Changelog

Every lgtmaybe release, newest first.

2.15.2

Released 2026-10-17 · Compare with the previous release

Bug Fixes

• ci: wait out PyPI index lag and winget token quota on release (#857) (140d197)

2.15.1

Released 2026-10-17 · Compare with the previous release

Bug Fixes

• ci: pin the tap checkout branch so it skips the REST API (#855) (8ff6b12)

2.15.0

Released 2026-10-17 · Compare with the previous release

Features

• gitea: resolve fixed findings' comments on Gitea 1.26+ (#826) (b1c55a0)
• provider: review with a ChatGPT subscription (#827) (15e830c)

Bug Fixes

• forgejo: read FORGEJO_ event names and stop own comments cancelling commands (#841) (7fea9b5)
• gitea: key resolved-comment dedupe on the comment's own author (#828) (2bb074b)
• land the rest of the merged stack and the second review pass (#820) (5967587)

Documentation

• fold stacked pull requests into the GitHub Action guide (#829) (ccac651)
• move long design history from the largest modules into docs (#854) (39b51ae), closes #840
• turn the ollama guide into a run-locally guide for ollama, llama.cpp and vLLM (#843) (4709365)

2.14.0

Released 2026-10-16 · Compare with the previous release

Features

• forgejo: keyless Bedrock, Azure and Vertex from a runner's OIDC token (#823) (75147cf)
• forgejo: review Forgejo pull requests from Forgejo Actions (#822) (a6d00b0)
• review stacked pull requests (#781) (52af2f4)

Bug Fixes

• azure_devops: count open finding threads in the PR context (#793) (3cfc1e3)
• config: reject a non-positive timeout at the config boundary (#802) (b74c3e2)
• disclose a temperature withheld under extended thinking (#801) (b74c3e2)
• engine: arbitrate opposed findings before the location merge (#786) (3cfc1e3)
• engine: budget the reflection audit diff to the input budget (#797) (3cfc1e3)
• engine: cap scan findings after diff scoping, not before (#788) (3cfc1e3)
• engine: keep scan findings on files triage or the cap dropped (#789) (3cfc1e3)
• engine: prefer the anchored finding when deduping a shared location (#785) (3cfc1e3)
• engine: rebuild the per-batch blocks for each split piece (#799) (b74c3e2)
• escape the describe fallback text before posting it (#791) (3cfc1e3)
• escape the finding path rendered in body-section bullets (#790) (3cfc1e3)
• fail fast on a response with no choices (#800) (b74c3e2)
• github: gate resolve-on-fix on a completed run (#783) (bd7aba4), closes #747
• github: keep validated resolution inside the incremental scope (#784) (3cfc1e3)
• gitlab: record a posted note in the run's notes cache (#805) (b74c3e2)
• keep a split piece's retry when a fallback model is configured (#798) (b74c3e2)
• read branch-mode head text from HEAD, not the working tree (#795) (3cfc1e3)
• redact five secret shapes the pattern battery missed (#792) (3cfc1e3)
• refuse a GitLab served under a sub-path on gitlab-ci (#806) (b74c3e2)
• require a self-note's formatting error to belong to the model's reply (#804) (b74c3e2)
• run slash commands only on created issue_comment events (#807) (b74c3e2)
• run the ast-grep subprocess on a scrubbed environment (#808) (b74c3e2)
• static-analysis: anchor zizmor findings on the Primary location (#787) (3cfc1e3)
• surface a hung local git as a ClickException (#796) (3cfc1e3)

Documentation

• openspec: propose Forgejo Actions support (#821) (5d0b766)

2.13.0

Released 2026-10-16 · Compare with the previous release

Features

• engine: add architecture review lens (#734) (111d3df)
• engine: arbitrate opposing findings and refresh ponytail lens for ponytail 5 (#731) (b2e5f8d)
• engine: write findings any engineer can act on (#736) (afa9f0d)
• static-analysis: static analysis now runs on every review by default, and every tool's output is a hint the reviewing model checks rather than a posted finding. Turn it off with static_analysis.enabled: false or --no-static-analysis, or post a tool directly with static_analysis.tool_mode (#779) (5671ce8)

Bug Fixes

• action: stop the review container before revoking its token (#745) (e15e373)
• comment: drop em dashes from the text lgtmaybe posts (#746) (2fe41a4)
• risk: count .jest. files as tests, thanks @kukielp (#730) (5e945a0)

Documentation

• add a glossary, a docs style guide and a readability gate (#737) (0dee3a0)
• add a What's New list and a Changelog page to the docs site (#732) (43dc53d)
• agents: cut AGENTS.md to the rules an agent acts on (#742) (f197899)
• drop lines about how the docs are built (#738) (e500665)
• rename CLAUDE.md to AGENTS.md (#739) (fbd41ba)
• rewrite the configuration and feature guides in plain language (#744) (0b8bf4d)
• rewrite the homepage, README, tutorial and explanation pages in plain language (#741) (ac92c47)
• rewrite the install, provider and host guides in plain language (#743) (2542169)

2.12.0

Released 2026-10-13 · Compare with the previous release

Features

• azure-devops: resolve finding threads once they are fixed (#725) (8834d94)
• azure-devops: review Azure DevOps Repos pull requests (#698) (daea409)
• calibrate reviews with a project profile (#687) (04a3a1c)
• cli: add --format sarif for code-scanning upload (#694) (f742b48)
• cli: first-run configuration: detecting config init, progressive help, no llama3 default (#696) (ac3af9f)
• cli: serve local reviews to coding agents over MCP (#695) (73a4f42)
• gitlab: incremental re-review and thumbs-down feedback (#697) (c70fa37)
• pre-commit framework hooks and a local --fail-on exit code (#693) (d6d44da)
• provider: say when a configured temperature never reached the model (#692) (5b315a7)
• provider: step reasoning effort only to levels the model accepts (#690) (f5f5507)

Bug Fixes

• cli: refuse non-github.com hosts for GitHub PR URLs (#707) (dea30fe)
• compress: split diff and file text on newline only (#713) (2ac53b8)
• engine: a rescued split piece doesn't re-send the oversized whole batch (#712) (7dd9f69)
• engine: filter before dedupe, keep distinct scan hits, gate unattributed code-health (#717) (a412155)
• forge: trust only lgtmaybe's own comments for ids and upserts (#721) (a412155)
• gitea: don't stop paging at a short page (#714) (a412155)
• gitea: only collaborators with write access can run slash commands (#706) (50079e3)
• gitea: refresh the overview on a push, which Gitea reports as synchronized (#701) (944540f)
• github: don't overwrite a completed summary when a run fails (#722) (a412155)
• github: read the last watermark and strip forged ones from model prose (#699) (0a2dc74)
• local: don't let a user's textconv filter rewrite the local diff (#729) (a78f514)
• local: pin git diff output format against user config (#708) (e4806e7)
• logging: actually redact registered secrets from logs (#716) (a412155)
• never report LGTM when nothing was reviewed (#720) (a412155)
• overview: render paths as single-line, escaped code spans (#719) (a412155)
• parse: one invalid finding no longer drops its siblings (#705) (c08178e)
• provider: drop prompt_cache_key when an endpoint rejects it (#711) (35f7b09)
• provider: fit a small context window instead of keeping the 100k default (#709) (b28da17)
• provider: handle Azure and Vertex ambient auth edge cases (#710) (6e79063)
• provider: honour reflect_model and triage_model (#702) (4ab4d03)
• redact: stop at line ends, bound the PEM scan, catch quoted keys (bee1c81)
• risk: security path signal matches words, not substrings; label scanner security hits (#715) (a412155)
• static-analysis: the PR can't reconfigure or blind its own scanners (#718) (a412155)
• suppress: a PR-authored pragma can't hide high/critical security findings (03d5869)
• trust only the finding ids the adapter appends, not ones in a suggestion (#724) (fee4cb9)

Performance Improvements

• github: post re-run findings in one review; skip deleted files' head text (#723) (a412155)

Dependencies

• bump the python-dependencies group across 1 directory with 13 updates (#686) (4a331ce)

Documentation

• full clarity and accuracy pass across the docs tree (#689) (100d087)
• gitlab,gitea: warn that MR/PR pipelines run the request's own CI config (#700) (98a5d76)
• refresh model ids for litellm 1.104 (#691) (44d9643)

2.11.1

Released 2026-10-12 · Compare with the previous release

Bug Fixes

• bind review summaries to authenticated native heads (#684) (0867057)
• update vulnerable release dependencies (#685) (879e4cc)

Dependencies

• bump the python-dependencies group with 7 updates plus urllib3 to 2.8.0 (#681) (66f1fff)

2.11.0

Released 2026-09-30 · Compare with the previous release

Features

• risk: risk-of-change verdict on every review (#678) (efc4e3f)

2.10.0

Released 2026-09-30 · Compare with the previous release

Features

• overview: skip the auto overview when the PR already diagrams the change (#675) (c32c70c)

Documentation

• release: winget token needs the workflow scope (#676) (a3b71bd)

2.9.1

Released 2026-09-30 · Compare with the previous release

Bug Fixes

• parse: ignore engine-owned fields a reviewing model fills in (#673) (9569687)

Dependencies

• bump the python-dependencies group with 6 updates (#672) (bf97f7d)

2.9.0

Released 2026-09-26 · Compare with the previous release

Features

• release: publish linux/arm64 image alongside amd64 (#671) (072b421)

Documentation

• refresh model recommendations from 2.8.5 benchmarks (#668) (f23bea4)
• simplify guidance and clarify privacy scope (#670) (1a8dc33)

2.8.5

Released 2026-09-20 · Compare with the previous release

Bug Fixes

• github: skip, not fail, when GitHub refuses to serve a PR's diff (#665) (381ef96)
• retain fetched grounding in large-diff reflection rechecks (#667) (a6ac60f)

2.8.4

Released 2026-09-20 · Compare with the previous release

Bug Fixes

• engine: retry a reasoning-bound step-down once with reasoning off (#661) (c53d5ac)

Dependencies

• bump the python-dependencies group with 8 updates plus anyio to 4.14.2 (#664) (5703a0e)

2.8.3

Released 2026-09-19 · Compare with the previous release

Bug Fixes

• engine: an active change is not evidence on its own, and unfit means could not fit (#657) (7e08c85)

2.8.2

Released 2026-09-19 · Compare with the previous release

Bug Fixes

• engine: a model's note to itself never posts, and the summary names the lenses that ran (#655) (d770c07)

2.8.1

Released 2026-09-17 · Compare with the previous release

Bug Fixes

• engine: remedy a parser-detected truncation like any other (#652) (48b7694)
• provider: walk the structured-output ladder on every trigger (#654) (eb86515)

Dependencies

• bump the python-dependencies group with 8 updates (#650) (9ab9242)
• raise litellm floor to 1.101.0 for corrected price map (#653) (916d4cf)

2.8.0

Released 2026-09-08 · Compare with the previous release

Features

• estimate and report review cost in the profile (#645) (9fb52b8)
• provider: use litellm's capability map for schema, window, and thinking (#647) (22f4820)

Dependencies

• bump the python-dependencies group with 9 updates (#649) (9ff8bb8)

Documentation

• add GLM-5.3 Flash benchmark results and the silent structured-output failure mode (#648) (e761d6d)

2.7.0

Released 2026-09-02 · Compare with the previous release

Features

• report High Impact Areas in a unified change overview comment (#644) (b1a13cb)

Bug Fixes

• action: keep config file optional (#637) (07a1eeb)
• action: reject unsupported GHES hosts (#625) (60a7857)
• aws: defer credentials to sdk (#622) (7ba588c)
• bound truncated finding salvage (#629) (a89bd5b)
• bound unpaginated Gitea reads (#638) (df479f8)
• budget context for reviewed files (#603) (8e53d1c)
• cache Docker dependency layer (#595) (ed3e0d1)
• cap files after triage ranking (#609) (7e043af)
• cli: disclose incomplete JSON reviews (#610) (7dee13b)
• compress: expand zero-length hunks correctly (#612) (437f956)
• distinguish finding identity sides (#617) (d9a05dd)
• enable manifest scan before prefetch (#626) (3496e13)
• engine: preserve scanner findings on model failure (#615) (97904bb)
• gitlab: preserve renamed file paths (#619) (beda46e)
• ignore root files in split heuristic (#611) (0981b51)
• neutralise findings before reflection (#633) (77100b2)
• ollama: preserve chat message roles (#631) (46f70fd)
• pair GitLab context line positions (#632) (e42f842)
• preserve Azure authentication errors (#623) (e7459a1)
• preserve duplicate forge findings (#621) (61de578)
• preserve header-like diff content (#639) (c6cfd5e)
• preserve JSON string content during repair (#642) (2fa7e52)
• prevent quadratic token redaction (#636) (cc6100f)
• protect scanner findings from feedback (#635) (27725d6)
• redact exact captured secret span (#643) (06afc05)
• reject truncated GitHub comparisons (#616) (6a511c8)
• reject unknown config get keys (#608) (7e5516f)
• remember rejected model parameters (#602) (3d0ae9f)
• remove needless review code (#597) (c1396b0)
• report unsplit parser truncation (#614) (96d7a8b)
• retain schema retry salvage (#613) (7fd5b26)
• retry GitHub writes after rate limits (#618) (9d64134)
• root local symbols at repository (#606) (583e5eb)
• salvage rejected initial review comments (#630) (7e813b5)
• security: isolate static analysis config (#640) (66aa950)
• security: keep clone token out of argv (#628) (d94a7f6)
• security: redact AWS secret credentials (#634) (38a354d)
• split diffs only on newline bytes (#641) (6cfe9dd)
• surface config file errors cleanly (#627) (7db160f)

Performance Improvements

• github: cache review thread walk (#607) (796dbbd)
• github: request larger review pages (#604) (c9c13b8)
• gitlab: cache merge request notes (#600) (6512ae7)
• local: remove redundant git subprocesses (#596) (21f9369)
• scan JSON delimiters once (#599) (3b1f256)
• skip supplied cache key derivation (#601) (deea912)

Dependencies

• bump the python-dependencies group with 7 updates (#546) (712ec3e)

Documentation

• clarify Azure credential precedence (#624) (3748f3f)
• correct automatic concurrency defaults (#605) (77489c7)

2.6.2

Released 2026-08-26 · Compare with the previous release

Bug Fixes

• diff: preserve paths containing header delimiters (#545) (9adfc7e)
• forge: preserve self-hosted URL schemes (#544) (02a18fd)
• gitlab: allow resolved findings to return (#542) (2caf5e7)
• gitlab: preserve CI server port (#543) (ea42610)
• gitlab: preserve rejected inline findings (#541) (f9d439c)

Documentation

• explain fallback model behavior (#539) (438b7ae)

2.6.1

Released 2026-08-26 · Compare with the previous release

Bug Fixes

• fall back to full review after merge (#538) (f4858ac)
• provider: read bedrock's Converse-level structured-output refusals (#536) (071282b)

2.6.0

Released 2026-08-25 · Compare with the previous release

Features

• escalate a truncated lens to the fallback model last, not first (#534) (8bdfebb)

2.5.1

Released 2026-08-24 · Compare with the previous release

Bug Fixes

• provider: send bedrock a schema subset its validator accepts (#533) (91a8cfc), closes #531

Dependencies

• bump the python-dependencies group with 9 updates (#530) (dfe686c)

2.5.0

Released 2026-08-20 · Compare with the previous release

Features

• hold the advisory lenses to the severity their prompt asks for (#529) (c9aab3c)

Bug Fixes

• count tokens in the model's own tokenizer and reserve prompt overhead (#528) (d41dab9), closes #509
• escape the file path when fetching file contents (#522) (c244bf7), closes #508
• honour resolve_fixed on GitLab (#525) (6f5efe5), closes #502
• read a fenced reply in /ask and follow-up validation (#523) (ae4e650), closes #510 #511
• read a self-hosted server's tool-call rejection wording (#526) (3e3a200), closes #499

Performance Improvements

• overlap the per-file boundary scans and guard the flood counter (#524) (70b87e5), closes #505 #506

Documentation

• correct the gitea token scopes and drop a flag that does not exist (#527) (fd36563), closes #500 #501
• rescore the model-choice tables and correct three recommendations (#512) (d9953cc)

2.4.0

Released 2026-08-19 · Compare with the previous release

Features

• review merge requests on GitLab (#482) (adb87b9)
• review pull requests on Gitea (#481) (b46d34a)

Bug Fixes

• bound runaway model output that floods a review (#484) (90f1511)

Documentation

• add benchmark model selection guide (#472) (9a47534)
• cover GitLab and Gitea across the documentation (#483) (baf0660)
• cut table-restating detail from the review-model notes (#478) (adfb0a9)
• define the benchmark suites before the tables that use them (#479) (4d30161)
• make the review-model guide read more naturally (#475) (0bf32a7)
• plainer opening for the choose-a-review-model page (#476) (364f1a6)
• remove marketing phrasing from the review-model guide (#477) (2b45195)

2.3.0

Released 2026-08-18 · Compare with the previous release

Features

• provider: keep structured output on routes that reject response_format (#470) (28830c2)

2.2.0

Released 2026-08-17 · Compare with the previous release

Features

• cli: give the profile a machine-readable form, and stop it corrupting stdout (#460) (ff86872)
• engine: re-ask a lens without the schema when its reply won't parse (#463) (2a241cd), closes #454
• evals: run each fixture N times so flaky and reproducible failures separate (#461) (b8d0568), closes #458
• make structured-output compliance failures diagnosable and survivable (#450) (d1c4f4d)

Bug Fixes

• engine: don't re-ask a lens whose schema the adapter already stripped (#465) (60d7944)
• provider: don't step down an effort the route would discard (#459) (464b5b6)
• provider: give a reasoning-bound truncation a lever when no effort was set (#452) (abe7dac)

Dependencies

• bump the python-dependencies group with 8 updates (#449) (047e8b4)

2.1.4

Released 2026-08-14 · Compare with the previous release

Bug Fixes

• release: make the homebrew gate wait for the PyPI publish (#447) (1806faf)

2.1.3

Released 2026-08-14 · Compare with the previous release

Bug Fixes

• github: post re-run inline findings when the run was incomplete (#445) (d145b70)

2.1.2

Released 2026-08-14 · Compare with the previous release

Bug Fixes

• trace review findings through profile (#441) (8f21f4d)

2.1.1

Released 2026-08-14 · Compare with the previous release

Bug Fixes

• github: preserve findings rejected by GitHub (#438) (801d195)

2.1.0

Released 2026-08-13 · Compare with the previous release

Features

• engine: retry a reasoning-bound truncation at a lower effort (#434) (83e3736)
• engine: tell every lens what it was not shown (#433) (495bf40)
• profiling: report the reasoning share of every call, not just failed ones (#436) (35fa139)

2.0.1

Released 2026-08-13 · Compare with the previous release

Bug Fixes

• align GitHub Action with v2 (#430) (386b9d4)
• prompt: close the seam between humility and the gap carve-out (#428) (0d09eb4)

2.0.0

Released 2026-08-13 · Compare with the previous release

⚠ BREAKING CHANGES

• remove answer_replies from configuration and Action inputs.
• remove legacy prompt cache path (#399)

Features

• clarify user-facing review responses (#405) (59a8b64)
• cli: report the running version with --version (#411) (ac15398)
• default to medium reasoning, and stop capping local models (#402) (3028133)
• ollama: cap local output so a review can't decode forever (#413) (c0c7428)
• validate findings on follow-up reviews (#412) (419460a)

Bug Fixes

• cli: scale the local per-call timeout by the fan-out width (#404) (3385dcd)
• deterministic ceiling tests, and delete the dead single-stream branch (#407) (334396e)
• engine: rescue a split piece that failed on the provider (#400) (160fcb0)
• give a reasoning-bound truncation both levers, and the dogfood cap the headroom (#418) (ef74cfe)
• make follow-up thread resolution fail safe (#419) (ccd0e62)
• providers: read a spent ceiling as a truncation (#417) (64513ce)
• remove automatic review thread replies (#408) (d435124)
• restore automatic PR diagrams (#406) (183b8ec)
• slash: stop /ask asking the model for a findings object (#415) (800df1a)

Code Refactoring

• remove legacy prompt cache path (#399) (86f276e)

1.14.1

Released 2026-08-13 · Compare with the previous release

Dependencies

• bump the python-dependencies group with 8 updates (#375) (50056c6)

1.14.0

Released 2026-08-13 · Compare with the previous release

Features

• engine: add a spec lens that checks the PR against its committed spec (#376) (d55e35e)

Bug Fixes

• stop one flaky provider call voiding a whole review round (#378) (6c59f86)

Dependencies

• bump the python-dependencies group and unblock pip-audit on aiohttp (#379) (d296f25)

1.13.1

Released 2026-08-04 · Compare with the previous release

Bug Fixes

• github: render finding confidence as a percentage (#373) (6b56e0f)

1.13.0

Released 2026-08-03 · Compare with the previous release

Features

• cli: post partial results when the process is signalled (#367) (1819896)
• engine: skip oversized files and more generated artefacts (#366) (bb03da6)

Bug Fixes

• providers: deliver the reasoning budget on OpenRouter, and never drop a param in silence (#369) (afcc0b8), closes #348

Performance Improvements

• ci: start no job for a comment with no slash command (#365) (e72903d)
• engine: stop splitting a truncation the thinking budget caused (#368) (85c5258)

1.12.2

Released 2026-08-03 · Compare with the previous release

Dependencies

• bump litellm in the python-dependencies group (#363) (d225fcc)

1.12.1

Released 2026-08-03 · Compare with the previous release

Bug Fixes

• deps: sync uv.lock with the 1.12.0 version bump (#340) (b310637)
• engine: exclude diff file headers from the triage line count (#341) (43965bb), closes #326

Dependencies

• bump the python-dependencies group across 1 directory with 7 updates (#353) (5c7f220)

1.12.0

Released 2026-08-02 · Compare with the previous release

Features

• config: bound reasoning with reasoning_effort (#323) (0e0c83c)

1.11.1

Released 2026-08-02 · Compare with the previous release

Bug Fixes

• engine: tell the intent lens which files it was not shown (#321) (cfa8742)

1.11.0

Released 2026-08-02 · Compare with the previous release

Features

• engine: report reasoning tokens on every provider call (#319) (24c4645)

1.10.0

Released 2026-08-02 · Compare with the previous release

Features

• cli: flatten the diagram's collapsible blocks for a terminal (#317) (961738b)
• config: scope review instructions and context to a directory (#314) (d3f79f1)
• diagram: add a sequence view beside the change flowchart (#309) (3fe0f87)
• engine: let a lens defer once for bounded context (#316) (008eca7)
• github: show the lens and confidence on posted findings (#313) (f72d310)

Bug Fixes

• config: stop the dogfood cap starving a reasoning model (#312) (a9e3593)
• engine: split and retry when a call blows the output ceiling (#315) (dd4b3c9)

1.9.1

Released 2026-07-28 · Compare with the previous release

Bug Fixes

• parse: keep the findings a truncated response finished emitting (#306) (b8d9193)
• provider: report a truncated response as truncated, and stop retrying it (#305) (cadca72)
• provider: survive a rejected response_format on Bedrock and keep --json stdout parseable (#303) (a11860c)

1.9.0

Released 2026-07-27 · Compare with the previous release

Features

• cli: report what a local review spent (#299) (78af30b)
• engine: cap a review's spend with max_review_tokens (#297) (78987a6)

1.8.1

Released 2026-07-27 · Compare with the previous release

Bug Fixes

• bugs found driving the CLI end-to-end across 10 scenarios (#291) (1a9f0f0)
• compress: merge hunks whose context pads overlap (#290) (6fcb279)
• context: only pad to a definition that still contains the hunk (#294) (7cb9129)
• identity: stop asking for branded identity on an event that cannot mint it (#296) (f4c3bdd)
• prompt: stop asking lenses to flag our own redaction marker (#293) (054a4ce)

Performance Improvements

• context: size the enclosing-definition reach against the fixed pad (#295) (eb586f1)

1.8.0

Released 2026-07-27 · Compare with the previous release

Features

• engine: name the lgtmaybe version alongside the model in posts (#284) (a295cd8)
• static-analysis: add mypy to the fusion runner (#287) (b8f8739)
• static-analysis: deterministic scan tools that replace LLM work (#288) (4eee72c)
• static-analysis: semgrep rule pack, ast-grep, osv-scanner, and prompt narrowing (#289) (1cddf6b)

Bug Fixes

• prompt: bind suggestion names to the file being changed (#286) (7bb2f02)

1.7.0

Released 2026-07-26 · Compare with the previous release

Features

• diagnostics: log the effective timeout's source; stop retrying a refusal (#271) (ee1e42c)
• provider: cap generated tokens and fail fast when credit runs out (#281) (ea84ba4)
• summary: stop claiming LGTM while business is still outstanding (#273) (1b50d93)

Bug Fixes

• dedupe: key re-run dedupe on a prose-free finding identity (#275) (7ee4eed)
• engine: report a split piece that failed instead of claiming the batch was reviewed (#279) (8885bbd)
• posting: make an incomplete review visible on the PR (#278) (85f2e7f)
• providers: fail a blown wall clock once, and pin the documented timeouts (#277) (1af7efc)
• workflows: stop lgtmaybe cancelling its own reviews (#276) (1ece55e)

Dependencies

• bump the python-dependencies group and migrate to cdk-nag v3 (#280) (bfd429b)

1.6.0

Released 2026-07-26 · Compare with the previous release

Features

• cache: widen prompt caching to vertex, zai, and openrouter routes (#266) (461c34a)
• preset: four distinct lenses on every provider (#268) (2d81f5e)
• timeouts: raise every wall-clock budget so slow calls finish (#264) (90591f4)

Performance Improvements

• overlap the three serial I/O stages, isolate resolve failures (#270) (f03c4bf)

Documentation

• preset: fix the mangled --preset help and sweep the stale call counts (#269) (44a4776)

1.5.5

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• homebrew: trust the tap in the smoke gate instead of disabling the check (#262) (a22bb02)

1.5.4

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• providers: make the wall-clock timeout clock-authoritative (#260) (979e0e5)

1.5.3

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• load broker from flattened Lambda bundle (#249) (a886683)

1.5.2

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• avoid deprecated GitHub App input (#246) (f97bb01)
• skip Winget update before initial publication (#248) (72c59fa)

1.5.1

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• pass GitHub identity timeout by keyword (#243) (6d9beea)

1.5.0

Released 2026-07-25 · Compare with the previous release

Features

• add public lgtmaybe bot identity (#239) (8ff8720)
• gate defect findings on failure scenarios (#238) (ea23a60)

Bug Fixes

• harden auxiliary model outputs (#241) (421c188)
• prevent bot comments cancelling reviews (#240) (0eeeee3)
• upgrade GitHub App token action to v3 (#237) (8b38186)

Documentation

• expand WinGet installation guidance (#235) (777163e)

1.4.1

Released 2026-07-25 · Compare with the previous release

Bug Fixes

• generate readable Mermaid flowcharts (#231) (e85b491)
• pass repository to release commands (#232) (1e4c633)

1.4.0

Released 2026-07-25 · Compare with the previous release

Features

• add Windows executable and winget distribution (cac1339)

Bug Fixes

• harden Windows release workflows (ee08c9f)
• support Windows compatibility (ab7185e)

Documentation

• simplify Marketplace Action setup (2b9c274)

1.3.0

Released 2026-07-25 · Compare with the previous release

Features

• diagram: green-border changed elements; recommend App identity; marketplace assets (#220) (17700fc)
• diagram: light-green relationship lines and labels in C4 diagrams (#224) (a1c40f5)
• full-screen link and density-aware layout for change diagrams (4667dc2)
• generous timeout defaults for slow providers and auto diagrams by default (18707a5)
• raise the direct-cloud timeout default to 300s (e656b32)

Bug Fixes

• diagram: loosen C4 layout and brighten rel styling for dark mode (#221) (00116a6)

Documentation

• link the Marketplace listing and feature the branded screenshots (#223) (ddb39ec)

1.2.0

Released 2026-07-25 · Compare with the previous release

Features

• answer author replies in review finding threads (#217) (3ecf858)
• gate merges on findings via a GitHub Check Run (#214) (4cc2604)
• parallelise correctness review (#210) (f63e0ec)
• suppress findings downvoted with thumbs-down reactions (#213) (fbbebce)
• write reviews in a configurable language (#215) (23abe0a)

Bug Fixes

• diagram: keep C4 relationship lines readable in GitHub dark mode (#216) (317ce54)
• harden review pipeline against watermark, expansion, and redaction bugs (#212) (f03e076)

Documentation

• move docs site to lgtmaybe.coles.codes (#218) (65960fc)

1.1.0

Released 2026-07-24 · Compare with the previous release

Features

• diff: skip generated llms.txt/llms-full.txt files in review (#202) (b06086c)
• enable diagrams in starter workflows (#204) (157c47d)

Bug Fixes

• align action with v1 release (#209) (46e94b5)
• bound default review runtime (#208) (d21e4e8)

Documentation

• add Google Search Console verification file (#201) (d20b6d3)
• clarify Marketplace provider setup (#206) (089e74f)
• improve homepage feature showcase (#207) (ab9984f)
• show change diagram on homepage (#205) (64f58f7)

1.0.0

Released 2026-07-24 · Compare with the previous release

Features

• action: post reviews as a GitHub App via app_id/app_private_key inputs (#197) (88ec1e4)

Documentation

• optimise the docs site for SEO and LLM crawlers (#200) (d46078f)

0.13.1

Released 2026-07-24 · Compare with the previous release

Documentation

• accuracy pass, Mermaid diagrams, and a light-stroke header logo (#194) (394ea3a)

0.13.0

Released 2026-07-24 · Compare with the previous release

Features

• C4-style change diagram (/diagram, auto_diagram, lgtmaybe diagram) (#191) (4d99649)

0.12.2

Released 2026-07-22 · Compare with the previous release

Bug Fixes

• homebrew: prefer wheels so litellm >=1.92 sdist never builds in the sandbox (#189) (df6e7e1)

Dependencies

• bump the python-dependencies group with 6 updates (#187) (5615a79)

0.12.1

Released 2026-07-17 · Compare with the previous release

Performance Improvements

• cut redundant work in the pipeline and trim prompt token waste (#185) (c2ebd97)

0.12.0

Released 2026-07-17 · Compare with the previous release

Features

• add openspec living specs anchored to code with ast-grep (#178) (b61813f)

Bug Fixes

• deps: cap litellm below 1.92 on Python 3.14 (#181) (0add942)

Dependencies

• bump the python-dependencies group with 5 updates (#180) (33d8ba0)

Documentation

• make the docs read more human and easier (#183) (137fb4a)
• product spec for simplifying end-user CLI configuration (#182) (25f6979)

0.11.0

Released 2026-07-06 · Compare with the previous release

Features

• cli: add help command with usage examples (#177) (beac277)

Bug Fixes

• evals: disable the review deadline in eval runs + one-command preset A/B (#175) (934b309)

0.10.0

Released 2026-07-06 · Compare with the previous release

⚠ BREAKING CHANGES

• engine: reviews now run the fast preset by default — all nine lenses covered in four grouped model calls (~half the calls and wall time), trading some recall on the softer lenses (performance, complexity, ponytail, deprecation, tests, documentation). Set preset: full (or --preset full / the Action's preset input) to restore the previous one-call-per-lens behaviour.

Features

• engine: first-class structured describe + opt-in auto-describe (#171) (35f1b08)
• engine: function-boundary context, per-tool lint floors, eval A/B coverage (#173) (684acf1)
• engine: review-effort/risk labels + declarative finding rules (#172) (1414322)
• engine: static-analysis fusion — deterministic linters ground the review (#169) (48d6ecf)
• engine: two-stage triage routing behind a security floor (#170) (00966fe)
• github: commit-scoped incremental review on synchronize pushes (#168) (f16f915)
• prompt caching + audit-driven review improvements (#166) (95acc2b)

Performance Improvements

• engine: cut review wall time — global fan-out pool, cached diff prefix, fast preset, deadlines (#174) (853917a)

Dependencies

• bump the python-dependencies group across 1 directory with 3 updates (#164) (bb83a4f)

0.9.2

Released 2026-07-01 · Compare with the previous release

Bug Fixes

• provider: fail fast on expired cloud credentials (#162) (c56fa7d)

0.9.1

Released 2026-07-01 · Compare with the previous release

Documentation

• streamline install + local-model guides (#160) (6425ad3)

0.9.0

Released 2026-06-30 · Compare with the previous release

Features

• homebrew: install lgtmaybe + deps from PyPI wheels (preserve_rpath) (#157) (fe10d12)

0.8.2

Released 2026-06-30 · Compare with the previous release

Documentation

• homebrew: document the required brew trust step for the tap (#155) (5099cd3)

0.8.1

Released 2026-06-30 · Compare with the previous release

Bug Fixes

• homebrew: make the tap actually publish (path, tap registration, first-publish commit) (#154) (4e71803)
• homebrew: make the tap publish reliably on release (#152) (b46619b)

0.8.0

Released 2026-06-30 · Compare with the previous release

Features

• add Homebrew install support (#143) (e4b9e1b)
• resolve cross-file symbols with ast-grep during reflection (#144) (0ad9cdf)

Bug Fixes

• surface clear errors for malformed payloads and tiny reflect budgets (#148) (dad6e7a)

Documentation

• align end-user docs with the codebase (#146) (fd7bf94)
• optimise documentation for SEO and LLM crawlers (#147) (b881877)

0.7.2

Released 2026-06-30 · Compare with the previous release

Documentation

• add how-to contents, default examples to Sonnet, and support Python 3.11 (#141) (a79bb12)

0.7.1

Released 2026-06-30 · Compare with the previous release

Dependencies

• bump the python-dependencies group with 5 updates (#139) (8b03a49)

0.7.0

Released 2026-06-24 · Compare with the previous release

Features

• cut false positives from real-project feedback (prompts, grounded reflection, retrieval, benchmark) (#136) (e98eddf)

0.6.2

Released 2026-06-23 · Compare with the previous release

Documentation

• add per-provider how-to guides and reframe the OpenAI-compatible guide (#133) (14c5470)

0.6.1

Released 2026-06-23 · Compare with the previous release

Bug Fixes

• diffparse: ignore "No newline at end of file" markers when anchoring (#128) (94a03e3)
• github: defang triple-backticks in finding title and body (#132) (0496939)
• models: coerce model severity case so mixed-case findings aren't dropped (#131) (82c9078)
• redact: bound scheme length to stop quadratic connection-string match (#130) (4701287)

Documentation

• correct stale defaults, dedupe key, and release/redaction descriptions (#127) (266137f)

0.6.0

Released 2026-06-23 · Compare with the previous release

Features

• providers: add first-class zai (GLM / Zhipu AI) provider (#125) (f103e4f)

0.5.0

Released 2026-06-23 · Compare with the previous release

Features

• engine: collapse same-location findings regardless of title wording (#123) (bfd2031)

0.4.1

Released 2026-06-22 · Compare with the previous release

Bug Fixes

• evals: resolve credentials so the harness runs keyless local endpoints (#121) (cfd840a)
• keep parsing past a non-finding JSON object in model output (#119) (8296fde)
• recover empty structured-output responses from local gateways (#118) (5b70a04)

Documentation

• correct the "review incomplete" error string in the ollama guide (#120) (1ae700b)

0.4.0

Released 2026-06-22 · Compare with the previous release

Features

• cut weak-model false positives with confidence gates (#115) (1911edc)

Bug Fixes

• e2e: use llama.cpp -hf quant tag instead of filename (#117) (c5b279f)

Dependencies

• bump the python-dependencies group with 5 updates (#114) (a81be06)

0.3.3

Released 2026-06-19 · Compare with the previous release

Documentation

• correct stale claims across the docs tree (#112) (cbc0bba)
• document codebase humility and eval false-positive scoring (#109) (ad42254)

0.3.2

Released 2026-06-19 · Compare with the previous release

Bug Fixes

• robustly parse findings JSON from gateways without JSON mode (#105) (f84fd77)

0.3.1

Released 2026-06-17 · Compare with the previous release

Bug Fixes

• reframe demoted-findings body section for customers (#102) (f296911)

0.3.0

Released 2026-06-17 · Compare with the previous release

Features

• deterministically anchor findings to the lines they describe (#100) (0cee324)

0.2.1

Released 2026-06-16 · Compare with the previous release

Bug Fixes

• anchor inline comments by line+side, not deprecated position (#97) (621a030)

0.2.0

Released 2026-06-16 · Compare with the previous release

Features

• add curated opt-in lens packs and inspired-by attribution (#91) (b75e82d)
• recursive (RLM) hunk-walk by default, with opt-out and an A/B benchmark (#95) (d855190)

0.1.11

Released 2026-06-16 · Compare with the previous release

Bug Fixes

• fail fast on permanent provider errors and bound the lens fan-out (#92) (0da9be0)

0.1.10

Released 2026-06-15 · Compare with the previous release

Features

• engine: log per-lens progress so a running Action shows it's not stuck (#87) (8f821e2)

0.1.9

Released 2026-06-15 · Compare with the previous release

Bug Fixes

• make the model emit code in suggestion, prose in body (#83) (f2c5697)

0.1.8

Released 2026-06-15 · Compare with the previous release

Bug Fixes

• deps: bump aiohttp to 3.14.1 for CVE fixes (#86) (542525e)
• retry without temperature when a model rejects the value (#84) (a79e91c)

0.1.7

Released 2026-06-15 · Compare with the previous release

Features

• add built-in ponytail lens (lazy senior dev / write less code) (666eb3e)
• add openai-compatible provider for custom OpenAI /v1 endpoints (a83c345)
• evals: add RLM-style recursive hunk-walking spike (4cb5fb7)
• give openai-compatible the long local-server timeout default (b223e01)
• support user-defined review lenses (BYO skills) (831683b)

Documentation

• add FOSS roadmap explainer with community open questions (28f01c6)
• ollama: recommend Qwen3.6-27B with RAM-based sizing guidance (ed410b3)
• remove the FOSS roadmap explainer (9964fc1)

0.1.6

Released 2026-06-14 · Compare with the previous release

Dependencies

• bump the python-dependencies group with 3 updates (5c376bf)

Documentation

• bedrock: require both inference-profile and foundation-model ARNs in IAM policy (4b34f97)

0.1.5

Released 2026-06-14 · Compare with the previous release

Performance Improvements

• e2e: single 9B local model + cached, single-fixture ollama run (64e49f9)

Documentation

• bedrock: clarify valid model ids and inference profiles (4ca94fd)
• bedrock: clarify valid model ids and inference profiles (417e64c)

0.1.4

Released 2026-06-14 · Compare with the previous release

Bug Fixes

• bundle boto3 + google-auth so keyless Bedrock and Vertex work (7da1070)

Documentation

• bump actions/checkout to v6 in docs and README (0d195e0)

0.1.3

Released 2026-06-14 · Compare with the previous release

Bug Fixes

• provider: drop model-unsupported params instead of failing the review (ff1bf3b)

0.1.2

Released 2026-06-14 · Compare with the previous release

Bug Fixes

• e2e: pool eval recall across fixtures so a one-finding miss can'… (f39eb66)

0.1.1

Released 2026-06-14 · Compare with the previous release

Features

• add performance and complexity review lenses (73d99ee)
• cli: add --uncommitted flag for reviewing only working-tree edits (8ce1cc0)
• e2e: real ollama CI run on a large multi-file diff + tunable context/timeout knobs (62733a2)
• engine: add intent lens and broaden review-prompt coverage (dc4b20b)
• gate example review workflows to trusted authors (ac61548)
• gate example review workflows to trusted authors (75a3dca)
• intent lens, broader scan prompts, and remote-primary-branch CLI comparison (321abbc)
• name provider+model and scope review marker per provider (ae2dc92)
• name provider+model and scope review marker per provider (568d863)

Bug Fixes

• align provider connector docs with the code (+ fix local Vertex/Bedrock cred detection) (c480f1a)
• cli: compare local reviews to the remote primary branch (5b06c7b)
• credentials: recognize ADC, VERTEXAI_PROJECT, and ~/.aws as ambient creds (5f30117)
• drop PyPI job from reusable release.yml (moved inline) (35bdf26)
• e2e: use qwen3:1.7b + --no-reflect so the ollama recall bar is reachable (28593f4)
• github: follow pagination when locating the existing review (e61b2ea)
• github: paginate the existing-review lookup, plus code-quality pass fixes (39f632f)
• inline PyPI publish for trusted publishing (95cc3af)
• inline PyPI publish so trusted publishing matches release-please.yml (da31c0f)
• ollama: raise default num_ctx to 32768 and tighten e2e recall floor (1095d78)
• security: broaden redaction, case-fold injection markers, escape suggestion fences (02a0cfa)
• security: broaden redaction, case-fold injection markers, escape suggestion fences (b00e2b4)
• surface the real provider error on a failed review (c264111)
• surface the real provider error on a failed review (6d166aa)

Performance Improvements

• avoid redundant PR re-fetch and cache prompt/tokenizer builds (89842a7)

Documentation

• add ARCHITECTURE.md and fix stale code comments (2b2713e)
• add Ayu-dark CLI cards and PR-comment mockups for every scan type (72ea4be)
• add PR review screenshots to README and what-gets-reviewed (af1a3f5)
• add Trust and Cost explanation and README security/cost callout (50c83c3)
• add Trust and Cost explanation and README security/cost callout (87c6a77)
• align provider connector docs with the code (90e0578)
• correct provider count to six hosted plus local ollama (d80c6b4)
• correct the stale num_ctx default from 16384 to 32768 (65fa85f)
• reframe trust/cost docs around choosing who can trigger reviews (cfe5fa4)
• refresh CLAUDE.md and releasing guide for current state (e0d8eb5)

0.1.0

Released 2026-06-07 · Compare with the previous release

Miscellaneous Chores

• release 0.1.0 (3e4fa01)

0.0.2

Released 2026-06-07 · Compare with the previous release

Features

• add local CLI review mode + user-level config (f0b8035)
• agent output format + local AI-fix loop (425ac2a)
• agent output format + local AI-fix loop; drop CLI-posting ollama workflow (6325033)
• configurable timeout, drop cost from summary, unblock small local models (5b79c4e)
• configurable timeout, drop cost summary, unblock small local models (239a93a)
• deterministic reviews (temperature=0) + skippable reflection (d214d38)
• deterministic reviews (temperature=0) and skippable reflection (91d8131)
• eval harness — measure whether a model produces usable reviews (e928cb0)
• eval harness — measure whether a model produces usable reviews (3a37efd), closes #27
• expand reviewer scan coverage (logic, tests, docs) and refresh docs (50a8eca)
• expand reviewer scan coverage (logic, tests, docs) and refresh docs (38ea0f9)
• expose timeout and temperature as GitHub Action inputs (fab5282)
• expose timeout and temperature as GitHub Action inputs (1641883)
• local CLI review mode + user-level config (5613c5e)
• parallel tracks — providers, github, engine, config/CLI, docs (8f9e2cd)
• parallel tracks — providers, github, engine, config/CLI, docs (3baffb6)
• per-category fan-out; remove cost cap and approximate cost (4a54f97)
• per-category fan-out; remove cost cap and approximate cost (d0536d9)
• review diff hunks with surrounding context lines (1849198)
• review diff hunks with surrounding context lines (07adc5e)
• step 3 integration — real adapters, slash commands, guards (4c33a19)
• step 4 packaging — Action, release pipeline, examples, current models (c60f0dc)
• step 4 packaging — action.yml, release pipeline, examples, current models (3a406fe)
• structured output for the reflection pass (c902f10)
• structured output for the reflection pass (4446ffb)
• structured outputs — constrain models to valid findings JSON (39d1958)
• structured outputs — constrain models to valid findings JSON (413f582), closes #27
• wire step 3 integration — real adapters, slash commands, guards (4c13ec8)

Bug Fixes

• reliable local reviews — provider-aware timeout/concurrency + fail loud (d3e0ccc)
• reliable local reviews — provider-aware timeout/concurrency + fail loud (8f8474b)
• use the factory-resolved model string at completion time (311d31b)

Dependencies

• bump litellm in the python-dependencies group (8e105f1)

Documentation

• add CONTRIBUTING.md (dee443e)
• add project logo (c9dcdab)
• add project logo (5368472)
• add raster favicons and apple-touch-icon (a302a32)
• drop max_cost_usd from the user-facing scope summary (dbad25b)
• explain what gets reviewed, scoping, and output shape (b43d0b3)
• fix CLI usage for the local-only review command (7e37abb)
• fold manual-steps.md into docs/, delete the file (c69ae5c)
• homepage — cover local + Action review, bullet the functionality (2fb0e58)
• homepage copy — tagline, functionality bullets, local + Action review (02c9e00)
• host the docs on GitHub Pages via MkDocs Material (9e11a46)
• left-align homepage tagline (d85597d)
• make the inline/summary feature a paragraph, not a bullet (0cddca3)
• make the inline/summary feature a paragraph, not a bullet (1ee8145)
• ollama: replace codellama with current model recommendations (0bbe319)
• record step 4 packaging in CLAUDE.md (9db1ba7)
• theme to logo colours, center hero, note context-line review (0956e61)
• theme to logo colours, center hero, note context-line review (b831b1e)
• trim manual-steps to remaining human-only actions (8ccd3d5)