 (run 1, floor ccn>=5, churn 12mo) — 1 active, 0 dormant
  risk      5.4  ccn   5 (  5 std)    5c/1a w   1.08  util/stats.py:1  bucket( value , low , high )

Before it, argparse reads the path as the subcommand name and exits 2 without ever mentioning --repo:

$ crapkit --repo /path/to/repo worklist --top 1
crapkit: error: argument command: invalid choice: '/path/to/repo' (choose from 'inventory', 'coverage', ...)

--json prints one sorted-keys JSON object on stdout, always carrying a schema field.

Reading the output

Flags: why a coverage number is missing

The coverage summary counts all four as measured / untested / no_lane / cc_only.

Remedy: what to do about it

Grade and CRAP load

The grade is the share of functions over their ceiling: A+ at exactly zero, A under 2%, B under 5%, C under 10%, D under 20%, F at 20% or more. crap_load beside it is the plain sum of every function's CRAP score, so it moves when a function gets better even if the letter does not.

Risk: what ranks the worklist

risk = ccn * churn weight. The weight is a time-weighted sum over the file's commits in the churn window: each commit contributes a logistic weight rising to 0.5 for the newest commit in the log and falling to near zero for the oldest, so five edits last month outrank fifty from two years ago. The window anchors on the newest commit, never on the wall clock, so a fixed tree ranks identically forever.

Age is not the input, position in the log is. A log whose commits all share one timestamp reads 0.0 everywhere, which is why a one-commit repo shows risk 0.0 on every row and falls back to ccn order. Commits minutes apart already rank. This repo was eight commits old, all made the same day:

$ crapkit worklist --scope util
worklist @ a7c5c85ac37 (run 1, floor ccn>=5, churn 12mo) — 3 active, 0 dormant
  risk      5.4  ccn   5 (  5 std)    5c/1a w   1.08  util/stats.py:1  bucket( value , low , high )
  risk      4.5  ccn   9 (  9 std)    1c/1a w   0.50  util/curve.py:1  curve( scores , mode , floor , ceiling , skip_none )
  risk      4.3  ccn   4 (  4 std)    5c/1a w   1.08  util/stats.py:13  spread( values , cap )  ok

bucket at ccn 5 outranks curve at ccn 9 because five commits touched it and one touched curve. That is the whole point of weighting by churn. spread carries the ok marker: already at or under its ceiling, listed anyway, and next-item would not hand it out.

The list splits in two: active (files with commits in the window) and dormant (zero churn, kept out of the queue but counted). Two rules reach under the worklist_floor. A file whose churn weight sits in the top 10% is promoted down to ccn 3, which is why spread appears above at ccn 4. And a function over its ceiling is admitted whatever its ccn, so the floor can never hold back debt.

The trusted baseline

Every verify measures the working tree against one earlier run, the trusted baseline. crapkit runs list marks which one that is today.

Which runs qualify. A coverage run, or a verify that passed. A failed verify never qualifies, and neither does a partial run (a lane failed, so some scope fell back to no-lane) nor a hook override record, which carries no scored rows at all. In runs list, verdict=- marks a run that produces no verdict rather than one that failed: only verify renders a verdict. Four readers ask this one question and get this one answer: the baseline pick here, ratchet seed, prune, and the tighten damping that compares a mark against the same commit's previous run. A mark can no longer be signed off a run verify refused.

What advances it. Any qualifying run. coverage writes one wherever HEAD is, so a dashboard cron advances the baseline exactly as CI does. A passing verify advances it and tightens the ratchet on the way.

The taint rule. A failed verify recorded findings against a tree. Until some verify passes, runs made after that failure do not become the baseline: choosing one would move the comparison point past the findings, the flagged function would stop counting as touched, and nothing would look at it again. verify says which run it refused and falls back to the newest run in front of the failure.

$ crapkit runs list
run   1 @ 88012a148f6 2026-08-23T09:27:46Z coverage  verdict=-      lanes=py  baseline
run   2 @ 803bdde8556 2026-08-23T09:27:53Z verify    verdict=FAILED lanes=py
run   3 @ 803bdde8556 2026-08-23T09:28:02Z coverage  verdict=-      lanes=py

$ crapkit verify
warning: run 3 is not the baseline: verify run 2 FAILED with 1 finding(s) and no passing verify has cleared it since — measuring against run 1 @ 88012a148f6 instead, so those findings stay visible. Fix them, or pass `--baseline 3` to accept the newer run deliberately.
verify FAILED @ d89068de7f3 vs baseline 88012a148f6 (2 changed files)
  GATE  crap     72.0  ccn   8 cov 0%  calc/legacy.py:7  legacy_router( a , b , c , d , e )  -> decompose
  findings: 1 committed / 0 dirty (uncommitted edits and untracked files)

Run 3 is a coverage run somebody took on the tree run 2 refused, and it scores the same ccn-8 function. Without the rule it would have become the baseline, legacy_router would have stopped being a touched function, and that gate line would never print again.

The escape, twice. Fix the findings and let a verify pass, which clears the taint for good. Or accept the newer run on purpose with verify --baseline 3: an explicit id bypasses the rule, and the run history records which run the verdict used. Nothing here touches a repo that has never run verify: with no failure to protect, coverage alone always advances the baseline.

When the id you pass cannot serve. A --baseline ID naming a real run that is not a candidate says which run it is, why, and which ones can:

$ crapkit verify --baseline 3
crapkit: run 3 is an inventory run (no coverage was measured) and cannot serve as a baseline; trusted runs: 1, 2; pass `--baseline 2` for the newest

Exit codes

Exit 1 means one of three things

CI cannot tell a crash from a clean policy verdict on the code alone. Which one you got depends on the command:

verify reports the first of 6, 7, 8, 9 that fires, in that order. A gate violation and a ratchet regression together report 6. A run that takes any of them fails, so it neither advances the baseline nor tightens the ratchet, exit 9 included.

Quickstart: Python

A repo with calc/grade.py, tests/test_grade.py, and a pyproject.toml. Commit first; crapkit reads git ls-files. Install the coverage plugin first, because the lane init writes runs pytest --cov and those flags come from pytest-cov:

pip install pytest-cov

(pip install "crapkit[py]" pulls both at once when crapkit shares the suite's venv.)

If your suite drives its own CLI through subprocess.run, add [tool.coverage.run] patch = ["subprocess"] to pyproject.toml and keep coverage>=7.10.6: pytest-cov 7.0.0 dropped subprocess measurement, so without that key every entry point scores 0% and nothing warns. docs/lanes.md has the whole rule.

1. Scaffold the config

$ crapkit init
wrote crapkit.toml with 1 scope(s): calc
detected 1 lane(s) from this repo's own files: py — next: run `crapkit coverage`
added to .gitignore: .crapkit/, .coverage, __pycache__/

init sniffs tracked source into one scope per top-level source directory, and detects a coverage lane from what the repo already has: a pytest marker file (pyproject.toml, pytest.ini, setup.cfg) writes a live [[lane]], and so does a test script or vitest/jest in package.json. A lockfile beside them names the environment: uv.lock, poetry.lock, pdm.lock or Pipfile.lock makes the lane uv run python -m pytest … (and the matching run for the rest), because a bare python binds to whichever venv the shell has active rather than the one the repo pins — see The interpreter a lane binds to. Whatever it detects, it also leaves commented templates for the runners it did not find, and those carry the same launcher, so uncommenting one cannot hand the bare python back. Every lane it writes reports into .crapkit/cov/, which is why the .gitignore list is so short: see Where artifacts live.

[crapkit]
target = 6

[[scope]]
name = "calc"
paths = ["calc"]
languages = ["python"]

[exclude]
globs = ["**/node_modules/**", "**/dist/**", "**/build/**", "**/vendor/**", "**/*.test.*", "**/*.spec.*", "**/test_*.py", "**/*_test.py", "**/conftest.py", "node_modules/**", "dist/**", "build/**", "vendor/**", "*.test.*", "*.spec.*", "test_*.py", "*_test.py", "conftest.py", "*_test.go", "**/*_test.go", "*.config.ts", "*.config.js", "*.config.mts", "**/*.config.ts", "**/*.config.js", "**/*.config.mts"]

[[lane]]
name = "py"
command = "python -m pytest --cov --cov-branch --cov-report=json:.crapkit/cov/py.json --junitxml=.crapkit/cov/junit-py.xml --continue-on-collection-errors"
artifact = ".crapkit/cov/py.json"
results_artifact = ".crapkit/cov/junit-py.xml"
parser = "coveragepy"
scopes = ["calc"]

# Declare one [[lane]] per coverage command, then run `crapkit coverage`.
# [[lane]]
# name = "js"
# command = "npx vitest run --coverage --coverage.reportsDirectory=.crapkit/cov/js --coverage.reportOnFailure --reporter=default --reporter=junit --outputFile=.crapkit/cov/js/junit.xml"
# artifact = ".crapkit/cov/js/coverage-final.json"
# results_artifact = ".crapkit/cov/js/junit.xml"
# parser = "istanbul"
# scopes = ["<your-scope>"]

# `crapkit test-scoped FILES` runs one command per scope, with {files}
# replaced by that scope's files, each quoted.
[crapkit.scoped_tests]
calc = "python -m pytest {files} -q -p no:cacheprovider"

The last block is the one an agent loop needs. crapkit test-scoped exits 3 for a file whose scope declares no template, and AGENTS.md makes it step 4 of the burn-down loop. Every key is in docs/configuration.md.

2. Check the config against the repo

$ crapkit doctor
ok   config keys all recognized
ok   scope 'calc': 1 file
ok   every tracked source file belongs to a scope
ok   1 lane(s) declared
ok   lizard 1.24.0
doctor: no problems found

doctor prints one line per check and exits 1 only on a FAIL. WARN and note report and exit 0.

3. Score the repo, and read the queue

$ crapkit coverage
run 1 @ fae4db93108: 2 functions scored — 2 measured / 0 untested / 0 no-lane / 0 cc-only, 1 over target 6, CRAP load 41.0, grade F

$ crapkit worklist
worklist @ fae4db93108 (run 1, floor ccn>=5, churn 12mo) — 1 active, 0 dormant
  risk      0.0  ccn  14 ( 14 std)    1c/1a w   0.00  calc/grade.py:7  classify( score , attempts , late , bonus )

Columns: risk, ccn with the standard-only ccn in parentheses, <commits>c/<authors>a in the churn window with w<weight>, path:line, the function's long name, then a marker on rows the burn-down queue will not hand out (ok, no-lane).

worklist is the risk map, not a to-do list. It ranks finished rows too, so it does not empty when the burn-down does. next-item is the other view of that run: it drops the no-lane rows, ranks by crap, and its empty: true is the stop condition.

4. Take the top item

$ crapkit next-item
{"commit": "fae4db93108b4841a00959f9117430679e7250ca", "empty": false, "item": {"authors": 1, "ccn": 14, "ccn_std": 14, "cognitive": 13, "commits": 1, "cov": 0.5, "crap": 38.5, "end": 28, "est_splits": 3, "est_uncovered_paths": 7, "flag": "measured", "function": "classify( score , attempts , late , bonus )", "handle": "classify", "nesting": 8, "nloc": 22, "path": "calc/grade.py", "remedy": "decompose", "scope": "calc", "start": 7, "target": 6, "uncovered_lines": [9, 11, 15, 17, 19, 24, 25, 26, 27, 28]}, "run_id": 1, "schema": 1, "skipped_no_lane": 0, "stale": false}

remedy: "decompose", est_splits: 3 (this needs roughly three pieces to fit under 6), and uncovered_lines naming the ten lines no test walks. handle is the name form to pass back, and stale: false says the run still describes HEAD. Every field is in docs/agent-json.md.

5. Seed the ratchet

Arm the debt gate before fixing anything. ratchet seed records every over-target function at its current score, and from then on nothing may get worse.

$ crapkit ratchet seed
crapkit-ratchet.tsv: added 1, tightened 0 — 1 mark(s) vs run 1 (fae4db93108)

$ git add crapkit.toml crapkit-ratchet.tsv .gitignore && git commit -m "adopt crapkit"

6. Fix it and verify

Extract until every piece sits at or under the ceiling. Here classify became _validate, _adjusted, _band and a classify that only sequences them, with the table of cases pushed into parametrized tests. Commit the fix, then:

$ crapkit verify
verify OK @ 8d10c13303d vs baseline fae4db93108 (5 changed files)

$ crapkit coverage
run 3 @ 8d10c13303d: 5 functions scored — 5 measured / 0 untested / 0 no-lane / 0 cc-only, 0 over target 6, CRAP load 19.0, grade A+

CRAP load 41.0 to 19.0, grade F to A+. verify reruns the lanes and checks three things against the trusted baseline: every function the diff touched sits at or under its ceiling, no marked function got worse, and no test that passed in the baseline fails now. Exit 0 advances the baseline and tightens crapkit-ratchet.tsv in place, so the repaid mark leaves the file: follow up with git commit -am "ratchet: classify repaid". The full mark lifecycle is in docs/ratchet.md.

crapkit next-item now comes back empty: true with a reasons object saying which ending you got. That is most of the stop condition, not all of it: AGENTS.md states the whole rule and reads the rest of reasons.

Quickstart: TypeScript

A vitest repo with src/grade.ts and test/grade.test.ts.

1. Scaffold the config

$ crapkit init
wrote crapkit.toml with 1 scope(s): src
detected 1 lane(s) from this repo's own files: js — next: run `crapkit coverage`
added to .gitignore: .crapkit/

The lane init wrote is npm run test -- --coverage --coverage.reportsDirectory=.crapkit/cov/js --coverage.reportOnFailure --reporter=default --reporter=junit --outputFile=.crapkit/cov/js/junit.xml. It reads vitest's json reporter from .crapkit/cov/js/coverage-final.json; the reportsDirectory flag is what keeps that report out of your root. The junit half is the lane's results_artifact, which the crashed-worker and no-new-failures checks read; both reporters are named because --reporter=junit alone would replace the console output you watch the suite through. Anything that produces an istanbul coverage-final.json works; see docs/lanes.md for the jest and pytest recipes, a package one directory down, and a crapkit root below the repo top.

2. Install a coverage provider

This is the step that stops most TypeScript users. vitest ships no coverage provider by default. Without one, init and doctor are both happy and coverage dies with exit 5:

$ crapkit coverage
crapkit: lane 'js' FAILED: lane 'js' produced no artifact at .crapkit/cov/js/coverage-final.json (command exit 1); full log: /repo/.crapkit/lane-js.log; last output: $ npm run test -- --coverage --coverage.reportsDirectory=.crapkit/cov/js --coverage.reportOnFailure --reporter=default --reporter=junit --outputFile=.crapkit/cov/js/junit.xml

 MISSING DEPENDENCY  Cannot find dependency '@vitest/coverage-v8'

(exit 1)
crapkit: every lane failed (1 of 1); the errors are above

That failure writes no run. Every lane failed, so coverage exits before it opens a store: there is no .crapkit/crap.sqlite yet and the run ids below still start at 1.

Install the provider, and pin the major yourself. Unpinned, npm resolves the newest provider against your older vitest and refuses the tree:

npm i -D "@vitest/coverage-v8@<your vitest major>"

The artifact crapkit wants is coverage-final.json, written by vitest's json coverage reporter, which is on by default. If your vitest config sets coverage.reporter explicitly, keep "json" in the list.

vitest writes no coverage report at all when the run fails. The lane init wrote already carries --coverage.reportOnFailure, so a red test still produces the artifact. If you write the lane by hand, or you would rather keep the switch beside your other coverage settings, coverage.reportOnFailure = true in the vitest config does the same job; either one is enough. The full block is in docs/lanes.md.

3. Score the repo

$ crapkit coverage
run 1 @ 8bfbe613fcd: 2 functions scored — 2 measured / 0 untested / 0 no-lane / 0 cc-only, 1 over target 6, CRAP load 56.68, grade F

$ crapkit worklist
worklist @ 8bfbe613fcd (run 1, floor ccn>=5, churn 12mo) — 1 active, 0 dormant
  risk      0.0  ccn  15 ( 15 std)    1c/1a w   0.00  src/grade.ts:8  classify ( row Row )

classify is ccn 15 against a ceiling of 6: one function holding the late-and-retry penalty, the letter bands, the demotion rule and the null case.

4. Seed the ratchet and commit

ratchet seed records every over-target function at the score it has today, so nothing can get worse while you burn this one down.

$ crapkit ratchet seed
crapkit-ratchet.tsv: added 1, tightened 0 — 1 mark(s) vs run 1 (8bfbe613fcd)

$ git add crapkit.toml crapkit-ratchet.tsv .gitignore && git commit -m "adopt crapkit"

5. Fix it

Above the ceiling, coverage cannot help, so classify gets split rather than tested. penalty, band and demote come out as their own exported functions, and classify keeps the null case and the bonus:

export function classify(row: Row): string {
  if (row.score === null) {
    return "N/A";
  }
  let score = row.score - penalty(row.attempts, row.late);
  if (row.bonus && score < 90) {
    score += 3;
  }
  return demote(band(score), row);
}

rescore --gate judges that edit on complexity alone, before the slow step:

$ crapkit rescore src/grade.ts --gate
rescore vs run 1 @ 8bfbe613fcd (coverage STALE, complexity fresh)
   ccn   cov     crap  remedy     function
     5    0%     30.0  add-tests  src/grade.ts:22  band ( score )
     5    0%     30.0  add-tests  src/grade.ts:38  demote ( letter , row Row )
     4    0%     20.0  add-tests  src/grade.ts:8  penalty ( attempts , late )
     4   45%      6.7  add-tests  src/grade.ts:48  classify ( row Row )
     4   75%      4.2  ok         src/grade.ts:59  average ( scores Array )

Exit 0: every piece is at or under 6. The crap column is loud because its coverage half is still run 1's, from before three of those functions existed, and add-tests is the literal instruction for step 6.

6. Cover the new pieces

rescore --gate passed on complexity, not on coverage. penalty, band and demote are three functions no test has ever called, so each gets a table test:

describe("band", () => {
  it.each([[95, "A"], [85, "B"], [75, "C"], [65, "D"], [10, "F"]])(
    "scores %i as %s", (score, expected) => expect(band(score)).toBe(expected));
});

Run the suite once before the slow step:

$ npx vitest run
 Test Files  1 passed (1)
      Tests  21 passed (21)

Skip this step and step 7 fails rather than passes. Run on a copy of this repo with step 6 left out, verify reruns the lanes against the real tree and three functions the old suite never called come back over the ceiling:

$ crapkit verify
verify FAILED @ 0296156ff21 vs baseline 0e646697946 (1 changed files)
  GATE  crap     17.8  ccn   5 cov 20%  src/grade.ts:38  demote ( letter , row Row )  -> add-tests
  GATE  crap     12.4  ccn   5 cov 33%  src/grade.ts:22  band ( score )  -> add-tests
  GATE  crap     10.8  ccn   4 cov 25%  src/grade.ts:8  penalty ( attempts , late )  -> add-tests

7. Verify

$ crapkit verify
verify OK @ 2af3433d979 vs baseline 8bfbe613fcd (3 changed files)

$ crapkit coverage
run 3 @ 2af3433d979: 5 functions scored — 5 measured / 0 untested / 0 no-lane / 0 cc-only, 0 over target 6, CRAP load 22.0, grade A+

CRAP load 56.68 to 22.0, grade F to A+, and the mark seeded in step 4 is gone: verify dropped it once classify scored under the ceiling, rewriting the tracked crapkit-ratchet.tsv in place. Commit it with your change. Marks only ever fall.

A verify may also print warning: N changed line(s) have no coverage above its verdict; that block is advisory unless diff_uncovered_max is set (docs/configuration.md).

Documentation

crapkit.schema.json is the authority on the config file shape.

Development

pip install -e ".[dev]"
pip install pytest-xdist
git config core.hooksPath git-hooks
python -m pytest -q

pytest-xdist is not optional: tests/fixtures/mini_repo declares a lane that shells out to pytest ... -n 2, and without it that subprocess dies on an unrecognized -n. The git config line arms the complexity gate on your own commits. Same steps, with what each one buys, in CONTRIBUTING.md.

License

MIT. See LICENSE.