The active tester for Build your own Web Application Firewall uses internal/waf and the stable course slug firewall. Keep this repository beside build-your-own-firewall, or set WAF_COURSE_DIR to the course checkout when running reference checks.
External client -> Learner WAF -> Local application origin
|
+----> Coursework ML detector
The reference checks use Go and the Python starter's uv launcher (Python 3.14). They start local listeners on port 8080, so stop any other learner process first.
make copy_course_file
make test
make buildEach new stage is checked against its corresponding Go and Python reference. Stages 1-3 use the course solutions. Stages 4-10 use the cumulative WAF fixtures in internal/waf/testdata/references, since the course publishes solutions only through stage 3. The preceding reference in each language must fail the next stage's new control. The final stage repeats the earlier contract cases in addition to checking the virtual patch.
test.sh and cmd/tester use the new WAF grader. It observes original request bytes, origin request counts, response status/body/headers, detector calls, timeout fallback, body-validation failures, spoofed diagnostics, route decisions, response leaks, and private-field export patches. No model-proposed tools are executed.
The main exercises use a tester-managed customer application and an embedded frozen classifier. Output compares direct application responses with firewall responses, diagnostics, and observed origin request counts. The final replay includes successful defenses, an unblocked detection probe, and a known false positive. Additional scripted detector responses check integration boundaries and failure behavior. Training source, held-out examples, and the reproducible model evaluation are maintained under model_development in the course repository; tests verify the embedded weights and scores match.
The integrated application exercises require the coordinated v4 tester release. Earlier releases can grade older stage orders and do not include the visible coursework replay. Publish the matching tester before applying the updated hosted course.
The Test workflow accepts a course_ref input for checking a coordinated course
commit or branch before merging. It now includes regressions for all supported
methods, exact fired-signal IDs, total detector deadlines (including trickled
headers and bodies), incomplete origin responses, and empty inspected bodies.
It also runs the frozen model checks and the real final-stage CLI in Go and
Python. The Publish workflow requires these checks against the course's main
branch before creating release artifacts.
Follow the course repository's RELEASING.md to coordinate publication and the
hosted tester selection. Passing local checks does not publish a new tester.