Repository navigation
v2.0.0
🔒 Security release
This release addresses security-relevant issues — updating promptly is recommended.
⚠️ Upgrade actions
Read all of this before you update. The update runs 11 database migrations, and all of them are additive (details below).
-
Back up first. The
start.shpath does not back up for you. Take a verified backup of both databases, as described indocs/05-runbooks/backups.md. -
New env keys are added for you. Two keys are new in v2.0.0:
SMTP_SECRET_KEY(encrypts the stored SMTP password) andAI_SECRET_KEY(encrypts an AI provider API key). The v2.0.0start.shappends both toinfra/env/.env.prodwhen they are missing. It backs up the file first (.env.prod.bak-<timestamp>), never changes or regenerates a key that is already there, and prints only the key names.AI_WORKER_CONCURRENCYis optional (default4) and is not written. No new key needs a value from you. -
Update.
git pull+start.shusers:git fetch --tags && git checkout v2.0.0 ./infra/start.shstart.shrecognises the existing install, keeps every secret, rebuilds the images and runs the migrations.- Do not use
./infra/update.shfor this release. Since v1.1.0 it cannot apply a release tag: it requires a signed tag and the release tags are unsigned, so it stops safely at the tag check after taking its backup (#1458). It also never edits.env.prod, so it would stop on the missingSMTP_SECRET_KEY. Use thestart.shpath above; the in-app Update button's command has the same limitation until #1458 ships.
-
What you will see.
- There is a brief outage while the migrations run and the containers are recreated. The first build takes longer because of the new Bun 1.4.2 base image.
- Search starts empty and fills itself. Meilisearch 1.53.2 cannot open the 1.12.3 data, so it starts on a new volume,
lazyit-prod_meili_data_v1_53_2. The API rebuilds every index from Postgres in the background, and results can be incomplete for a few minutes. There is nothing to run. - Signed-in users stay signed in. Sessions from before the update do not appear in the new sessions list until the user signs in again.
- Diagrams use the Mermaid 12 look. Node labels are now plain text, so HTML inside a label is shown as text (SEC-084).
- The web app sends a
Content-Security-Policy-Report-Onlyheader, which reports and does not block. If you run your own reverse proxy in front of lazyit, do not add a second CSP. - Caddy now sends
/mcp,/.well-known/oauth-*,/.well-known/openid-configuration*,/oauth/token|register|revokeand/authorize|token|registerto the API. Streamed responses are no longer compressed. If you have your own proxy in front of Caddy, pass those paths through unchanged. - Signing out of the web app no longer revokes MCP or OAuth grants. Those end when the password changes, when the account is deactivated or offboarded, or when they are revoked from
/account/ai. - The assistant and MCP stay off instance-wide until an admin enables them. Once enabled, MEMBERs get
ai:useandai:connectby default (seeded once), and VIEWERs do not. Permissions you revoked earlier stay revoked.
-
Local-auth instances: rotate passwords (SEC-085). From v1.3.0 through v1.11.0, user reads returned the argon2id
passwordHash. The worst path wasGET /assets/:id, which any VIEWER can read, and which included the hash of each asset's owner, admins included. v2.0.0 closes the leak, but a hash that was already read can be cracked offline. After updating, reset passwords for admins first, then everyone else. Changing a password ends that user's other sessions and their MCP credentials. OIDC instances are not affected, because no hash is stored there. -
Clean up the old search volume when you are ready. The 1.12.3 volume is left in place so you can roll back to v1.11.x. Once you are staying on v2.0.0, remove it:
docker volume rm lazyit-prod_meili_data
Migrations (all additive): notification_read_dismissed_at, add_applied_role_permission_defaults (backfilled from current grants and revokes), add_ai_assistant_and_oauth (new tables, plus nullable aiInvocationId on asset_history and user_history), mcp_credential_epoch (defaulted column), mcp_allow_any_https_client_default, consumable_deliveries (nullable columns, 2 enum values, CHECKs that every existing row already satisfies), user_activation_history (2 enum values, view recreated), ai_conversation_settings, ai_web_search, user_ui_preferences (nullable locale and theme), and user_sessions (new table, 1 enum value). Nothing is dropped, renamed or made NOT NULL without a default. The only non-trivial locks are the index and CHECK builds on consumable_movements, which take seconds and happen during the outage anyway.
New & Changed
- feat(web): document your sessions in the Manual (#1420)
- feat(web): label and tint the session_ended activity (#1420)
- feat(web): explain a revoked session on change password (#1420)
- feat(web): list and end my sessions on the account security panel (#1420)
- feat(web): forward the browser's User-Agent and X-Forwarded-For at sign-in (#1420)
- feat(api): cap sessions per user and purge long-idle remember-me rows (#1420)
- feat(web): language and theme follow the user to a new browser (#1422)
- feat(web): edit my own first and last name on /profile (#1421)
- feat(api): order the user-sessions migration after the latest one on dev (#1420)
- feat(api): list your own sessions and end one (#1420)
- feat(api): open a session on sign-in and end them with the epoch levers (#1420)
- feat(api): check the session id of local tokens in the guard (#1420)
- feat(api): add the per-device session contract to the shared package (#1420)
- feat(api): add the UserSession model and its additive migration (#1420)
- feat(api): save each user's language and theme on the server (#1422)
- feat(api): let a signed-in user edit their own first and last name (#1421)
- updt(brand): new demo video and poster
- feat(web): add the refusal.mutationCapBatch sentence to the en and es catalogs (#1384)
- feat(web): show the localized sentences on approval cards, tool details and refused calls (#1384)
- feat(web): render server-built sentence lists in the user's locale, falling back to the English (#1384)
- feat(web): add the en and es catalogs for the assistant's server-built sentences (#1384)
- feat(web): the asset model form can clear the category (#1315)
- feat(web): an explicit no-category row in the category combobox (#1315)
- feat(web): the asset create form shows the next tag to members (#1315)
- feat(web): read the member-safe asset-tag scheme summary (#1315)
- feat(api): write tools build preview sentences and summaries from codes (#1384)
- feat(api): carry sentence codes through results, events and core refusals (#1384)
- feat(api): closed list of localizable AI sentence codes in the shared contract (#1384)
- feat(web): list the changed fields on asset UPDATED history (#1382)
- feat(web): document the client domain and the loopback install path (#1315)
- feat(web): lead with the download when no marketplace is served (#1315)
- feat(web): show the verified domain at consent and in connected apps (#1315)
- feat(web): lead the preview table with the asset column for update batches (#1409)
- feat(web): approve all and reject all cover untrusted-source changes (#1409)
- feat(api): document CIMD as built and in the Manual (#1315)
- feat(api): test CIMD fetch, SSRF refusals, cache, fallback and the end-to-end flow (#1315)
- feat(api): resolve https client_ids through CIMD at the authorization endpoint (#1315)
- feat(api): fetch, validate and cache Client ID Metadata Documents through the egress guard (#1315)
- feat(api): share the client display sanitizers between DCR and CIMD (#1315)
- feat(api): CIMD audit actions and consent verifiedDomain in the shared contract (#1315)
- feat(web): Manual absorbs the text moved off Settings → AI (#1407)
- feat(web): cut Settings → AI down to a configuration page (#1407)
- feat(api): label asset_update_batch and its asset column in en and es (#1409)
- feat(api): add asset_update_batch, many asset edits in one card (#1409)
- feat(web): paged approval card with approve all and reject all for several changes (#1409)
- feat(web): plan paged approvals, bulk eligibility and refused-call collapsing (#1409)
- feat(web): pointer-following glow on the AI assistant entry points (#1405)
- feat(web): draw the AI assistant with Hugeicons (#1406)
- feat(web): reach the account hub from the user menu (#1404)
- feat(web): account hub page at /account (#1404)
- feat(web): shared account sub-navigation across /account and /profile (#1404)
- feat(api): tell the assistant the pending-approval limit and to work in batches (#1409)
- feat(api): defer proposals past the pending-approval limit instead of failing them (#1409)
- feat(web): show web search sources under the assistant's answer (#1389)
- feat(web): web search card in Settings -> AI (#1389)
- feat(api): web search in the chat runtime, frozen per conversation (#1389)
- feat(api): prompt rules for unknown terms and web search, version 5 (#1389)
- feat(api): declare the provider-native web search on model steps (#1389)
- feat(api): web search switch in the instance AI settings (#1389)
- feat(shared): web search contract for the AI assistant (#1389)
- feat(web): link the asset tag scheme ref to the instance settings page (#1394)
- feat(api): tell asset_create and asset_create_batch to leave the tag to the scheme (#1394)
- feat(web): label the asset tag scheme tools, fields, entity and warning (#1394)
- feat(api): read and follow the instance asset tag scheme from the AI tools (#1394)
- feat(shared): add the INSTANCE_CONFIGURATION warning and the assetTagScheme entity type (#1394)
- feat(web): render the assistant's input form card in the chat (#1388)
- feat(web): answer the assistant's input forms through the API (#1388)
- feat(web): table approval card for bulk assistant changes (#1387)
- feat(web): shape list-of-record previews into a table model (#1387)
- feat(api): exact asset tag and serial list filters on GET /assets (#1387)
- feat(web): label request_input and the AWAITING_INPUT tool status (#1388)
- feat(api): tell the chat assistant to ask for missing data with a form (#1388)
- feat(api): pause chat runs AWAITING_INPUT and answer input requests (#1388)
- feat(api): add the chat-only request_input tool (#1388)
- feat(shared): add the input-request contract and the AWAITING_INPUT status (#1388)
- feat(web): en/es labels for the taxonomy tools (#1390)
- feat(api): taxonomy management tools for the assistant (#1390)
- feat(api): planning rules in the assistant prompt, prompt version 3 (#1386)
- feat(web): labels for the bulk asset creation tool and its preview fields (#1387)
- feat(api): default new-asset status and bulk asset creation tool (#1386, #1387)
- feat(web): per-chat model picker, effort, temperature and auto-approve switch with consent (#1373, #1376)
- feat(web): show auto-approved writes as an applied-automatically record (#1376)
- feat(web): slash commands with one argument, /model and /auto (#1373, #1376)
- feat(web): chat settings data layer — model catalog, create body, settings PATCH (#1373, #1376)
- feat(api): conversation settings routes and GET /ai/models (#1373, #1376)
- feat(api): per-conversation model and auto-approve in the agent runtime (#1373, #1376)
- feat(api): auto-approve path in the AI core with AUTO provenance (#1376)
- feat(api): additive columns for conversation settings and approval mode (#1373, #1376)
- feat(shared): per-conversation AI settings and model catalog contract (#1373, #1376)
- feat(api): filter GET /users and user_search by activation (#1375)
- feat(web): label the new KB folder AI tools and the path preview field (#1377)
- feat(web): en/es messages for the AI panel width, slash commands, tool and field labels (#1371, #1372, #1377)
- feat(web): slash-command palette in the AI composer with /copy, /new and /help (#1372)
- feat(api): add kb_folder_create and kb_folder_rename AI tools (#1378)
- feat(web): embed the shared Claude Code install panel in the MCP card (#1315)
- feat(web): consumables on the offboarding sheet and the Return Act, with per-row selection (#1364)
- feat(web): adopt the stable /config/ai error codes, the curated MCP clients and the server-known endpoint (#1315)
- feat(web): label the consumable delivered and returned activity verbs (#1364)
- feat(web): consumable deliveries panel with deliver and return on user, asset and location detail (#1364)
- feat(web): deliver to a person, asset or location from the Remove dialog, returnable flag and ledger destinations (#1364)
- feat(web): consumable deliveries data layer, payload builders and target resolver (#1364)
- feat(api): cover the any-HTTPS default in the OAuth client policy tests (#1315)
- feat(api): default mcpAllowAnyHttpsClient to true in the ai_settings column (#1315)
- feat(shared): accept any HTTPS MCP client by default (#1315)
- feat(web): consumable delivered and returned events on the asset timeline (#1364)
- feat(api): revoke MCP credentials on credential events, not on logout (#1315)
- feat(api): bind OAuth grants and personal tokens to mcpCredentialEpoch (#1315)
- feat(api): add the mcpCredentialEpoch counter for MCP credentials (#1315)
- feat(api): server-known MCP endpoint and marketplace URL on GET /ai/status (#1315)
- feat(api): consumable deliveries to a user, asset or location, returns and the deliveries read (#1364)
- feat(web): let a sign-out land on a given sign-in path (#1315)
- feat(api): stable machine codes on every /config/ai refusal (#1315)
- feat(api): curated MCP client allowlist defaults move to the shared contract (#1315)
- feat(shared): consumable delivery targets, returns and deliveries read contract (#1364)
- feat(api): add consumable delivery targets, returnable and returns migration (#1364)
- feat(web): add the per-Service-Account AI access control (#1315)
- feat(web): add Settings → AI — setup wizard, editor, limits, MCP and turn-off (#1315)
- feat(web): add the AI hub card and per-account AI access copy (#1315)
- feat(web): add the Settings → AI form, error-mapping and MCP helpers (#1315)
- feat(web): add the /config/ai endpoints and hooks (#1315)
- feat(web): vendor the Radix radio-group primitive (#1315)
- feat(web): link AI & connected apps from the user menu (#1315)
- feat(web): add the oauth message catalog in en and es (#1315)
- feat(web): add /account/ai with the MCP install panel and connected apps (#1315)
- feat(web): add the OAuth consent page (#1315)
- feat(web): add the OAuth endpoints and connected-apps hooks (#1315)
- feat(web): keep the query string in the sign-in callbackUrl (#1315)
- feat(web): mount the AI chat in the assistant panel slot (#1315)
- feat(web): add the AI chat panel, history, messages and composer (#1315)
- feat(web): add the AI approval card with warnings and password step-up (#1315)
- feat(web): render chat markdown without images, raw HTML or auto-links (#1315)
- feat(web): add the AI chat copy in English and Spanish (#1315)
- feat(web): add the AI conversation, run, stream and decision endpoints and hooks (#1315)
- feat(web): add the approval preview, untrusted text, link and history presenters (#1315)
- feat(web): map AI run and approval decision errors to recoveries (#1315)
- feat(web): plan the chat's UI effects from tool results (#1315)
- feat(web): add the AI run event parser and the chat stream reducer (#1315)
- feat(web): build safe entity links and the page context for the AI chat (#1315)
- feat(web): add the unsaved-changes registry the chat checks before navigating (#1315)
- feat(api): cover the owner delete of a conversation (#1315)
- feat(api): add DELETE /ai/conversations/:id through the W3-6 purge service (#1315)
- feat(api): export the mapper's template path parser for the AI previews (#1315)
- feat(api): pass PREVIEW_CHANGED and addedWarnings through the decision endpoint (#1315)
- feat(web): label and route the mcp.client_connected notification (#1315)
- feat(api): serve the MCP resource server at /mcp (#1315)
- feat(api): email the new AI agent connection notice (#1315)
- feat(api): mint, list, revoke and verify personal MCP tokens on lan (#1315)
- feat(shared): add the mcp.client_connected notification type (#1315)
- feat(shared): let a personal MCP token be read-only or read & write (#1315)
- feat(api): serve the authenticated plugin download and the HTTPS-only public marketplace (#1315)
- feat(api): render the Claude Code plugin from templates, the primer and the tool registry (#1315)
- feat(api): add the chat-only workflow authoring tools (#1315)
- feat(api): add AI conversation retention sweeper and purge service (#1315)
- feat(api): cover the AI HTTP surfaces over the real runtime (#1315)
- feat(api): add the audited per-Service-Account AI access setting (#1315)
- feat(api): add /ai/runs with the SSE event stream and the decision endpoint (#1315)
- feat(api): add the owner-only /ai/conversations surface (#1315)
- feat(api): project AI transcripts to the provider-neutral wire shape (#1315)
- feat(api): cover the workflow operations tools end to end (#1315)
- feat(api): add the workflow operations AI tools (#1315)
- feat(api): cover critical applications and user lookup in the access spec (#1315)
- feat(api): refuse off-chat writes on critical applications and resolve users by email or name (#1315)
- feat(api): refuse MCP and headless writes on critical applications (#1315)
- feat(api): explain access automation in the domain primer, prompt version 2 (#1315)
- feat(api): pre-create the workflow operations and authoring toolsets (#1315)
- feat(api): require step-up for critical-application actions on any write preview (#1315)
- feat(shared): add OUTBOUND_INTEGRATION and CRITICAL_APPLICATION preview warnings (#1315)
- feat(api): cover the AI runtime loop, recovery and step-up with specs (#1315)
- feat(api): wire the AI runtime module and export RUN_EVENT_BUS (#1315)
- feat(api): add the ai-run worker and the run sweeper (#1315)
- feat(api): add the AI approval service with password step-up (#1315)
- feat(api): add the AI agent loop and run orchestrator (#1315)
- feat(api): add the AI run event bus, limits and runtime building blocks (#1315)
- feat(api): show the application's criticality on a revoke preview (#1315)
- feat(api): test the asset and reference tools for route parity, previews and the write path (#1315)
- feat(api): add the asset tools: search, get, create, update, archive, restore, check-out and check-in (#1315)
- feat(api): add the reference tools reference_lookup, asset_model_create and location_create (#1315)
- feat(api): cover the access toolset with route parity and write-path specs (#1315)
- feat(api): add the access AI toolset for applications, grants and requests (#1315)
- feat(api): cover the knowledge-base AI tools against the real folder ACL (#1315)
- feat(api): add the knowledge-base AI tools and their catalog rows (#1315)
- feat(api): add the activity AI toolset — dashboard summary and activity feed (#1315)
- feat(api): add the users AI toolset — search, get, create, update, offboard, restore (#1315)
- feat(api): add the consumables AI tools with route-parity and write-path specs (#1315)
- feat(api): require step-up for privilege grants and credential delivery in core (#1315)
- feat(shared): add PRIVILEGE_GRANT and CREDENTIAL_DELIVERY preview warnings (#1315)
- feat(api): fail the suite when the AI SDK is imported outside the provider layer (#1315)
- feat(api): implement ChatModelPort over AI SDK 7 and bind CHAT_MODEL_PORT (#1315)
- feat(api): add the Anthropic, OpenAI, Gemini and OpenAI-compatible provider definitions (#1315)
- feat(api): add the provider contract, error classification and egress-guarded provider fetch (#1315)
- feat(api): redact OAuth token and revoke credential fields from logs (#1315)
- feat(api): add the OAuth 2.1 authorization server endpoints, tokens and connected apps (#1315)
- feat(api): scope OAuthGrant reads to live grants in the soft-delete extension (#1315)
- feat(api): add the OAuth issuer gate, credential primitives and MCP client trust policy (#1315)
- feat(api): mark the AI settings and status routes unexposed to the tool catalog (#1315)
- feat(api): add GET /ai/status (#1315)
- feat(api): add /config/ai with the enable gate, write-only key and config audit (#1315)
- feat(api): add the AI connection tester over the chat-model port (#1315)
- feat(api): add a generic AES-256-GCM envelope cipher keyed by an env var (#1315)
- feat(api): ledger-backed AI writes and the propose/approve/reject lifecycle (#1315)
- feat(api): add the AI reference resolver and rt.resolve (#1315)
- feat(api): add the insert-only AiActionLog writer with input redaction (#1315)
- feat(api): add the infra read tools infra_node_search and infra_node_get (#1315)
- feat(api): add the versioned system-prompt, MCP instructions and turn-context builders (#1315)
- feat(api): add the lazyit domain primer for the AI assistant (#1315)
- feat(api): register AiModule, OAuthModule and McpModule (#1315)
- feat(api): pre-create the AI, OAuth and MCP module skeleton (#1315)
- feat(api): stamp aiInvocationId on asset and user history (#1315)
- feat(api): add the reference toolset and pre-create every domain toolset (#1315)
- feat(api): declare the model-call, run-event-bus and settings ports (#1315)
- feat(api): add the AI tool registry, in-process dispatcher and executor (#1315)
- feat(api): prove the delegated branch is DB-first and unreachable from the network (#1315)
- feat(api): add the delegated-identity branch to JwtAuthGuard (#1315)
- updt(api): extract the principal loader and SA authenticator from JwtAuthGuard (#1315)
- feat(web): mount the AI assistant shell in the app layout (#1315)
- feat(web): register the ai, aiSettings and oauth message namespaces (#1315)
- feat(web): add the AI status read with a fail-closed gate (#1315)
- feat(web): add a fetch-based SSE reader and apiFetchStream (#1315)
- feat(shared): allow private-use redirect schemes on explicit MCP allowlist entries (#1315)
- feat(api): store the MCP client allowlist overlay on ai_settings (#1315)
- feat(shared): add the MCP client allowlist contract (#1315)
- feat(api): add the AI assistant and OAuth data model (#1315)
- feat(shared): add the AI assistant, MCP and OAuth wire contracts (#1315)
- feat(web): label the AI permissions in the role and permission settings (#1315)
- feat(shared): add the ai:use and ai:connect permissions (#1315)
- feat(api): add the seed-once ledger for default role permissions (#1314)
- feat(infra): generate the optional AI_SECRET_KEY in start.sh (#1322)
- feat(infra): route MCP and OAuth paths to the API and stream SSE uncompressed (#1322)
- feat(shared): add the dismiss-all upTo query contract (#1309)
- feat(web): dismiss notifications from the bell and clear all (#1309)
- feat(web): add optimistic dismiss hooks for the notification bell (#1309)
- feat(api): let users dismiss notifications from their own bell (#1309)
- feat(web): revoke the session server-side on sign-out (#1307)
- feat(web): add "Keep me signed in" to the local login form (#1307)
- feat(shared): add the dismiss-notifications result contract (#1309)
- feat(api): add nullable dismissedAt to the notification read join (#1309)
- feat(api): add keep-me-signed-in local sessions and server-side sign-out (#1307)
- feat(shared): add rememberMe to the login contract and expose session expiry (#1307)
- feat(web): give topology edges a decorative packet flow (#1295)
- feat(web): draw user endpoints on the topology map by default (#1295)
- feat(web): create, rename and move KB folders from the folder tree (#1291)
- feat: expose the ADR index to every agent (#1279)
- feat: add path-scoped rules per boundary (#1279)
- updt: make AGENTS.md the canonical instruction source (#1279)
- feat: add the repository charter (#1279)
- updt(api): move every @nestjs/* package to v12 in one step
- updt(shared): @noble/hashes 2.4.0 (#1276)
- updt(web): next + eslint-config-next 16.3.3, @noble/hashes 2.4.0 (#1276)
- feat(api): expose password-reset capabilities and the optional delivery body (#1268)
- feat(api): let an admin choose the password-reset delivery in local mode (#1268)
- feat(api): resolve the origin an admin reset link is built against (#1268)
- feat(api): mint + send an admin-initiated password-reset link (#1268)
- updt(web): password-reset copy that is true in both auth modes, en + es (#1268)
- feat(web): add the local-mode password-reset delivery dialog (#1268)
- updt(web): reuse the shared reveal in local onboarding (#1268)
- feat(web): extract the shown-once temporary-password reveal (#1268)
- feat(web): add the password-reset capabilities query (#1268)
- feat(web): carry the reset delivery choice through the mutation hook (#1268)
- feat(web): type the admin reset outcome and read its capabilities (#1268)
- feat(shared): admin password-reset delivery contract (email link vs temp password) (#1268)
- updt: sync BullMQ v5 lockfile
- updt: raise BullMQ v5 floor
Fixes
- fix(web): the workflow list never enables without the reviewed version (SEC-077)
- fix(infra): test start.sh key appending and run the start.sh tests in CI (#1459)
- fix(infra): start.sh appends missing, safely generatable keys to an existing .env.prod (#1459)
- fix(api): boot self-heal waits for Meilisearch before probing the indexes (#1216)
- fix(api): make change-password conditional on the caller's session epoch (#1420)
- fix(web): add an opt-in hydration smoke script (#1448)
- fix(web): take "now" from the server-seeded useNow instead of Date.now (#1448)
- fix(web): read config status hydration-safely outside /settings/instance (#1448)
- fix(web): make the caller's user and permission queries hydration-safe (#1448)
- fix(api): select asset owners through the public user allowlist (SEC-085)
- fix(api): serialize User through a public column allowlist (SEC-085)
- fix(api): decide the AI exposure of PATCH /users/me (#1421)
- fix(web): clear a model's SKU and description on edit (#1441)
- fix(web): document the Content-Security-Policy for operators in the Manual (#1440)
- fix(web): send a nonce-based Content-Security-Policy, report-only first (#1440)
- fix(api): asset_model_update clears the SKU and description with null (#1441)
- fix(api): cover clearing an asset model's SKU and description on PATCH (#1441)
- fix(shared): UpdateAssetModelSchema accepts sku and description null to clear them (#1441)
- fix(deps): override lodash-es to 4.18.1 under mermaid's chevrotain (#1402)
- fix(api): scan in-flight runs without the removed
pausedjob state (#1402) - fix(api): asset_model_update clears the category with category null (#1315)
- fix(api): cover clearing an asset model's category on PATCH (#1315)
- fix(shared): UpdateAssetModelSchema accepts categoryId null to clear it (#1315)
- fix(web): Mermaid labels render as SVG text, not HTML (SEC-084)
- fix(api): count a batch's rows toward the per-SA mutation cap (SEC-081)
- fix(web): label the toolResult untrusted-source marker (SEC-080)
- fix(api): mark reads of other-authored text as untrusted sources (SEC-080)
- fix(api): bound the /mcp query-token scan behind the per-IP limiter (SEC-083)
- fix(api): consent admin step-up uses the shared backoff and audits failures (SEC-082)
- fix(api): make the password step-up verifier one shared auth primitive (SEC-082)
- fix(api): format and lint the changed files for the bumped prettier (#1315)
- fix(web): let diagrams take the mermaid 12 look and follow the theme toggle (#1402)
- fix(api): member-safe asset tag scheme read for asset creators (#1315)
- fix(api): specific refusal when the provider has web search disabled (#1315)
- fix(api): re-check approval card impact counts at approve (#1315)
- fix(web): neutral label for the asset UPDATED history detail (#1382)
- fix(api): plain asset edits write an UPDATED history row with the changed field names (#1382)
- fix(api): don't report the plugin marketplace on a loopback origin (#1315)
- fix(infra): route the OAuth fallback and OIDC discovery paths to the API (#1315)
- fix(api): document the CIMD review fixes in mcp-and-oauth §15 (#1315)
- fix(api): time-box CIMD fetch incl. DNS, bundle only on network failure, narrow race catch (#1315)
- fix(api): connected apps use the consent verified rule and carry verifiedDomain (#1315)
- fix(api): never sweep a client with a pending code; age CIMD rows by last refresh (#1315)
- fix(web): say that connected AI apps survive signing out on all devices (#1404)
- fix(ai): stop a tool call that keeps failing the same way (#1403)
- fix(ai): normalize request_input before validating it (#1403)
- fix(ai): send OpenAI function tools with strict false (#1403)
- fix(web): web search disclosure covers search partners, the lasting no-auto-approve rule and OpenAI (#1389)
- fix(api): run OpenAI web search without live page fetches (#1389)
- fix(api): keep the web search untrusted mark for the whole conversation (#1389)
- fix(web): require a real session user in server-side auth guards (#1399)
- fix(web): upgrade next-auth to 5.0.0-beta.32 for Auth.js advisories (#1399)
- fix(web): keep a sent input form closed and expire it client-side (#1388)
- fix(web): let Combobox and EntityMultiSelect take describedby and required (#1388)
- fix(web): name both causes of an incomplete duplicate check, voseo in es (#1387)
- fix(web): never build a table from a redacted preview value (#1387)
- fix(api): wrap answer values picked from lazyit name lists as untrusted (#1388)
- fix(web): stop following a run paused for input and track its forms (#1388)
- fix(api): wrap lazyit option labels in an input answer as untrusted (#1388)
- fix(web): sort the batch labels and label the incomplete duplicate check (#1387)
- fix(api): bind the batch create to its approved card (#1387)
- fix(api): honour cancel and AI-off before auto-approving; record the run model under the lock (#1373, #1376)
- fix(api): re-check auto-approve in the claim and refuse after an untrusted read (#1376)
- fix(web): localize AI tool and preview field names and collapse repeated tool lines (#1377)
- fix(web): fill the viewport with the AI panel and make it resizable as an overlay (#1370, #1371)
- fix(web): label and tint the user activation activity verbs (#1375)
- fix(api): let AI search tools list with filters only (#1374)
- fix(api): audit activation and legajo/username changes on user update (#1375)
- fix(api): add DEACTIVATED/REACTIVATED user history events to the activity view (#1375)
- fix(shared): add activation events to the user-history and activity contracts (#1375)
- fix(web): keep the Return Act's consumables notes when rows are unfetched or hidden (#1364)
- fix(web): step the deliveries panel back when a return empties its last page (#1364)
- fix(web): word the address warning for both connection modes (#1315)
- fix(web): build install snippets from the /ai/status endpoint and marketplace URL (#1315)
- fix(web): warn when a connection URL carries credentials (#1315)
- fix(api): keep legacy userinfo connection URLs working at run time (#1315)
- fix(api): flag legacy connection URLs with userinfo on read (#1315)
- fix(web): move the key-hygiene save hooks to a client module so the page builds (#1315)
- fix(web): update the oauth copy for the origin check and the consent sign-out (#1315)
- fix(web): show a custom redirect scheme with its host on the consent page (#1315)
- fix(web): keep the minted token and the consent answer out of the mutation cache (#1315)
- fix(web): build OAuth install snippets from the server-known origin (#1315)
- fix(web): keep a Service Account's write cap across access changes and explain a blocked save (#1315)
- fix(web): keep limits edits across other saves and clear stale save errors (#1315)
- fix(web): wait for /ai/status before naming the MCP mode, and explain the WEB_ORIGIN rule (#1315)
- fix(web): drop the provider key from the mutation cache once a request settles (#1315)
- fix(web): show the full destination of external links in the AI chat (#1315)
- fix(web): use the step-up password once and keep it out of the mutation cache (#1315)
- fix(web): read the AI conversation fresh on every panel open (#1315)
- fix(web): send expectedVersion on enable and round-trip redacted header values (#1315)
- fix(api): AI workflow tools send version preconditions and share the userinfo refusal (#1315)
- fix(api): dry-run refuses offboarded or revoked samples and redacts header values (#1315)
- fix(api): honour expectedVersion on workflow update and baseVersion on authoring (#1315)
- fix(api): refuse URL userinfo on connection write and at workflow egress (#1315)
- fix(api): redact connection header values and gate header changes under workflow:secrets (#1315)
- fix(shared): add URL userinfo helpers, redaction sentinel and version preconditions (#1315)
- fix(api): cover article restore in the SEC-074 folder-hidden 404 tests (#1315)
- fix(api): run the folder ACL before the author 403 in article restore (SEC-074, #1315)
- fix(api): test folder-hidden articles 404 on every KB write path (SEC-074, #1315)
- fix(api): run the folder ACL before the author 403 in loadOwned (SEC-074, #1315)
- fix(api): show in-flight runs on a re-point and refuse legacy userinfo connections (#1315)
- fix(api): remember a connection notice only once it exists (#1315)
- fix(api): count the personal-token cap under a lock and re-assert scopes (#1315)
- fix(api): hold the per-SA write cap on /mcp as writes per rolling hour (#1315)
- fix(api): revoke on sight a token presented in a /mcp query string (#1315)
- fix(api): keep a credential sent in a query string out of the request log (#1315)
- fix(api): close the workflow operations golden block after the merge (#1315)
- fix(api): refuse an Idempotency-Key reused for another request (#1315)
- fix(api): drop a stream whose client left during load, re-check its caller each heartbeat (#1315)
- fix(api): make the workflow authoring cards describe exactly what is sent (#1315)
- fix(api): refuse fenced shell blocks, keep tool-index cells on one line and name the marketplace after the host (#1315)
- fix(api): re-derive step-up from the fresh preview at approve (#1315)
- fix(api): never purge a conversation holding an in-flight tool invocation (#1315)
- fix(api): address the G2 review of the workflow operations tools (#1315)
- fix(api): give the event stream spec room under a loaded runner (#1315)
- fix(api): require step-up / refuse offboarding across critical applications (#1315)
- fix(api): show the title and whole body on a published-move card and bound KB summary fields (#1315)
- fix(api): tell the model critical applications change only in the chat (#1315)
- fix(api): number run events from one process-wide counter and create no buffer on subscribe (#1315)
- fix(api): give every tool call of a run a non-empty unique id (#1315)
- fix(api): neutralize forged turn-context tags with whitespace, attributes or no closing bracket (#1315)
- fix(api): answer calls of a faulted run UNKNOWN_OUTCOME and mark its interrupted writes (#1315)
- fix(api): re-check the run before resolving calls or pausing, and keep the sweeper off live runs (#1315)
- fix(api): persist the turn's untrusted sources in every step record so a resume keeps them (#1315)
- fix(api): refuse an approval while AI is off or when the run is not awaiting a decision (#1315)
- fix(api): count a step-up attempt before the KDF and allow one in flight per user (#1315)
- fix(api): a partial directory scan never resolves a username or legajo (#1315)
- fix(api): scope the primer's critical-application confirmation to the chat (#1315)
- fix(api): say step-up applies to write and elevated previews alike (#1315)
- fix(api): log AI runtime faults by error class and code, never the raw message (#1315)
- fix(api): harden the asset tools' references, spec keys and agent-written text after the G2 review (#1315)
- fix(api): cover the access tools review fixes in the spec (#1315)
- fix(api): name the grantee on access grant and approval cards (#1315)
- fix(api): hide inert SA-ungrantable grants in the service account read shape (SEC-073, #1315)
- fix(api): strip SA-ungrantable permissions at principal load (SEC-073, #1315)
- fix(api): name the KB folder and its audience on the card and refuse blind creates (#1315)
- fix(api): pin the manager by id, flag restored roles and the vault drop, resolve usernames with @ (#1315)
- fix(api): pin KB tool references, show the full visible body and page by serialized size (#1315)
- fix(api): refuse a consumable reference resolved from a partial page (#1315)
- fix(api): a manager-only user_update needs no step-up — LEDGER_APPEND instead of IDENTITY_CHANGE (#1315)
- fix(api): keep the provider module spec off the generated Prisma client (#1315)
- fix(api): harden the AI SDK import-boundary spec (#1315)
- fix(api): give model listing its own deadline and abort signal (#1315)
- fix(api): refuse SDK file downloads and turn off SDK telemetry on model steps (#1315)
- fix(api): cap every provider response body and classify the overflow (#1315)
- fix(api): parse OAuth redirects with URL, refusing userinfo and unparseable URIs (#1315)
- fix(api): harden the AI write path per review and derive step-up in core (#1315)
- fix(api): extend the ledger redaction keys and mask URL credentials (#1315)
- fix(api): harden AI base URL validation and honor shim mode in the connection test (#1315)
- fix(api): report the chat unavailable when the stored AI key does not decrypt (#1315)
- fix(api): make the AI settings save a conditional write (#1315)
- fix(shared): classify MCP redirect URIs from the parsed host and refuse userinfo (#1315)
- fix(api): cover the directory offboard last-admin skip, SEC-021 (#1319)
- fix(api): skip the last active admin in the directory offboard sweep (#1319)
- fix(api): expose the last-admin predicate as hasAnotherActiveAdmin (#1319)
- fix(api): apply default role permissions once per instance (#1314)
- fix(api): compare asset specs iteratively in jsonDeepEqual (#1321)
- fix(shared): bound asset specs structure on write (#1321)
- fix(api): rewrap the last-admin guard docstring (#1319)
- fix(api): cover the last-admin guard against isActive, SEC-021 (#1319)
- fix(api): guard deactivation and count only active admins in the last-admin guard (#1319)
- fix(shared): close the host:port carve-out bypass in isSafeApplicationUrl (#1320)
- fix(shared): cover the SEC-051 url scheme guard bypass in tests (#1320)
- fix(web): hand the real API client back after endpoint tests stub it (#1307)
- fix(web): keep keyboard focus in the bell after Clear all (#1309)
- fix(web): send upTo with Clear all so newer notifications stay (#1309)
- fix(api): retry dismiss-all once when retention prunes a target (#1309)
- fix(api): bound dismiss-all to an optional upTo datetime (#1309)
- fix(web): tidy the comments and test layout of the 401 callbackUrl change (#1307)
- fix(web): keep the remember-me warning's live region mounted (#1307)
- fix(web): keep the destination when the 401 handler sends a user to /login (#1307)
- fix(api): revoke local sessions when the directory sync offboards a user (#1308)
- fix(web): end expired sessions server-side and stop the /login bounce loop (#1307)
- fix(api): revoke local sessions on deactivation and offboarding (#1307)
- fix(api): gate the KB article move on destination folder access (#1296)
- fix(web): name .docx and .zip where the article import is chosen (#1292)
- fix(web): split the KB reading rail on available width, not on
xl(#1292) - fix(deps): pin @types/bun to a semver range instead of the latest dist-tag (#1288)
- fix(web): trace the @swc/helpers ESM half into the standalone output (#1274)
- fix(web): the create-form password hint claimed lazyit never stores it (#1268)
- fix(web): the password reset was permanently disabled in local auth mode (#1268)
- fix(infra): document SMTP_SECRET_KEY in the prod env template (#1269)
- fix(infra): generate SMTP_SECRET_KEY in start.sh, preserving an existing one (#1269)
- fix(infra): route the API's local-auth /api/auth endpoints to the API in Caddy (#1250)
Removed
- del(api): leave DELETE /ai/conversations/:id to the W3-6 purge service (#1315)
- del: collapse the Cursor rules to a pointer (#1279)
Full changelog: v1.11.0...v2.0.0