Skip to content

Changelog

Every release, both packages.

Release notes for the quelltest npm and PyPI packages, newest first. Each release has a permalink, and each package has an Atom feed.

npm

v0.3.1

Apache-2.0
npm install -D quelltest

PyPI

v2.0.0

MIT
pip install quelltest

Subscribe: all · npm · PyPI

npm Unreleased

unreleasedbreaking

Requires a newer Node.js. Merged on main, not yet published to npm (the latest published version is 0.3.1).

Breaking

  • Requires Node.js 22.18+ or 24.11+ (engines.node is ^22.18.0 || >=24.11.0). Node 20 and earlier 22.x are no longer supported: the parser dependency moved to @babel/parser 8, which declares that range. A test keeps engines.node equal to the parser's.

Changed

  • CI runs on Node 22 and 24 (Linux), plus Windows and macOS on Node 22.
  • Dev tooling: jest 30, mocha 12, vitest 5 (clears the open Dependabot alerts except a no-patch sprintf-js advisory in jest's dev-only config loader).

Upgrade notes

  • Check node --version before upgrading. On Node 20 or Node 22 older than 22.18, stay on 0.3.1.

PyPI Unreleased

unreleased

Changes on master since 2.0.0 that are not on PyPI yet (pyproject.toml already says 2.0.1). Entries are written from the merged pull requests; PR numbers are in parentheses.

Added

  • Measured per-line coverage is the ground truth for gaps (#165); gaps are ranked and suppressed by actionability (#167), quell find ranks gaps and offers measured coverage (#168), and process-control exits are suppressed (#170).
  • Synthesis mines real construction sites for custom types (#162), builds guard-aware mocks for parameters you do not supply (#163), reuses the project's own conftest.py fixtures (#157) and generates native async tests instead of asyncio.run (#158).
  • The verifier classifies gate failures and retries instead of discarding the test (#171).
  • MCP server for the Claude connector with OAuth PKCE and 8 tools (#123), and cloud sync with the quell sync command and --sync flag (#122).
  • Benchmarks: an ablation harness for the synthesis ladder (#172) and a recorded measurement of quelltest on itself (#169).

Changed

  • The Production Readiness Score scores the codebase, not quelltest's own output (#156).
  • The rule engine types every previously silent skip (#164).
  • quell find reports verified tests as WRITTEN, uses the real PRS and no longer crashes on Windows (#161).
  • Licence: the project is relicensed from MIT to Apache-2.0 (#178; applies from the next release; releases up to 2.0.0 stay MIT).

Removed

  • Breaking: quell scan and quell check are removed and now fail with an error; use quell find (#124).

Fixed

  • The verifier runs candidate tests inside the project's test tree (#160).
  • Boundary injection no longer corrupts mined constructions (#173); synthesis rung 3 is now actually called and composed with rung 2 (#174).
  • Six defects found by adversarial review (#176); CI: the fake self-check is fixed, lint is wider, the synthesis ladder is gated and the CLI contract is pinned (#175); a mis-attributed benchmark claim for rung 1 is retracted and marginal contribution is measured instead (#177).

npm 0.3.1

patch

Fixes from the 0.3.0 effectiveness report, and a research-backed answer to code that depends on API calls and services (see docs/dependencies.md).

Added

  • generate: property guards (typeof body.lat !== "number", !branch.geofences?.length) become tests with a composed argument.
  • generate: path-condition stub synthesis for injected dependencies (parameters, their methods, global fetch): the guards on a call's result define the conditions, a small solver picks the values, and the first guard to fire decides the expected outcome. Stubs are labelled as derived from guards, not from the real service. A stubbed globalThis.fetch is always restored.
  • Rules unguarded-network-call and deep-response-chain; inline literal secrets in jwt.sign/createHmac; the plain missing-await form.
  • verify: survivors on mocked code carry mockBoundary and an explanation in the handoff; --command runs your own integration command (Testcontainers, in-memory server) against every mutant; --timeout; mutants capped per kind (maxPerKind, default 8).
  • quell mocks --examples FILE: checks stubs annotated // quell-mock GET /path against recorded examples (HAR, JSON map, OpenAPI JSON). Exits 6 when a stub has drifted.
  • quell eval and the eval/ fixtures ship in the package; findings carry confidence and --min-confidence filters them.

Fixed

  • module.exports.name = fn, exports.name, bracket form and Object.assign(module.exports, ...) are recognised as exports.
  • Route request cases see through asyncHandler(...), catchAsync(...), wrap(...), middleware chains and curried wrappers.
  • gate reports a newly added .only and shows a deleted-and-rewritten test as a replacement.
  • A sanitiser applied through a variable (const p = escapeRegex(q)) no longer triggers regex-injection; a redirect checked with startsWith('/') is not an open redirect.
  • From the independent review of this release: generation for if (!body) ...; if (body.lat > 90) ... composes an object root instead of true; the security gate allows globalThis.fetch only as an assignment target or as const originalFetch = globalThis.fetch (never called or aliased); prototype-walking member names (__proto__, constructor, prototype) are rejected so the process cannot be polluted; guards nested in else-branches, loops or other conditions, tests of reassigned parameters and functions that call the network without a stubbable result are refused with a reason; a stubbed parameter that is also guarded keeps its guard cases; duplicate local result names are ambiguous; an imported or local fetch is not the global one; --command passes the test path in an environment variable (no shell interpolation) and --infra-exit marks environment-failure exit codes; missing-await needs a Model-like receiver or an ORM in the module and ignores Array.find; a redirect needs both startsWith('/') and startsWith('//') checks; only a try block (not catch/finally or detached callbacks) guards a call; gate pairing does renames first and ignores generic names; computed export keys are skipped; quell mocks escapes routes, avoids backtracking and survives malformed example files.
  • A verification mirror that cannot be deleted (a child process still holds it after a timeout) no longer fails the run.

Upgrade notes

  • Drop-in upgrade from 0.3.0. Requires Node.js 20+. Findings now carry confidence; pass --min-confidence to filter.

npm 0.3.0

minor

Repositioned as the free, offline gate for AI-written tests (see docs/agents.md).

Added

  • quell gate --diff REF: detects deleted/skipped tests, .only, removed and loosened assertions and changed expected values in a change; --json, --sarif, --strict; exit code 5. // quell-allow-weaken exempts a test.
  • quell mcp: MCP server on stdio with find, verify and gate tools.
  • verify: survivor handoff report (--handoff FILE), --min-score, --findings, --in-place, // quell-ignore-mutant, relative path support, the failing command in error messages.
  • Risk-aimed mutants: sanitiser unwrap, await removal, ??/|| swaps, ordered ahead of generic operator flips.
  • Rules: nosql-injection, path-traversal, open-redirect, missing-await, hard-coded-secret, eval-input, unparsed-json; one hop of data flow from req.body/query/params (aliasing, spread, $set, destructuring, JSON.parse, new URL).
  • // quell-ignore [rule] and // quell-ignore-file.
Show 2 more
  • generate: request cases for Express route handlers in TODO notes, actionable reasons (export / extract / unsupported expression), skips candidates already covered (--include-covered overrides).
  • Verifier benchmark (eval/gate), runner fixture matrix for Jest, Mocha, Vitest and node:test.

Changed

  • verify and generate --verify run in a temporary copy of the project by default; the real source is never written (falls back to in-place with a backup for very large projects).
  • Findings are ranked by per-rule risk; falsy-fallback fires only for numeric defaults; nullish ignores code that already has a fallback or guard; unguarded-call respects asyncHandler-style wrappers.
  • README rewritten around the verifier; honest comparison with Stryker, Flawd and LLM generators.

Fixed

  • Verification can no longer write through a symlinked source file; sources under build//dist/ are mirrored; mutants that do not parse are dropped instead of counted as kills; return await and plain coercions are no longer prioritised equivalent mutants; refuses to overwrite an existing test in place; warns when it falls back to in-place.
  • Scanner: per-sink sanitisers, whole-body vs single-field taint, reassignment and parameter shadowing, same-site redirects, route/i18n constants, process.env defaults; quell-ignore-mutant no longer suppresses scanner findings.
  • Gate: git renames, describe.only, template-literal and repeated test names, expect.assertions/t.mock/t.plan no longer counted, unparsable files are errors; the PR workflow's gate step now fails the job.
  • MCP: malformed messages, batches and stray responses are handled without crashing.

Upgrade notes

  • Nothing from 0.2.0 was removed. New exit codes: 4 (verify not verified, introduced in 0.2.0) and 5 (gate found weakened tests).

npm 0.2.0

minor

AST scanner, coverage-based score, mutation-based verification and real test generation for the detected runner.

Added

  • AST scanner (replaces the regex scanner) and a coverage-based score read from coverage-final.json or lcov.info.
  • Mutation-based verification: quell verify runs a test against source mutants and requires it to fail on them.
  • Real test generation for the detected runner (Jest, Vitest, Mocha or node:test).
  • --diff to restrict the report to changed lines, SARIF output, and an --llm hook for findings the rules cannot handle.

Changed

  • Licence changed from MIT to Apache-2.0.

Upgrade notes

  • quell verify exits with code 4 when a test is not verified.

npm 0.1.0

minor

First release: regex scanner and placeholder scaffolds.

Added

  • Regex-based scanner for risky edges in JavaScript and TypeScript.
  • Placeholder test scaffolds.

PyPI 2.0.0

major

Three-bucket output, a 5-gate verification pipeline and the Production Readiness Score.

Added

  • quell find: new primary command; auto-detects all spec sources; --fix writes WRITTEN tests, --auto skips confirmation, --format github emits CI annotations, --use-llm enables the Groq fallback.
  • Three-bucket output: WRITTEN (verified, written to disk), SCAFFOLDED (stubs in tests/scaffold/ with # quell: complete assertion), FLAGGED (one-line reason per item; nothing is dropped silently).
  • 5-gate pipeline: Gate 1 AST validity, Gate 2 originality, Gate 3 security, Gate 4 passes on correct code, Gate 5 fails on violated code. Only tests that pass all five become WRITTEN.
  • Production Readiness Score (PRS): a 0-100 score across WRITTEN tests and confidence values; tiers GREEN at 80 and above, YELLOW at 60 and above, RED below 60; posted as a PR comment by the GitHub Action.
  • Groq LLM provider (llama-3.3-70b-versatile), configured with quell auth set --provider groq --key .... The LLM is opt-in (use_llm = false by default).

Changed

  • quell init writes the v2.0.0 defaults (groq, prs_threshold=60, scaffold_dir) and adds the scaffold directory to .gitignore.
  • quell score is rewritten: it reads the PRS from quell-report.json; --badge writes an SVG and --json prints machine-readable output.

Upgrade notes

  • Requires Python 3.11+. The LLM stays off unless you pass --use-llm or set use_llm = true.

PyPI 1.0.0

major

Infrastructure-aware verified testing: QuellGraph and an ephemeral container engine.

Added

  • QuellGraph: persistent SQLite code-intelligence graph at .quellgraph/graph.db; incremental sha256-based AST builder; BFS infra-tag propagation across call chains (3-hop verified); quell graph build / show / why / stale / stats.
  • Container engine: starts throwaway Docker containers (postgres, redis, localstack, mongo, smtp, rabbitmq, elasticsearch) with hardcoded ephemeral credentials only; keep-alive lockfile; quell teardown; pytest fixture injection into conftest.py.
  • Confidence scorer: six factors (annotation coverage, constraint clarity, dependency clarity, graph coverage, docstring quality, mutation strength); tiers HIGH at 85 and above, MEDIUM at 70, LOW at 50, SKIP below 50; a write gate and a CI gate.
  • CLI flags: --with-containers, --min-confidence, --ci-confidence, --keep-containers, --graph-rebuild.
  • Environment detection for 8 runtime types (LOCAL_DOCKER, GITHUB_ACTIONS, GITLAB_CI, CIRCLECI, DEVCONTAINER, DOCKER_IN_DOCKER, KUBERNETES_POD, NO_DOCKER) with a per-environment strategy.

Upgrade notes

  • No breaking changes: all new flags are opt-in.

PyPI 0.9.9.2

patch

Rolls up the unpublished 0.9.9 and 0.9.9.1, and fixes the release pipeline.

Fixed

  • Engine accuracy: nested function scanning, syntax fixes, bare return violation, silent_fail stubs.
  • CI: ruff E501, mypy arg-type and no-untyped-def fixes.
  • Release workflow publishes with pypa/gh-action-pypi-publish.

PyPI 0.9.7

patch

Uploaded to PyPI on 2026-05-14, after 0.9.8 (the code dates from 2026-05-12).

Fixed

  • Functions whose required parameters are all of unknown type are skipped before any test is generated, so they no longer produce guaranteed-failing stubs.

PyPI 0.9.6

patch

Uploaded to PyPI on 2026-05-14, after 0.9.8 (the code dates from 2026-05-12).

Fixed

  • Written tests go to the project root tests/ directory (found via pyproject.toml, setup.py, setup.cfg or .git), not src/tests/; the CLI prints the full path.

PyPI 0.9.8

patch

Added

  • GitHub Action: a composite action that scans PRs, posts inline annotations and an idempotent PR comment; fail-on-gaps: true blocks merges.
  • GitHub App rewrite: webhook server with no per-repo YAML; fetches changed files via the Contents API.
  • --format github flag that outputs GitHub Actions annotation syntax.

Changed

  • source_line added to the Requirement model.

PyPI 0.9.6.1

patch

Added

  • CUSTOM guard rules: assert statements now generate concrete stubs.

Changed

  • Concrete class preference: non-abstract implementors are picked over base classes.
  • Assert violation injection comments out the assert line for Phase 2.

PyPI 0.9.5

patch

Added

  • Builtin exception guard: ValueError, TypeError and similar are handled directly.

Changed

  • Abstract stubs are skipped; ABC classes are never instantiated.

Fixed

  • PYTHONPATH=src/ is set automatically in the pytest subprocess, which fixes src-layout projects.

PyPI 0.9.4

patch

Added

  • Detects try/except/raise patterns and generates must_raise tests.
  • Detects standalone raise statements in guard branches.

PyPI 0.9.3

patch

Fixed

  • pytest is auto-detected when it is not in the sys.executable environment (conda, venv, pipx).

PyPI 0.9.2

patch

Includes the unpublished 0.9.1.

Added

  • Auto-loads the .env family (.env, .env.example, .env.template, .env.local, .env.secrets) into the pytest subprocess, so tests no longer fail on missing environment variables.

Changed

  • The report surfaces the real failure reason (from pytest output) for rejected tests.

PyPI 0.9.0

minor

Dual-engine architecture.

Added

  • Dual-engine architecture: the rule engine handles known patterns deterministically; the framework engine handles FastAPI and Flask route guards.

PyPI 0.8.1

patch

Fixed

  • The pytest subprocess uses sys.executable instead of a hardcoded python, so it runs in the same environment as Quell.
  • Subprocess output is decoded as UTF-8, which fixes silent errors on Windows (cp1252).

PyPI 0.8.0

minor

Async support and full violation coverage.

Added

  • Full violation coverage: every ConstraintKind gets a targeted injection.
  • Async function support: the test body is wrapped in asyncio.run().

Fixed

  • Windows encoding root cause in the verifier.
  • Duplicate kwargs in not_null stubs.

PyPI 0.7.0

minor

Fixed

  • Duplicate kwargs in not_null stubs.
  • silent_fail verification now correctly tests None-return paths.

PyPI 0.6.9

patch

Fixed

  • Pydantic @classmethod validators are handled correctly in stubs.
  • Enum kwarg name: stubs use the correct field name.
  • Duplicate Optional[X] in generated code is removed.

PyPI 0.6.8

patch

Fixed

  • Async functions are skipped, which was the root cause of verified=0 on FastAPI backends. Async support arrives in 0.8.0.

PyPI 0.6.5

patch

Added

  • quell-report.json is written after every run.

Fixed

  • Smarter stubs; guards on self.attr are skipped.

PyPI 0.6.2

patch

Fixed

  • Asyncio event loop conflict: the scan always runs in a dedicated thread (Jupyter and IPython).

PyPI 0.6.0

minor

CodeGuardReader.

Added

  • CodeGuardReader: scans if/raise, assert and try/except/raise patterns directly from source; no docstrings or annotations needed.
  • quell scan command, which works on any Python file.
  • FixSuggester: interactive fix recommendations.

PyPI 0.5.0

minor

Auth and PySpark.

Added

  • Auth system: quell auth login with PKCE OAuth; credentials are stored in ~/.quell/credentials.json. QUELL_API_KEY for CI.
  • PySpark reader: extracts requirements from StructType schemas (nullable=False, DoubleType, and so on) without importing pyspark; install with pip install quelltest[pyspark].
  • quell pr: posts scan results as a GitHub PR comment.
  • --no-llm flag: disables all LLM calls; rule engine only.

PyPI 0.4.2

patch

Changed

  • Targeted violation injection: each ConstraintKind gets a precise mutation.
  • Tests for optional-return functions are skipped.

PyPI 0.4.1

patch

Added

  • Signature inspection: real parameter names and types are used in generated stubs, and real tests are generated.
  • Diagnostic report: quell-report.json records per-requirement outcomes.

PyPI 0.4.0

minor

Spec-first architecture.

Changed

  • Unified Requirement model: every spec reader returns list[Requirement].

PyPI 0.3.0

minor

GitHub integration and PyPI packaging. Includes the unpublished 0.2.0.

Added

  • GitHub integration: PR comment poster and webhook listener (optional github extra, quell-github-app entry point).
  • VS Code extension scaffold.
  • CI score tracking: quell score --badge generates a badge (from 0.2.0).
  • Repair mode: auto-writes verified tests to disk (from 0.2.0).
  • MCP server: exposes quelltest as a tool for Claude Desktop (optional mcp extra, from 0.2.0).
  • SDK: from quell import Quell; q = Quell(); q.check('src/') (from 0.2.0).

PyPI 0.1.0

minor

Initial release.

Added

  • Python 3.11+, Typer CLI, Pydantic v2 and libcst.
  • Docstring reader: extracts Raises: and Returns: blocks.
  • Pydantic reader: extracts Field constraints and Literal types.
  • Two-phase verifier: Phase 1 must pass on correct code, Phase 2 must fail on violated code.
  • AST-safe writer: libcst injection, backup before write, restore on failure.

Dates are UTC publish dates from the registries. Sources: quelltest-npm and quelltest-lib.