lasa 0.9.0

Lambda Attack Surface Analyzer — risk analysis for AWS Lambda packages. Zero pip dependencies.

lasa — risk analysis for AWS Lambda packages

Analyze a Lambda's deployed bytes for supply-chain risk: third-party dependencies, distinct PyPI maintainer accounts your code trusts, KEV- listed CVEs in installed versions, and the declared AWS API surface your handler actually calls.

Outputs a 0-100 score, a catalog.conf-shaped attack-surface listing, and a punchy human-readable report.

$ lasa risk --aws-lambda my-handler --region us-east-1

Lambda: my_handler.handler  (semcom score: 94/100)
==================================================
Third-party code:  1.98 MB  (100% of deployed package)
First-party code:  6.5 KB  (111 LOC)

Top contributors:
  urllib3                    724.6 KB  1.26.15 · transitive · 2 maintainers
  idna                       464.8 KB  3.4 · transitive · 1 maintainers
  requests                   335.5 KB  2.28.1 · transitive · 2 maintainers
  certifi                    280.7 KB  2022.12.7 · transitive · 2 maintainers
  charset-normalizer         221.1 KB  2.1.1 · transitive · 1 maintainers

Reachable maintainers: 8 distinct PyPI accounts
KEV-listed affected:   0 installed versions (0 CVEs)
Declared AWS surface:  (none detected)

Score breakdown:
  Third-party bytes:   -2.0 points  (1.98 MB)
  Maintainers:         -4.0 points  (8 accounts)
  KEV CVEs:            -0.0 points  (0 affected)
  AWS API surface:     -0.0 points  (0 distinct)
                        --------
  Total penalty:       -6.0 points
  Score:                94 / 100

What it analyzes

Input	What it does
path/to/lambda.zip	Read the zip; analyze contents in-memory
path/to/source/	Walk the directory; same analysis as for a zip
--aws-lambda <name>	Fetch the deployed zip via boto3, then analyze

What the score means

The score combines four penalty axes (each capped at the value in parens, total possible penalty = 100):

• Third-party bytes (cap 40) — every megabyte of vendored code is a megabyte you didn't audit.
• Maintainers (cap 20) — every distinct PyPI account that can push to a dependency is an account whose laptop, if compromised, ships to your prod.
• KEV CVEs (cap 20) — CVEs in CISA's Known Exploited Vulnerabilities catalog. We surface only KEV-listed (not all advisories) — the floor below which "this is fine" stops being defensible.
• AWS API surface (cap 20) — every distinct (service, method) pair your code can reach is an action your IAM role must permit. Smaller surface ⇒ smaller blast radius if creds are compromised.

Full methodology: docs/lasa/METHODOLOGY.md.

Subcommands

lasa analyze   <input>                emit raw analysis JSON
lasa catalog   <input> [--out FILE]   emit catalog.conf-shaped text
lasa risk      <input> [--json|--score-only]
lasa verify    <input> <catalog>      check source still implies catalog
lasa batch     [--region X]           leaderboard across an account
lasa ratchet   <new> --against <baseline>   CI gate: PR can't lower score
lasa version

CI ratchet (GitHub Action)

Drop this in a PR workflow to make sure no one merges a regression:

- uses: lanethompson/lasa@v0.7
  with:
    new: dist/lambda.zip
    baseline: baseline/lambda.zip   # checkout from main
    max-drop: 5                     # allow up to 5 pts of slip
    comment-on-pr: true             # posts the breakdown as a PR comment

The action exits non-zero (failing the PR check) when the new score falls below the baseline minus max-drop, and posts a markdown table comparing baseline vs new scores per axis. See .github/workflows/example-ratchet.yml for a complete example.

<input> is one of: a path to a .zip, a path to a source directory, or --aws-lambda <name> [--region <region>] [--profile <profile>].

Install

Requires Python 3.10+. Zero pip dependencies — stdlib only, including SigV4 signing for the AWS Lambda API.

pip install lasa
lasa version

Or from source:

git clone https://github.com/Apilify/lasa
cd lasa
make install            # creates .venv, installs the package
.venv/bin/lasa version

The supply-chain claim we make about other people's Lambdas applies to ourselves first: lasa has no third-party transitive dependencies to be hijacked by the next worm. pip install lasa adds nothing to your Python environment beyond lasa itself.

Why catalog.conf

The catalog is informational — it describes what the code does (detected from source), not what it's structurally constrained to do. The catalog file is not enforced by the runtime; nothing prevents the Lambda from doing things outside this list at runtime.

Catalog-disciplined deployments — where the catalog IS structurally enforced by the linker / runtime — are a separate product. lasa analyzes existing Python Lambdas; it does not transform them.

Caching

PyPI maintainer counts and OSV vulnerability lookups are cached to ~/.cache/semcom/ for 7 days. Override with --no-cache or --cache-ttl SECONDS.

Versioning

This is v0.1. The Python implementation is the oracle — it defines the score and catalog format. v1 is Forth + dlsym; v2 is verified IR caps. Field names and types in the JSON output are the public ABI both later versions must match.

lasa version prints the implementation version + the methodology version; the methodology version bumps whenever weights or rules change.

License

MIT.