Skip to main content

tackbox

tackbox logo

publish verify-release pypi

Every failure must report, propagate, or explain itself.

Coding agents write error handling that looks right and silently isn't: a swallowed exception, a fatal exit with nothing logged, a report with the cause stripped out. tackbox catches it the moment it's written: hooked into the agent's edit loop it flags the finding before the turn ends, and the same rules gate pre-commit and CI - one coverage bar for hand-written and agent-written code.

And there is no quiet way around any of it: no flags, no config. The only escape is an explicit // no-report: <reason> at the site - and the agent hook asks for your approval before a new suppression lands.

resp, err := client.Do(req)
if err != nil {
    return nil // looks handled; the failure just vanished
}
client.go:42: ERC001: err-branch must propagate, capture, carry the
error into a terminal exit, or carry `// no-report: <reason>` (err=err)

One command brings the whole stack across Go, Python, Java, JS, TS, Svelte, and Markdown - no go install, no npm i, no external opengrep:

uvx tackbox@latest lint .

The wheel is hermetic: a consumer needs only git on PATH (plus a Go toolchain if the repo has .go files, and a Java 17+ runtime if it has .java files) and, the first time a given engine version runs, network access to fetch the engine payload once. Rules roll out via @latest - a new safety rule reaches every repo on its next run.

What it catches

  • Swallowed errors - the catch {} or if err != nil { return nil } that makes a failure vanish. Every path must report, propagate, or carry an explicit // no-report: <reason>.
  • Silent exits - os.Exit, log.Fatal, System.exit, or a local die reached with an unreported error, so the process dies and your error tracker never hears about it.
  • Double reports - capturing an error and re-throwing it, so the same failure hits Sentry/glitchtip twice and drowns the signal.
  • Broken cause chains - a new exception thrown from a catch that drops the original (only its message survives), erasing the stack you'd actually debug from.
  • Leaked secrets - a fingerprint or report argument that names a secret or raw user input, quietly shipping tokens/PII into telemetry.
  • Silently killed tests - the it.skip with no explanation, the failing test reborn as test.todo, the it.only that quietly turns off the rest of the suite. Every skip must state a reason; focused tests are always an error.

Wiring into a repo

Call tackbox lint from the repo's dev.py lint, next to the project's own linters:

def lint():
    sh("uvx tackbox@latest lint .")
    sh("uv run ruff check .")   # project-owned, if Python

Pre-commit runs a single language-agnostic hook; dev.py check (= lint + test) decides what to scan:

# .pre-commit-config.yaml in the consumer repo
repos:
  - repo: local
    hooks:
      - id: dev-check
        name: dev.py check
        entry: python3
        args: [dev.py, check]
        language: system
        pass_filenames: false
        always_run: true

CodeClimate report

tackbox lint --codequality <path> also writes a CodeClimate-format JSON array of every finding to <path> (console output and exit code unchanged; the report is written even when findings exist). Wire it into GitLab CI as a codequality report so the MR widget renders the findings:

lint:
  script: uvx tackbox@latest lint . --codequality gl-code-quality.json
  artifacts:
    reports:
      codequality: gl-code-quality.json

Distribution

uvx tackbox@latest installs one small wheel; the engine payload is fetched separately and cached per version:

  • tackbox (thin) - the Python CLI (including the pyrules flake8 plugin), the erclint / erclint-opengrep binaries, the javalint.jar, the opengrep rule yamls, and the ESLint and markdownlint plugins and presets. Platform-specific, bumped on every push.
  • tackbox-engines (fat, ~350 MB unpacked) - the bundled Node runtime, the opengrep binary, and the vendored third-party node_modules. Published as a PyPI wheel but not a pip dependency of thin. On the first run for a given engine version, tackbox resolves the wheel via the PyPI JSON API, verifies its unpacked payload against the tree sha256 pinned in the thin wheel's engines.json, and unpacks it once into $XDG_DATA_HOME/tackbox/engines/<version>/ (default ~/.local/share/...; override TACKBOX_ENGINES_DIR). Every later thin version reuses that one copy, so a stream of @latest patch bumps never re-materializes the engines. Bumped only when an engine changes.

After the first fetch tackbox runs fully offline until the engine version changes. Platform wheels cover Linux x86_64/arm64 (manylinux), macOS x86_64/arm64, and Windows x86_64. engines.json in the thin wheel records the source, version, sha256, and license of every bundled binary and dependency; tackbox doctor fetches the store if absent and verifies the payload against it.

What the rules enforce

Covers ERC001-008 (Go, via erclint), JV001-007 (Java, via the native javalint engine), ERC006 fingerprint rules (Go, Python, JS, TS, via the opengrep wrapper), Python exception and test-skip rules (via the pyrules flake8 plugin), frontend swallow and test-skip rules (JS, TS, Svelte, via ESLint), and Markdown (MD001-059 + ASCII).

See go/README.md for the Go ruleset. The specs these rules implement (error-reporting-and-coverage, error-handling-frontend) live outside this repo (private notes); the public summary:

  • Every err != nil branch must propagate, capture, or carry an explicit // no-report: <reason> marker.
  • Common parser results that fall through to nil must capture or carry // parse-skip: <reason>.
  • Terminal exits (log.Fatal*, os.Exit, project-local die) must be preceded by a capture call or carry a // no-report: <reason> marker (e.g. for the normal os.Exit(0) at the end of main).
  • Bare return nil from a single-result function must carry // nil-return: <reason> or use (val, ok) / (val, err).
  • A single err-branch may not both capture and return err.
  • Capture-call arguments (message, tags, dedupKey) must not reference secret-named identifiers or raw user input.
  • A skipped test must state a reason: t.Skip("why") / t.Skipf, or // test-skip: <reason> above a bare t.SkipNow(). The same contract holds in every language (skip / todo / xfail / @Disabled); focused tests (it.only, fit) are an unconditional error.

The same model is enforced beyond Go:

  • Java (javalint, JV001-006) on a typed javaparser AST: JV001 swallow (every catch path must propagate, report, print, or carry // no-report), JV002 chain (a thrown exception must carry the caught as its cause), JV003 throwable (a catch of Throwable / Error must rethrow), JV004 useless-catch (a catch that only rethrows the caught unchanged - deleted, not annotated), JV005 exit (System.exit in a catch needs a preceding capture; port of ERC003), JV006 double-capture (no path may both report and rethrow; port of ERC005), and JV007 skip (@Disabled / @Ignore must carry a non-empty reason string).
  • Python exception and test-skip rules ship as the pyrules flake8 plugin (TBX codes). A skip reason is accepted in any of the natural forms: @pytest.mark.skip(reason=...), @pytest.mark.skipif(cond, reason=...), @pytest.mark.xfail(reason=...), pytest.skip(...), or @unittest.skip(...). contextlib.suppress is flagged as a cosmetic dodge of the swallow rule; the one allowlisted use is asyncio.CancelledError around await task after task.cancel(), where the CancelledError on the await IS the confirmation that the cancel propagated, not an error to log.
  • JS / TS / Svelte swallow and test-skip rules run under ESLint. A skip reason is accepted in the call itself: node:test options ({ skip: 'reason' } / { todo: 'reason' }) and Playwright's test.skip(cond, 'reason') / test.fixme(cond, 'reason').

No configuration

By design, the ruleset is a single non-negotiable bundle. There are no flags to disable individual rules. Suppressing a finding requires the explicit per-site marker (// no-report, // parse-skip, // nil-return, // test-skip) with a non-empty reason.

Capture helpers are recognized by origin, not by name: a Go call counts only when its callee resolves (type info / import) to the github.com/nikitatsym/tackbox/go/report package, a JS/TS call to tackbox/report, and a Java capture when the caught reaches a known logger sink (e.g. slf4j, java.lang.System.Logger) at ERROR / WARNING - tier-1. Every language also honors a function declared in a repo-root .tackbox-reporters file (file#function: reason) - tier-2. A declaration names a report sink - it is not an exclude: it disables no rule, and a declared call is honored only when the caught error flows into its arguments.

Agent hook (Claude Code)

tackbox hook wires the rules into an agent's edit loop. It reads a Claude Code hook event on stdin and dispatches by hook_event_name:

  • PostToolUse re-lints the edited file (Go: its package). On a finding it exits 2 with the finding on stderr, so the model sees it and fixes it in-loop. The authoritative gate stays pre-commit / CI.
  • PreToolUse asks for approval before a new suppression marker (// no-report, // parse-skip, // nil-return, // test-skip, // long-comment) or a new .tackbox-reporters line lands; removing one is free.

The hook is a no-op unless the edit's cwd is a git repo with a dev.py at its root. Wire it once, globally, in ~/.claude/settings.json:

{
  "hooks": {
    "PreToolUse": [
      {"matcher": "Edit|Write|MultiEdit",
       "hooks": [{"type": "command", "command": "uvx tackbox hook"}]}
    ],
    "PostToolUse": [
      {"matcher": "Edit|Write|MultiEdit",
       "hooks": [{"type": "command", "command": "uvx tackbox hook", "timeout": 120}]}
    ]
  }
}

uvx tackbox hook runs the cached tackbox (no @latest): the hook is fast in-loop feedback, not the authoritative gate.

Layout

dev.py                                 # lint / test / e2e / check (dev-script)
hygiene.py                             # dev.py lint hygiene (conflict/yaml/ws/newline)
go.mod                                 # Go module
package.json                           # npm package (ESLint plugin + report helper)
eslint.config.preset.js                # default config used by tackbox-eslint bin
bin/tackbox-eslint.js                  # ESLint CLI wrapper with bundled preset
bin/tackbox-mdlint.js                  # markdownlint wrapper with bundled preset
go/
  cmd/erclint/                         # native Go analyzers (ERC001-005, 007)
  cmd/erclint-opengrep/                # opengrep wrapper, embedded rule yamls
    rules/                             # multi-language ERC006 yamls
  analyzers/                           # per-rule go/analysis packages
  internal/                            # markers + AST helpers
  report/                              # Go capture helper (Sentry/glitchtip)
java/
  pom.xml                              # Maven module -> shaded javalint.jar
  src/main/.../javalint/               # typed-AST analyzer (JV001-006)
    rules/                             # per-rule checkers
js/
  eslint-plugin.js                     # ESLint plugin entry
  rules/                               # 12 frontend rules
  markdownlint-rules/                  # custom markdownlint rules
  report.js                            # browser capture helper (@sentry/browser)
  tests/                               # RuleTester + node:test
py/
  tackbox/                             # lint / hook / doctor CLI, cache, engines
    pyrules/                           # flake8 TBX plugin (python exception rules)
  tests/                               # pytest suite

Repo conventions

  • Versioned via git tags (vMAJOR.MINOR.PATCH); CI auto-bumps the patch tag on every green push to main and publishes the wheels. Consumers track @latest, never a pinned version.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

tackbox-0.1.54-py3-none-win_amd64.whl (11.9 MB view details)

Uploaded Python 3Windows x86-64

tackbox-0.1.54-py3-none-manylinux_2_28_x86_64.whl (11.6 MB view details)

Uploaded Python 3manylinux: glibc 2.28+ x86-64

tackbox-0.1.54-py3-none-manylinux_2_28_aarch64.whl (10.6 MB view details)

Uploaded Python 3manylinux: glibc 2.28+ ARM64

tackbox-0.1.54-py3-none-macosx_11_0_arm64.whl (10.9 MB view details)

Uploaded Python 3macOS 11.0+ ARM64

File details

Details for the file tackbox-0.1.54-py3-none-win_amd64.whl.

File metadata

  • Download URL: tackbox-0.1.54-py3-none-win_amd64.whl
  • Upload date:
  • Size: 11.9 MB
  • Tags: Python 3, Windows x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for tackbox-0.1.54-py3-none-win_amd64.whl
Algorithm Hash digest
SHA256 186903685934cc14215ad18baa9199d2858a0e5be851accf2468fec15f2e0937
MD5 0ea50e2a8a575f3cbf3bfb000e7b47bc
BLAKE2b-256 67f19d24def529a2cbace94cb147994f4d2104871b90e2c2575791cb847b4319

See more details on using hashes here.

Provenance

The following attestation bundles were made for tackbox-0.1.54-py3-none-win_amd64.whl:

Publisher: publish.yml on nikitatsym/tackbox

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file tackbox-0.1.54-py3-none-manylinux_2_28_x86_64.whl.

File metadata

File hashes

Hashes for tackbox-0.1.54-py3-none-manylinux_2_28_x86_64.whl
Algorithm Hash digest
SHA256 91650281f9e58c92c398ecfa5252e4aa30210e77ee1404078e9c4416aee68301
MD5 7a4bfbcd65208520f6980dd434aeed22
BLAKE2b-256 85d980c802c45f0fac8b0c675c868118f32568d72435dc642fb5b3c11059bd1c

See more details on using hashes here.

Provenance

The following attestation bundles were made for tackbox-0.1.54-py3-none-manylinux_2_28_x86_64.whl:

Publisher: publish.yml on nikitatsym/tackbox

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file tackbox-0.1.54-py3-none-manylinux_2_28_aarch64.whl.

File metadata

File hashes

Hashes for tackbox-0.1.54-py3-none-manylinux_2_28_aarch64.whl
Algorithm Hash digest
SHA256 202ebe7f40f622e9d31d530f2c98e953866cdf19df134dbcd7aa5f759c48ed14
MD5 204ed26c7a84a777f29a870552cf7f1a
BLAKE2b-256 1e937069b5b63ad7754a855bc7fd839c62a15e0d6fa1f1cec458e76679066ba0

See more details on using hashes here.

Provenance

The following attestation bundles were made for tackbox-0.1.54-py3-none-manylinux_2_28_aarch64.whl:

Publisher: publish.yml on nikitatsym/tackbox

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file tackbox-0.1.54-py3-none-macosx_11_0_arm64.whl.

File metadata

File hashes

Hashes for tackbox-0.1.54-py3-none-macosx_11_0_arm64.whl
Algorithm Hash digest
SHA256 e036518c5d678e737ea70af4aafbca51a71915f4d3269c3d77feb9c0a554e66b
MD5 d91b6b16a0b66fe2902181f7f24eeda6
BLAKE2b-256 ad9f147f995eb68c96f34100b05376d51578717486e6b0db344366c01c79085f

See more details on using hashes here.

Provenance

The following attestation bundles were made for tackbox-0.1.54-py3-none-macosx_11_0_arm64.whl:

Publisher: publish.yml on nikitatsym/tackbox

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

0.1.97

4 files

0.1.96

4 files

0.1.95

4 files

0.1.94

4 files

0.1.93

4 files

0.1.92

4 files

0.1.91

4 files

0.1.90

4 files

0.1.89

4 files

0.1.88

4 files

0.1.87

4 files

0.1.86

4 files

0.1.83

4 files

0.1.82

4 files

0.1.81

4 files

0.1.80

4 files

0.1.79

4 files

0.1.78

4 files

0.1.77

4 files

0.1.76

4 files

0.1.75

4 files

0.1.74

4 files

0.1.73

4 files

0.1.72

4 files

0.1.71

4 files

0.1.70

4 files

0.1.69

4 files

0.1.68

4 files

0.1.66

4 files

0.1.65

4 files

0.1.64

4 files

0.1.63

4 files

0.1.62

4 files

0.1.61

4 files

0.1.60

4 files

0.1.59

4 files

0.1.58

4 files

0.1.57

4 files

0.1.56

4 files

0.1.55

4 files

This release

0.1.54 This release

4 files

0.1.53

4 files

0.1.52

4 files

0.1.51

4 files

0.1.50

4 files

0.1.49

4 files

0.1.48

4 files

0.1.47

4 files

0.1.46

4 files

0.1.45

4 files

0.1.44

4 files

0.1.43

4 files

0.1.42

4 files

0.1.41

4 files

0.1.40

4 files

0.1.39

4 files

0.1.38

4 files

0.1.37

4 files

0.1.36

4 files

0.1.35

4 files

0.1.34

4 files

0.1.32

4 files

0.1.31

4 files

0.1.30

4 files

0.1.29

4 files

0.1.28

4 files

0.1.27

4 files

0.1.26

4 files

0.1.25

4 files

0.1.24

4 files

0.1.23

4 files

0.1.22

4 files

0.1.21

4 files

0.1.20

4 files

0.1.19

4 files

0.1.18

4 files

0.1.17

4 files

0.1.16

4 files

0.1.15

4 files

0.1.14

4 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page