Skip to content

Tokentoll

v0.8.3 Breaking

This release includes breaking changes for platform teams planning a safe upgrade.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

anthropic cost-optimization devtools github-action llm mlops
+3 more
openai python static-analysis

ReleasePort's take

Light signal
editorial:auto 4d

The diff engine now includes an identity‑shape pre‑pass to eliminate false REMOVED/ADDED reports for refactored call sites with unchanged shape.

Why it matters: Reduces noise in diff outputs, improving reliability of change detection when code is refactored but logically identical. (Fact severity: 40)

Summary

AI summary

Fixed false REMOVED/ADDED reports for refactored call sites with unchanged shape.

Changes in this release

Bugfix Medium

Diff engine adds identity-shape pre-pass to reduce phantom REMOVED/ADDED noise.

Diff engine adds identity-shape pre-pass to reduce phantom REMOVED/ADDED noise.

Source: llm_adapter@2026-05-30

Confidence: high

Bugfix Low

Line-proximity matcher still runs for remaining calls after identity-shape pre-pass.

Line-proximity matcher still runs for remaining calls after identity-shape pre-pass.

Source: granite4.1:30b@2026-05-30-audit

Confidence: low

Bugfix Low

Verified that five open PRs now have clean diff reports without phantom noise.

Verified that five open PRs now have clean diff reports without phantom noise.

Source: granite4.1:30b@2026-05-30-audit

Confidence: low

Bugfix Low

PR #1671 and PR #1756 correctly report genuine removed call sites in base.py.

PR #1671 and PR #1756 correctly report genuine removed call sites in base.py.

Source: granite4.1:30b@2026-05-30-audit

Confidence: low

Full changelog

Bug fix

Refactors that shifted call sites by more than 10 lines were reported as REMOVED plus ADDED pairs even when the call shape did not change. On gpt-researcher's open PRs this produced 5 to 10 noisy entries netting to zero monthly delta, which is the worst-case PR comment.

The diff engine now does an identity-shape pre-pass that pairs calls with matching sdk, call_type, model, model_is_literal, max_tokens, and raw_expression in order, regardless of line distance. The existing line-proximity matcher still runs for remaining calls so model swaps within plus or minus 10 lines still surface as MODIFIED.

Verified against gpt-researcher

Five open PRs that previously surfaced phantom +/- noise are now clean. Two of them (#1671, #1756) correctly report a small handful of genuinely removed call sites at the end of base.py.

Upgrade

pip install -U tokentoll

GitHub Action consumers on @v0.8.3 or @main pick up the new pin automatically.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track Tokentoll

Get notified when new releases ship.

Sign up free

About Tokentoll

All releases →

Related context

Beta — feedback welcome: [email protected]