1.58.0 (2026-11-03)

Features

• error-feed: send the agent's prompt and tool calls with each simulation call (bb3162d)
• simulate: fix my agent improvements tab, fix page and launch form [TH-8955] [agent] (f28588d)
• simulate: let a fix protect cases that must still pass (2747c4b)
• simulate: prove a fix for a failing run and open its pull request (4a4f226)
• simulate: ship a candidate the must-pass check vetoed, with a reason (fec9b17)

Bug Fixes

• agentic_eval: keep empty few-shot inputs skipped (a51310b)
• agentic_eval: keep zero and false few-shot inputs, skip empty ones by the eval rule (8102cea)
• agentic_eval: match few-shot text inputs as strings so structured inputs find feedback (47c94b3)
• agentic_eval: send JSON eval inputs to the embedder as text, not objects (214b225)
• agentic_eval: send JSON eval inputs to the embedder as text, not objects [TH-8996] [agent] (61923d5)
• error-feed: keep a tool-heavy call under the evidence page limit (25f1cb2)
• gateway: give every Gemini tool call its own id (88e8397)
• gateway: give every Gemini tool call its own id (7a2c1be)
• simulate: group accepted findings and revisit deferred batches (478e11a)
• simulate: offer only passing cases to protect [TH-8955] (0cbc866)
• simulate: protect only cases that passed and were scored (5277fd7)
• simulate: read the frozen plan for a refused candidate's held-back tier (aed84c6)
• simulate: remove hover tooltips from the environment workspace tabs (bd36504)
• simulate: remove hover tooltips from the environment workspace tabs [TH-8994] [agent] (0071539)
• simulate: ship a must-pass refusal only on a whole final check (d54614a)
• simulate: ship exactly the accepted candidate's files (773a190)
• simulate: skip the fix list when the org lacks fix my agent [TH-8955] (cb13120)

1.57.0 (2026-11-02)

Features

• eval-tasks: Slack alert when a continuous eval task fails or stalls (4bf7724)
• eval-tasks: Slack alert when a continuous eval task fails or stalls (453a037)
• tasks: open a run's trace in a drawer from Task Usage [TH-8959] (a98e201)
• tasks: open a run's trace in a drawer from Task Usage [TH-8959] (7208af5)

Bug Fixes

• agentcc: decrypt guardrail credentials only from the config's own organization (b9cc5cb)
• agentcc: edit guardrails in place when the org config uses the checks map (7a7fdd8)
• agentcc: let gateway request contracts accept scalar values inside config dicts (d900f62)
• agentcc: report warn-mode guardrail hits in the guardrail Test tab (0bb7603)
• agentcc: store each gateway request log under its own organization (3274a45)
• annotations: align label toolbar filter order [agent] (a07b523)
• datasets: prevent silent blank rows from Falcon and MCP (TH-8841) (a853c1d)
• datasets: prevent silent blank rows from Falcon and MCP (TH-8841) (df96ffb)
• error-feed: address review of eval membership retire and eval issue scores (0bcc4c5)
• error-feed: retire deleted eval results from issues and fix eval issue scores (215b195)
• gateway: fill pricing catalogue gaps from models.dev (4efb205)
• gateway: hide global models from keys that cannot call them (59ce271)
• gateway: hide global models from keys that cannot call them (65662da)
• gateway: keep every org-config section when saving guardrails or fallbacks (185abdc)
• gateway: keep the Future AGI Eval Base URL empty unless the user sets one (4162b8e)
• gateway: load guardrail logs and budget activity without the undeclared gateway_id (0a0bb56)
• gateway: price search requests per query and flag unpriced ones (67c4a26)
• gateway: price search requests per query and flag unpriced ones (68f20bd)
• gateway: show the gateway's real status code in the guardrail Test tab (da74570)
• gateway: treat a CrowdStrike token reply without access_token as a token failure (53015b9)
• gateway: warn when an org-config save or rollback did not reach the gateway (6039909)
• integrations: fail SQS exports on unacknowledged entries (ca932f0)
• integrations: preserve SQS checkpoints on partial batch failure [agent] (c8d4ce5)
• repo: pin LF for shell scripts so the stack boots on Windows (d027f00)
• simulate: read the environment scenario count from the server detail [TH-8912] (78b085b)
• simulate: show the live scenario count in the environment workspace [TH-8912] (5ee05b4)

1.56.0 (2026-10-27)

Features

• mcp: add error feed read tools (TH-8798) (3330cfa)

Bug Fixes

• observe: OpenRouter Broadcast list/session fallbacks, users, small costs, pricing, span types [TH-8787] (8ce2e1c)
• pricing: refresh the model catalogue without losing entries or capability flags (783e3d3)

1.55.0 (2026-10-27)

Features

• error-feed: export an issue's traces as a dataset-ready CSV (72b59a8)
• error-feed: export an issue's traces as a dataset-ready CSV (57f68c8)
• error-feed: open the share dialog from the issue page (231374b)
• error-feed: share an issue through a shared link (a6ec656)
• shared: read one trace inside a shared Error Feed issue (6e0aad3)
• shared: read-only shared Error Feed issue page (ef5ccfe)

Bug Fixes

• contracts: validate the shared sidebar query and match CI's swagger order (b7d5db2)
• error-feed: copy the cluster ID from the issue page header (d0887e7)
• error-feed: share an Error Feed issue, hotfix to main [TH-8796] (6da856d)
• error-feed: show a check on Copy cluster ID instead of a toast (3983cef)
• shared: fit the hotfix onto main (b69f98d)
• shared: no generic error toast over the shared page's own error screen (d138caa)
• shared: review fixes for the shared Error Feed issue (205a8f3)
• shared: review round 1 for the shared Error Feed issue (f663db7)
• shared: show the right shared-link error and keep signed-out viewers on the page (037a337)
• shared: show the saved analysis in the real Fix tab and stop cutting short IDs (589c1f1)

1.54.0 (2026-10-22)

Features

• simulate: agent layer and voice observability, phase 1 (TH-7957) (be6f2f3)
• simulate: real data and designer updates on the environment detail page [TH-8684] (4428549)

1.53.0 (2026-10-21)

Features

• simulate: agent layer and voice observability, phase 1, into main (TH-7957) (ffa6fd2)

Bug Fixes

• fi-collector: give the default project config dev's target_type field (TH-7957) (528c88b)
• gateway: answer 504 on a provider timeout, and keep the error writable (f904189)
• gateway: enforce the token rate limit instead of only accepting it (d872f7b)
• gateway: judge every redirect hop against the configured origin (06554b1)
• gateway: keep the Gemini API key out of URLs and error messages (635f0ad)
• gateway: send a routing alias's model_override upstream; warn on unread config keys (e9190c5)
• gateway: stop dropping caller tools and provider response fields (d727ba0)
• gateway: timeout classification, credential handling and token limits (e9e0d3a)
• simulate: the observability toggle also updates the version copies the config form reads (TH-8683) (d0cb2d7)
• simulate: the observability toggle also updates the version copies the config form reads (TH-8683) (728a068)
• tests: bring the branch's tests in line with dev (TH-7957) (4bf61b5)
• tests: the gateway's fake job carries dev's environment link (TH-7957) (1560164)
• tests: two more gateway fakes return dev's plain simulator config (TH-7957) (ac97da7)

1.52.0 (2026-10-19)

⚠️ Contains database migrations

Features

• error-feed: add Slack issue alert rules (ba5003c)
• error-feed: add Slack issue alert rules (985f806)
• simulate: add builder-session setting; sync saved scenario count (18b312b)
• simulate: environment workspace layout cleanup [TH-8603] (b82168c)
• simulate: expose scoring status on the calls API and time out stuck scoring (90e948a)
• simulate: one builder session per environment (c354847)
• simulate: send fresh builds through the builder session behind ALK_HOSTED_BUILDER_SESSION, on by default (3bcb185)
• simulate: show scoring state in run detail cells and drawers (2e3ea90)

Bug Fixes

• align generated contracts with CI organization limit (0defe42)
• e2e: validate settings module per stack (9f93367)
• error-feed: address PR review and CI blockers (25e5f6f)
• error-feed: reconcile dev migrations and contracts (885043f)
• error-feed: type alert trigger thresholds in API contract (3b5177d)
• evals: auto-select newly added evals and restore saved model on edit hydration (97c479d)
• experiments: enforce workspace scope in generated get tool (3e1bdab)
• experiments: enforce workspace scope in generated MCP reads (8356b31)
• gateway: hash org configs before taking the tenant store's write lock (bdbe945)
• gateway: hold every request at one timeout ceiling (558d677)
• gateway: hold every request at one timeout ceiling (TH-8662) (142391d)
• gateway: keep an org's last good config when the control plane cannot build it (TH-8487) (6af8261)
• gateway: keep an org's last good config when the control plane cannot build it (TH-8487) (f41296c)
• gateway: re-admit a provider after its circuit breaker cooldown (c909938)
• gateway: report the successful fallback provider timeout (TH-8662) (95346d4)
• gateway: share a failed Vertex token exchange and ignore stale invalidations (cbab74a)
• gateway: share failed Vertex token exchanges, re-admit tripped providers, hash outside the tenant lock (0de18dd)
• make OAuth invalid-attempt lockout atomic (e8e0546)
• mcp: harden OAuth throttle fallbacks (2ce9437)
• mcp: keep proxy trust scoped to OAuth throttling (792a3dd)
• mcp: make OAuth throttle fallbacks deterministic (6a73d38)
• mcp: make OAuth throttles atomic (42c189a)
• mcp: pass request into OAuth invalid paths (a53f492)
• mcp: require distributed rate limiting in production (2d7f372)
• mcp: throttle OAuth token endpoint (48623fe)
• observe: load preset dates fresh in saved views and links [TH-8634] (30247db)
• simulate: keep a run's scenarios as they ran when its chat saves, and start one run per chat request (798e849)
• simulate: keep the payload scenario count in step with the column after a chat save (2a70ec9)
• simulate: run dev's chat, adjust and recovery when the builder session is off (8224748)
• system_evals: count hits from the alignment in word_info_preserved (e97ccbf)
• system_evals: count hits from the alignment in word_info_preserved (a370f9a)
• system-evals: align SDK word information metrics (487ee32)
• system-evals: handle null transcript inputs (4e0871f)
• system-evals: sequence the system eval version (458f6da)
• system-evals: support long transcript alignment (7d10426)
• throttle MCP OAuth token endpoint (20b9dfa)

1.51.0 (2026-10-15)

⚠️ Contains database migrations: accounts/0026_authtoken_scope (lock-safe), saml2_auth/0002_idp_generation, saml2_auth/0003_login_state, tracer/0115_traceinvestigationreport_grouping_gate.

Features

• error-feed: exact grouping retrieval and the unrecovered-finding gate (#3503) (c4f633d)
• simulate: reach every hosted model through AgentCC (c4ff1b6)

Bug Fixes

• agentic_eval: sandbox JsonValidation LLM prompt rendering with SandboxedEnvironment (#3436) (d5a318c)
• feedback: tie feedback vector lifecycle to Feedback record [agent] (c0fb3bb)
• gateway: fix the native, audio and translated routes simulation traffic uses (9066a7c)
• saml: tenant isolation and session boundary (TH-8356) (ecdbe24)
• tests: point feedback vector lifecycle tests at the CH sidecar (c9e8f75)
• tests: point feedback vector lifecycle tests at the CH sidecar (c2a9220)

Performance Improvements

• simulate: speed up uncached run calls v3 listing (e1fc145)

1.50.0 (2026-10-14)

⚠️ Contains database migrations. 0002_boot_event adds one nullable boot_event_at column — purely additive, no data touched. 0009_resync_builtin_skills re-runs the builtin-skill seeder and deactivates (never deletes) any built-in Falcon skill whose tool references no longer resolve after the MCP catalog rename, since Conversation.active_skill can still point at those rows.

Features

• falcon: cut over in-app tools to the generated MCP catalog (TH-7893) (53f6f14)
• gateway: add Request Logs columns for declared custom properties [agent] (362ba31)
• observe: exact eval count tiles and per-span rollups (28b805b)
• simulate: edit an eval on environment runs and the Evaluations tab [TH-8340] (4ea4e17)
• telemetry: add OSS customer journey analytics MVP (ea35594)

Bug Fixes

• deps: redo the dev merge's lockfile resolve incrementally, not from scratch (5893418)
• deps: upgrade backend image dependencies to clear critical and high CVEs (07e68e8)
• gateway: bound metadata and header accessibility (6c9e6ce)
• gateway: retain request columns after refetch failures (a6ecd1d)
• observe: align eval filters and rollups (d820239)
• observe: name cut-off filter fields and chosen values in full on hover (b76c610)
• observe: one filter-name tooltip, only when the name is cut off (7e7bbb3)
• observe: resolve eval tile review regressions (aa4bdd6)
• observe: reveal Save view for custom date bounds and compare-only edits (TH-4883) (0d31dce)
• observe: show a cut-off filter value's full name on hover (10f001c)
• observe: show full filter names on hover (TH-8543) (53147d9)
• telemetry: isolate shared outboxes across instances (ed5f1fb)
• tests: stop NLTK corpus test depending on preinstalled data, shrink body-limit test payload (315c47b)