v0.3.2

• Change: certificates are now read with Claude Haiku 5.5 (claude-haiku-5-5) instead of
  Claude Haiku 4.5. It reads better on Anthropic's benchmarks and costs about a tenth as much per
  token: $0.10 / $0.50 per million input / output tokens for prompts up to 100,000 tokens (any normal
  certificate), against $1 / $5. Very long PDFs over that are billed at $0.50 / $2.50, half the old
  price. Its tokenizer counts the same text about 30% higher, so a read costs a little more than a
  tenth of what it did.
• Change: the read request now leaves room for the model's thinking. Haiku 5.5 thinks by default
  and its thinking counts toward the output cap, so the 4096-token cap tuned for Haiku 4.5 could
  have cut a dense certificate off mid-read — reported as "nothing could be read". The cap is now
  16,000 tokens, and the request asks for low effort, the level the docs recommend for simple
  extraction. Override with the new coa_vault_claude_effort filter (low | medium | high |
  xhigh | max, or '' to use the model's default; it also receives the model). xhigh and
  max exist only on some models — where the chosen model lacks the level, high is sent instead.
  Effort is never sent to models that reject it (Claude Haiku 4.5, and the deprecated Claude Sonnet
  4.5), so a site pinned to one of those keeps working.
• To stay on Haiku 4.5, set define('COA_VAULT_CLAUDE_MODEL', 'claude-haiku-4-5'); in
  wp-config.php or use the coa_vault_claude_model filter.

What's Changed

• 0.3.2: read certificates with Claude Haiku 5.5 by @uddhabh in #3

Full Changelog: v0.3.1...v0.3.2