CI PyPI version Python versions License: MIT
SpecShift detects changes in OpenAPI and Swagger contracts, classifies each one as breaking, warning, or info, and can optionally summarize them in plain language.
When an API changes from one version to the next, the real question isn't "what changed" but "will this break me". SpecShift is built to answer exactly that: it takes two specifications, evaluates every difference between them individually, and tells you which ones actually matter.
$ specshift diff examples/old_api.yaml examples/new_api.yaml
Bookstore API : 1.0.0 -> 2.0.0
13 breaking, 2 warning, 3 info changes found.
[BREAKING] DELETE /books/{bookId} :: HTTP method removed
[BREAKING] GET /books > parameter 'category' :: Parameter 'category' is now required
[BREAKING] GET /books > response 200 > field 'author' :: field removed from response
...
Result: 13 breaking change(s) make this update risky.
Every team that keeps evolving its API eventually hits the same problem: a field gets removed, a parameter becomes required, an enum value disappears, and nobody notices until a client breaks in production. Most existing diff tools just show you a raw JSON diff and leave it up to you to figure out what actually matters.
SpecShift doesn't do that. It evaluates every change based on its context:
- Removing a field from a response is breaking, because clients may depend on it being there.
- Removing the same field from a request body is usually just a warning, because clients that send it are simply ignored, not broken.
- Adding a new required field to a request is breaking, but adding a new field to a response is just informational.
These context-aware rules are the core of SpecShift, and they work completely free, with no API key required. The optional AI-powered natural-language summary is an additional layer on top, never a requirement.
- Comprehensive structural diff: deep comparison at the path, HTTP method, parameter, request body, response, and schema level.
- GraphQL support: diff GraphQL SDL schemas with
specshift diff-graphql, covering types, fields, arguments, enum values, interfaces, unions, and deprecations, no external GraphQL library required. - Protobuf/gRPC support: diff
.protofiles withspecshift diff-proto, using protobuf's own documented wire-compatibility rules (field numbers, wire type categories, zigzag encoding) rather than generic JSON-shape heuristics, so severity reflects what will actually break on the wire. - Context-aware classification: the same change is weighted differently depending on whether it occurs in a request or a response (or, for GraphQL, an input type versus an object type).
$refresolution andallOfmerging: correctly follows the reference and composition patterns common in real-world specifications.- Detects enum, format, nullable, security scheme, content-type,
additionalProperties,readOnly/writeOnly, validation constraint, default value, andoneOf/anyOfchanges. - Works entirely for free: no API key or paid service is required.
- Optional AI summary: can generate a natural-language summary using Groq, the Gemini free tier, or any OpenAI-compatible endpoint. If no key is set, it automatically falls back to a rule-based summary and never stops working.
- CI/CD integration: the
specshift checkcommand compares the current specification against a branch and fails the build if a breaking change is found. An official GitHub Action is also available for one-line setup with automatic PR comments (see below). - Live monitoring: the
specshift watchcommand periodically checks a remote API's specification and sends a Slack or Discord notification when it changes. - Local web dashboard:
specshift serveopens a browser-based tool for comparing specs interactively (drag in files or paste text), with zero extra dependencies since it runs entirely on Python's standard library and stays on localhost. - Auto-generated changelog:
specshift changelogwalks your spec file's history across git tags and produces a single Markdown document summarizing what changed (and whether it was breaking) between every release, plus an "Unreleased" section for the current working tree. - Spec linting:
specshift lintreviews a single OpenAPI document for spec-validity problems (undeclared path parameters, missing response descriptions, undefined security schemes, duplicate operationIds) and best-practice gaps (missing summaries, missing error responses, inconsistent path naming, unused schemas), independent of any diff. - Five output formats: a colored console table, a Markdown report (ideal for PR comments), JSON (for integrating with other tools), a standalone filterable HTML report, and SARIF 2.1.0 for GitHub Code Scanning.
pip install specshift
For colored console output (optional, works fine without it too):
pip install "specshift[pretty]"Installing from source:
git clone https://github.com/Lethe044/specshift.git cd specshift pip install -e .
Compare two specifications directly:
specshift diff old_openapi.yaml new_openapi.yaml
You can also compare specifications from URLs:
specshift diff https://api.example.com/v1/openapi.json https://api.example.com/v2/openapi.json
To use it in CI, create a configuration file in your repo:
specshift init
This produces a .specshift.yml file similar to:
spec_path: openapi.yaml base_ref: main fail_on: breaking
Then, in your CI pipeline:
specshift check
This command compares the current openapi.yaml file against its version
on the main branch and returns exit code 1 if a breaking change is found.
SpecShift can use free-tier AI services to generate a natural-language summary of the changes. This never requires any payment:
export GROQ_API_KEY="your-groq-api-key" specshift diff old.yaml new.yaml --ai
You can also use Google Gemini's free tier instead of Groq:
export GEMINI_API_KEY="your-gemini-api-key" specshift diff old.yaml new.yaml --ai --ai-provider gemini
If you want to use a more powerful (paid) model, you can connect any OpenAI-compatible endpoint:
export SPECSHIFT_API_KEY="your-api-key" export SPECSHIFT_OPENAI_BASE_URL="https://api.openai.com/v1" specshift diff old.yaml new.yaml --ai --ai-provider openai_compatible --ai-model gpt-4o-mini
If no key is configured, the --ai flag still works, it simply produces a
rule-based summary instead of waiting on a network call. AI support is an
optional enhancement, never a requirement.
Compares two specifications. <old> and <new> can be a file path, an
http(s) URL, or raw JSON/YAML text.
Useful options:
| Option | Description |
|---|---|
--format console|markdown|json|html|sarif |
Output format (default: console) |
--output <file> |
Writes the output to a file |
--ai |
Adds a natural-language summary |
--ai-provider groq|gemini|openai_compatible |
Chooses the AI provider |
--fail-on breaking|warning|none |
Determines at which level exit code 1 is returned |
--quiet |
Only prints the summary line |
Compares two GraphQL SDL schemas. <old> and <new> can be a file path,
an http(s) URL, or raw SDL text. Supports the same --format, --output,
--ai, and --fail-on options as diff.
specshift diff-graphql old_schema.graphql new_schema.graphql
Compares two Protobuf/gRPC .proto schemas using protobuf's own
wire-compatibility rules (field numbers, wire type categories) rather
than generic JSON-shape heuristics. Supports the same --format,
--output, --ai, and --fail-on options as diff.
specshift diff-proto old_service.proto new_service.proto
Starts a local, dependency-free web dashboard for comparing specs interactively. Opens a browser tab where you can drag in or paste two files (OpenAPI, GraphQL, or Protobuf, auto-detected), and view the same styled HTML report the CLI produces, with a short history of past comparisons in the current session.
specshift serve
# or: specshift serve --port 9000 --no-browserEverything runs in-process on your machine; nothing is uploaded anywhere,
and the server binds to 127.0.0.1 by default.
Designed for CI/CD. Compares the current specification file against a git
reference (branch, tag, or commit) defined in .specshift.yml.
specshift check --spec openapi.yaml --base-ref origin/main
Periodically checks a remote specification, compares it against the previous snapshot, and sends a notification if a difference is found.
specshift watch https://api.example.com/openapi.json \
--interval 600 \
--slack-webhook "$SLACK_WEBHOOK_URL"Walks the given file's history across git tags and generates a single Markdown changelog, diffing each consecutive pair of tagged versions. By default it also appends an "Unreleased" section comparing the latest tag to the current working tree, if they differ.
specshift changelog openapi.yaml --output CHANGELOG-API.md
Options: --tag-pattern (glob, default *, e.g. v* to only consider
tags starting with v), --no-working-tree to skip the "Unreleased"
section, and --output to write to a file instead of stdout.
Reviews a single OpenAPI/Swagger document for spec-validity problems and
best-practice gaps, independent of comparing it to any other version.
Supports the same --format, --output, --ai, --fail-on, and
--artifact-path options as diff.
specshift lint openapi.yaml
Bookstore API : 1.0.0
2 breaking, 3 warning, 1 info issues found.
[BREAKING] GET /books/{bookId} :: Path template parameter '{bookId}' is used in the path but not declared as a path parameter for this operation.
[BREAKING] GET /books/{bookId} > response 200 :: Response has no description. The OpenAPI spec requires every response to have one.
[WARNING] POST /books :: Operation has no operationId; many code generators rely on this to name functions.
...
Result: 2 issue(s) should be fixed for spec validity.
See When SpecShift calls something breaking below for what each severity means in lint results specifically.
Creates a sample .specshift.yml file.
SpecShift ships its own composite GitHub Action that runs the contract check and automatically posts (and keeps updated) a Markdown report as a pull request comment:
name: API Contract Check on: pull_request: paths: - "openapi.yaml" permissions: pull-requests: write jobs: contract-check: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: fetch-depth: 0 - uses: Lethe044/SpecShift@v1 with: spec-path: openapi.yaml fail-on: breaking # optional: ai-summary: 'true'
Available inputs: spec-path, base-ref (defaults to the PR base branch),
fail-on (breaking | warning | none), ai-summary (true/false),
ai-provider, comment-on-pr (true/false), upload-sarif
(true/false, see below), github-token, and specshift-version to
pin a specific release. Outputs: has-breaking-changes, breaking-count,
warning-count, info-count, report-path, and sarif-path.
To also surface results in GitHub's Code Scanning tab, enable
upload-sarif and grant the extra permission it needs:
permissions: pull-requests: write security-events: write jobs: contract-check: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: fetch-depth: 0 - uses: Lethe044/SpecShift@v1 with: spec-path: openapi.yaml upload-sarif: 'true'
If you'd rather not use the action (or want to combine it with other tooling), the CLI works just as well on its own:
name: API Contract Check on: pull_request: paths: - "openapi.yaml" jobs: contract-check: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: fetch-depth: 0 - uses: actions/setup-python@v5 with: python-version: "3.12" - run: pip install specshift - run: specshift check --base-ref origin/${{ github.base_ref }}
If you want to add AI-powered PR comments, you can generate a Markdown
report with specshift check --ai --format markdown --output report.md
and post it as a PR comment using an action like
peter-evans/create-or-update-comment.
spec_path: openapi.yaml base_ref: main fail_on: breaking # optional ai_provider: groq ai_model: llama-3.3-70b-versatile slack_webhook: https://hooks.slack.com/services/... discord_webhook: https://discord.com/api/webhooks/... ignore_paths: []
The table below summarizes which severity level applies in the most common scenarios:
| Change | In a request | In a response |
|---|---|---|
| Field removed | Warning | Breaking |
| New required field added | Breaking | Info |
| New optional field added | Info | Info |
| Field no longer required | Info | Breaking |
| Field became required | Breaking | Info |
| Data type changed | Breaking | Breaking |
| Enum value removed | Breaking | Breaking |
| Endpoint or method removed | Breaking | Breaking |
The same table applies to GraphQL: input types behave like requests,
type/interface fields behave like responses.
Protobuf/gRPC follows a different model, based on the wire format rather
than request/response direction: adding fields or rpc methods is always
safe, changing a field's number is always breaking, and type changes are
judged by wire type category (see the diff-proto section above).
specshift lint follows a different model too, since it reviews one
document rather than comparing two: breaking means the document
violates something the OpenAPI spec itself requires (undeclared path
parameters, a response with no description, an undefined security
scheme), which can cause strict tooling or code generators to reject the
document outright; warning means a common best-practice gap (missing
summaries, missing operationIds, undocumented error responses,
inconsistent naming); info means a minor, low-priority suggestion
(an apparently unused schema).
| SpecShift | Raw JSON/YAML diff | oasdiff / openapi-diff style tools | |
|---|---|---|---|
| Context-aware classification | Yes | No | Partially |
| GraphQL support | Yes | No | Rarely |
| Protobuf/gRPC support (wire-aware) | Yes | No | Rarely |
| Single-document linting (not just diffing) | Yes | No | Rarely |
| Natural-language summary | Yes (optional) | No | No |
| Standalone HTML report | Yes | No | Rarely |
| Local interactive dashboard | Yes | No | Rarely |
| Auto-generated changelog across git tags | Yes | No | Rarely |
| GitHub Code Scanning (SARIF) integration | Yes | No | Rarely |
| Free to use | Fully free | Free | Usually free |
| CI integration | Built-in (check + official Action) |
Manual | Varies |
| Live URL monitoring | Built-in (watch) |
No | Rarely |
This project is under active development. Some planned areas:
- Path parameter pattern (regex) and discriminator-level rule refinements
- Deeper GraphQL support: directive-aware deprecation reasons, custom scalar compatibility hints, and federation-aware schema composition
- Extending
specshift lintto GraphQL and Protobuf schemas - Optional shared/hosted history for the
servedashboard (currently in-memory and per-session by design) - Changelog generation from raw commit history (not just tags), for repositories that don't tag every release
Feel free to open an issue if you have a feature request.
See CONTRIBUTING.md for the contribution guide. Bug reports, feature requests, and pull requests are always welcome.
This project is licensed under the MIT License.