v3.1.5: 3.1.5 release

Changes in v3.1.5

Fixes

• fix(product): QA round-2 — string-aware py-literals, definition-line gates, anchor v2, fullwidth paths

Full methodology for any metrics referenced: METRICS.md.

v3.1.4: 3.1.4 release

Changes in v3.1.4

Fixes

• fix(bench,guard): unicode-safe decoding on raw_search/rev-list/git sites — batch-2 sweep caught the same crash class on 2 more repos (slimtoolkit, argo-cd)
• fix(compactor): path extraction for @-scoped dirs, '+'-suffixed filenames and spaces — deterministic anchor to the source prompt

Full methodology for any metrics referenced: METRICS.md.

v3.1.3: 3.1.3 release

Changes in v3.1.3

Fixes

• fix(benchmark): refine git diff gates, diff slicer pass-through, json truncation repair and strictly english footer rule
• fix(product): QA-sweep wave — 506-repo dogfooding found and fixed 10 real defects

Other

• chore(registry): bump bench scenarios to v5 — kv_cache gate renamed to prefix_stability_ge_95 (methodology note included)

Full methodology for any metrics referenced: METRICS.md.

v3.1.2: 3.1.2 release

Changes in v3.1.2

Fixes

• fix(bench): NameErrors in measure_local_test_output (bench --real --here crashed on repos with tests/)

Docs

• docs(bench): registry metadata v4 — version bump + methodology note for 10-language expansion

Full methodology for any metrics referenced: METRICS.md.

v3.1.1: 3.1.1 release

Changes in v3.1.1

Features

• feat(bench): registry v4 — 10 repos on 10 supported languages (auto mode)

Docs

• docs(readme): v4 measured table — 15 repos, 10 languages, 25.7M tokens

Full methodology for any metrics referenced: METRICS.md.

v3.1.0: 3.1.0 release

Changes in v3.1.0

Features

• feat(bench+guard): v3 scenarios (5 new angles) + SECRET_FILE_READ guard rule
• feat(bench): v3 — 5 new scenario angles + SECRET_FILE_READ guard rule

Performance

• perf(bench): skeleton 1-hop deps -> signatures, failure-noise strip; measured gains

Docs

• docs(readme): v3 measured table — 3.06M tokens across 12 scenario angles

Full methodology for any metrics referenced: METRICS.md.

v3.0.1: 3.0.1 release

Changes in v3.0.1

Features

• feat(bench): real-repo benchmark — pre-registered scenarios + measured runs
• feat(bench): first measured real-repo run — 5 repos, 11/15 gated OK
• feat(bench): v2 scenarios — architecture, raw-search leak, git-diff churn, write-egress
• feat(bench): v2 measured run — 35 rows, 7 scenario angles on 5 real repos
• feat(cli): tokensafe bench --real --here (local-repo mode wiring)

Fixes

• fix(release): compute client pin upper bound from new major
• fix(bench): per-repo test venvs + final measured run (34 rows, 1 documented FAIL)

Docs

• docs(readme): measured real-repo benchmark section

Full methodology for any metrics referenced: METRICS.md.

v3.0.0: 3.0.0 release

Changes in v3.0.0

Features

• feat(adoption): /signup CTA path, community attribution, tokensafe status
• feat(v3): Tool Result Ledger, tool pruning, egress guard, history distillation
• feat(proxy): Anthropic /v1/messages ingest + one-command daemon wiring
• feat(cli): interactive proxy enable prompt in install and update flows

Fixes

• fix(cli): status joins the command elif chain (no Unknown-command fallthrough)

Refactoring

• refactor(audit): remove dead code, wire cache stats and session recalc

Full methodology for any metrics referenced: METRICS.md.

v2.9.0: 2.9.0 release

Changes in v2.9.0

Features

• feat(context): PostToolUse stdout compaction, scoped_search auto-regex, rules dedup, schema diet

Fixes

• fix(release): stage .codex-plugin/plugin.json in cloud sync commit

Full methodology for any metrics referenced: METRICS.md.

v2.8.4: 2.8.4 release

Changes in v2.8.4

Fixes

• fix(release): add 30s CDN replication buffer to wait_for_pypi_simple
• fix(security): default proxy bind host to 127.0.0.1 instead of 0.0.0.0

Docs

• docs(mcp,skills,rules): clarify tool parameters, relative path support, and usage guidance

Full methodology for any metrics referenced: METRICS.md.