v1.21.0

1.21.0 (2026-10-30)

Features

• core: add experimental CodeGraph review tools behind RUSTY_CODEGRAPH (#217) (78a81b3)
• core: raise the convention file limit to 15k tokens (#214) (b2b30b9)

v1.20.0

1.20.0 (2026-10-29)

Features

• cli: analyze-findings command for consensus raw-findings artifacts (7173e02)
• consensus: file sink, cluster variants and env knobs for experiments (ad62261)
• consensus: tag raw-findings records with run id and cluster pass membership (3784abd)

Bug Fixes

• consensus: bound experiment knobs and attribute pass records to the right model (3d107ed)
• consensus: measure agreement as mean pairwise overlap between passes (799b3c3)
• consensus: treat two empty passes as agreeing (4a4fe66)
• core: distinguish unavailable code search from no matches (4fd0401)
• diff: review deletion-only hunks and match ignore patterns at any depth (97d0ada)
• github: fail fast on code-search rate limits and report unavailable searches (8735bf3)
• github: start the webhook server from a dedicated entry module (e75e229)
• prior-context: only trust the context marker in the comment header (3332fa4)
• prompts: delimit untrusted PR content in every agent prompt (292f8d4)
• review: fail fast on tool-loop exhaustion, summarize LLM errors in logs, reset step counter per attempt (0f12c3e)

Refactor

• core: replace Mastra with direct AI SDK calls (#212) (cce6852)

v1.19.0

1.19.0 (2026-10-20)

Features

• consensus: warn and flag homogeneous ensembles (1921a05)

Bug Fixes

• consensus: apply token budget to every consensus pass (e9942fd)
• consensus: clamp threshold and validate consensus config (845f8ca)
• consensus: preserve caller order through compressDiff (bbf1b0a)
• consensus: warn only on invalid configured thresholds (0ff5533)
• review: bound each review pass with a time budget (c471e29)

Refactor

• core: move normalizePath to a leaf module to break an import cycle (393e35d)

Documentation

• document RUSTY_FLAG_SCOPE_CREEP in .env.example (05e638c)
• document RUSTY_REVIEW_PASS_TIMEOUT_MS and RUSTY_REVIEW_DEADLINE_MS (df9a308)

v1.18.0

1.18.0 (2026-09-28)

Features

• models: accept :effort suffix for azure-openai models (#207) (26ad310)

v1.17.1

1.17.1 (2026-09-25)

Bug Fixes

• review: force a final answer step when RUSTY_LLM_MAX_STEPS is unset (#205) (7e72daf)

v1.17.0

1.17.0 (2026-09-25)

Features

• models: accept model:effort suffix for reasoning effort (#204) (cf9b4a5)
• review: promote skim files to the deep tier when skim fails (0db6409)

Bug Fixes

• build: clear dist/prompts before copying so template edits take effect (539597e)
• prompts: make the system prompt tier-aware (b663838)
• review: fail fast on empty completions, degrade tier failures (e9c5c1f)

v1.16.1

1.16.1 (2026-05-21)

Bug Fixes

• model: supply placeholder apiKey to createAzure for Entra ID paths so @ai-sdk/azure's pre-flight loadApiKey doesn't throw (#200) (6853c7f)

v1.16.0

1.16.0 (2026-05-21)

Features

• model: Entra ID auth for multi-deployment azure-openai/azure-foundry/azure-anthropic setups (#198) (7a3e1c5)

v1.15.0

1.15.0 (2026-05-20)

Features

• github-server: webhook delivery dedupe + concurrency semaphore (#197) (606478e)

Bug Fixes

• bug fixes and API improvements (clean subset) (#193) (9ca798b)
• consensus: honest ticket-compliance fallback when pass 0 fails (#194) (9d3d4b6)
• description: augment existing description in-place on incremental re-reviews (#190) (0d7746c)
• github-storage: serialize concurrent writes through a promise-chain mutex (#196) (fdc81a3)
• prompt: treat linked tickets as context, not a per-PR checklist (#192) (8100d7e)
• review: move anti-meta-narrative rules into the base system prompt (#189) (b64e388)

Performance

• github: cache PR-comments fetch within a provider instance (#195) (d0d8ebe)

v1.14.3

1.14.3 (2026-05-14)

Bug Fixes

• prior-context: carry forward filesReviewed so incremental re-reviews keep coverage memory (#187) (7a17a8d)