v0.2.0 — Dogfood-ready MVP milestone

v0.2.0 — Dogfood-ready MVP milestone

zhihu-grabber-toolkit has reached its first dogfood-ready milestone:
the core workflow from Zhihu discovery to verified capture and
large-corpus research has been exercised on real multi-page questions.

Highlights

Reliable Zhihu capture

• Search for relevant Zhihu questions.
• Capture complete multi-page answer sets.
• Batch multiple questions with failure isolation.
• Resume interrupted/completed captures without rewriting canonical data.
• Produce structured canonical JSON and human-readable rich Markdown.
• Verify artifacts before downstream handoff.

Content fidelity and safety

• Preserve canonical server content in answers.json.
• Render structured Markdown for human use.
• Treat scraped content as untrusted external data.
• Apply deterministic Markdown and URL safety boundaries.
• Keep credential handling local and outside generated artifacts.

Verified handoff and corpus workflows

• verify-output is the artifact-validity authority.
• Verified artifacts can be handed to corpus-anthology.
• Large corpora support inspection, representative sampling,
  digest workflows and mechanical archive workflows.

Real-world dogfood

The toolkit has been exercised against real multi-page Zhihu questions,
including production-like batches ranging from dozens to hundreds of
answers.

The largest dogfood question in this milestone contained:

• 538 captured answers
• 29 pages

A real verifier false positive discovered during dogfood
(VERIFIER_FALSE_POSITIVE_FENCED_CODE) was reproduced, fixed,
independently reviewed, merged, and the same 538-answer artifact now
verifies successfully.

Portability

Machine-facing artifact paths have explicit Windows cross-volume safety
semantics and do not fall back to leaking absolute machine paths.

Current limitations

This milestone intentionally does not claim every explored feature.

• Browser smoke is a best-effort diagnostic, not the artifact validity
  authority.
• Very large full-coverage digests can still be expensive; representative
  sampling is often the practical workflow.
• Search candidate metadata does not yet expose enough scale information
  for perfect pre-capture selection.
• End-to-end research orchestration still involves several explicit steps.
• Reported-vs-captured answer-count mismatch is intentionally visible as
  a non-blocking warning.

These are backlog / dogfood frictions rather than blockers for the
current MVP.

Version scope

v0.2.0 is the GitHub repository milestone version for
zhihu-grabber-toolkit.

The two independently usable modules keep their existing component
package versions; this release does not publish new npm package versions.