v0.97.3 · 2026-07-07

The second adversarial pass — the review reviews the reviewers

Two more agents attacked the never-reviewed halves of the F5/OTel/rates
arc (client) and the nika dap server itself. The server's protocol
layer came back hardened (~30 hostile inputs, zero panics); the client
side pays its findings now:

• Preflight rates were dead on the wire — parseCheckReport never
  copied pricing: the exact class that hit requirements one review
  earlier, recurred one field later. Fixed — and a full-wire
  round-trip ratchet now types the fixture Required<CheckReport>,
  so any future field the parser forgets fails at compile time AND at
  test time. The class is structurally dead.
• The generated launch.json no longer hijacks F5. workflow: "${file}" reached the resolver before variable substitution, read as
  a literal path, missed the name, and silently replayed the newest run
  of ANY workflow. Resolution now runs in the substituted hook — and
  the newest-overall fallback is gone when the workflow's name is
  known: F5 says « no recorded run of name » instead of silently
  debugging a foreign journal (fork's never-silent-runs law, applied
  to the other direction).
• Quoted workflow names match their journals — workflow: "deploy #7" was truncated at the # by the line-scan extractor
  (the real parser handled it), so such workflows could never
  exact-match in the F5 direction.
• OTel export trusts the engine's own answer — the exported path
  is parsed from the engine's exported → … line instead of assumed
  from a suffix rule (a custom .jsonl traces glob made the assumed
  path point at the raw journal); a timed-out export no longer shows
  an empty error.
• Rates guard hardened (typeof, an omitted key renders nothing —
  never $undefined) · journal scan cap raised 100 → 500 stat-first.

---

📦 VS Marketplace · Open VSX · CHANGELOG · engine: docs.nika.sh/changelog/releases