Version Packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@tangle-network/agent-core@0.5.0

Minor Changes

• b449679: Carry prompt-cache tokens through readTokenUsage instead of discarding them.

  TokenUsageCounts gains cacheReadTokens and cacheWriteTokens, and the three
  prompt-side fields are now disjoint and additive: inputTokens is the freshly
  billed tail, and inputTokens + cacheReadTokens + cacheWriteTokens is the size
  of the prompt the model saw. Two helpers expose that directly — promptTokens()
  and cacheHitRate().

  Behavior change on an existing field. For OpenAI-compatible payloads,
  inputTokens previously returned the reported prompt_tokens, which already
  includes the cached share; it now excludes it, so a warm call's inputTokens
  drops to the part actually billed at the input rate. Measured on
  router.tangle.tools with glm-5.2, one append-only turn:
  prompt_tokens=8656, cached_tokens=8576 used to read as 8,656 input tokens and
  now reads as 80 input + 8,576 cache-read.

  The error flips direction, so callers must be updated, not just re-read. A
  caller that prices inputTokens at the full input rate previously OVERCHARGED a
  warm call (8,656 × full rate against a true 80 × full + 8,576 × cache rate); it
  now UNDERCHARGES, because the 8,576 cache-read tokens are no longer in the field
  it prices and it does not yet charge for them anywhere. Correct usage is to
  price all three terms:

  cost = inputTokens × inputRate
       + cacheReadTokens × cacheReadRate
       + cacheWriteTokens × cacheWriteRate
  

  and to use promptTokens(usage) — never inputTokens — wherever the number
  means "context size".

  genAiUsageAttributes emits gen_ai.usage.cache_read_input_tokens and
  gen_ai.usage.cache_creation_input_tokens beside gen_ai.usage.input_tokens,
  so total prompt volume stays recoverable on the attribute bag rather than being
  silently under-reported once input_tokens narrows to the billed tail. Both are
  omitted entirely for a producer that reported no cache information, because a
  synthesized 0 would read as "measured, no cache".

  The reader normalizes two provider conventions that reach callers through the
  same endpoint — OpenAI-style, where the cached count is inside the reported
  prompt total, and Anthropic-style, where it sits beside it — and does not
  double-count the router's normalized prompt_cache echo of a provider-native
  counter.

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-interface@0.43.1

Patch Changes

• 682814e: Offer codex none and pi ultracode in the reasoning-effort sets.

  Both were measured against the pinned CLI binaries and both were reachable all along: the codex API enumerates none first in its own error, and pi --thinking accepts max, which canonical ultracode maps to. The picker hid them, so turning thinking off on codex and reaching pi's top rung were not selectable.

@tangle-network/agent-provider-cli-bridge@0.3.5

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-provider-computesdk@0.2.37

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-provider-daytona@0.2.37

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-provider-e2b@0.2.37

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-provider-tangle@0.4.11

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

@tangle-network/agent-provider-testkit@0.5.3

Patch Changes

• Updated dependencies [682814e]
  • @tangle-network/agent-interface@0.43.1

[tangletools]

✅ Auto-approved GitHub Actions PR — 64e9dd70

This PR was opened by trusted GitHub Actions automation.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: github_actions_author · 2026-08-04T05:55:37Z}