You are an X07 coding agent building `x07crawl`.
GOAL
Build a crawler-shaped project that proves the RR workflow:
- keep crawl planning + parsing pure and deterministic (solve-pure)
- record HTTP responses once into rr cassettes
- replay offline (solve-rr) to validate extraction output deterministically
START PROJECT
https://x07lang.org/docs/getting-started/agent-quickstart
FUNCTIONAL REQUIREMENTS
- Planning module:
- normalize URLs deterministically
- enforce max-pages and stable dedup
- expose a pure function that turns newline-separated seeds into a deterministic plan
- Extraction module:
- extract at least one structured record from HTML
- emit canonical JSON bytes
- CLI wrapper:
- parse `--mode record|replay`, `--seeds`, `--out` (ext.cli specrows)
- emit canonical JSON summary bytes
RR REQUIREMENTS
- Provide rr policy under arch/rr/.
- Commit rr cassettes under tests/fixtures/replay/rr/.
- Solve-rr tests must:
- open cassette via std.rr.with_policy_v1
- call std.rr.next_v1 and validate response bytes
- run extraction against replayed HTML
TESTS
- unit tests for URL normalization, planning, and extraction
- 1 PBT test for an idempotence property (e.g. url normalize(normalize(x)) == normalize(x))
- 3 solve-rr tests that replay committed cassettes
DELIVERABLES
- README.md with record + test commands
- ci/telemetry.sh + ci/trust.sh + ci/review_diff.sh