fix(schema): keep an unset column default unset instead of stringifying it - #15225

…out one (langflow-ai#16144)

fix(schema): keep an unset column default unset instead of stringifying it

Column.set_formatter_from_type ran str() on a None default, so a column with no default reached the frontend as the string "None" and the table editor prefilled new rows with it. Return early when the default is None. Declared defaults keep their current normalization.

Ported from langflow-ai#15225. Adds tests for an explicit formatter and for the TableInput template path.

Co-authored-by: Lengshuang <[email protected]>

)

* chore: bump release versions to 1.12.9

* fix: backport 1.13.1 fixes and image security updates to 1.12.9 (#16119)

* fix: backport 1.13.1 security and bug fixes to 1.12.9

* fix: repair CI for the 1.12.9 backport

* fix: settle discarded browser requests in backport CI

* fix: address backport review findings

* test: bind background service fixtures to their database

* fix: address backport feedback and CI compatibility

* fix: absorb lxml and Mako image workarounds for 1.12.9 (#16123)

* fix: build lxml with system libraries and require patched Mako

* fix: enforce and verify system-linked lxml builds

* chore: bump release versions to 1.13.2

* fix: preserve base image dependencies in extension checks

* ci: centralize Chromium runner defaults on Ubuntu 24.04 (#16155)

ci: pin Chromium workflow defaults to Ubuntu 24.04

* fix: authenticate starter project MCP clients in auto-login mode (#16151)

* fix: authenticate starter project MCP clients in auto-login mode

* fix: keep starter MCP credentials scoped to the owner

* fix: reuse and repair starter MCP owner credentials

* fix: preserve document identities during knowledge base ingestion (#16134)

* fix: preserve document identities during knowledge base ingestion

* [autofix.ci] apply automated fixes

* fix: guard ingestion lookups and document upgrade behavior

* docs: separate stable ingestion IDs from storage removal notes

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

* fix: preserve loop items and distinct outputs in list inputs (#16147)

* fix(loop): preserve other sources alongside loop items

* fix(graph): resolve each list entry through its source output

---------

Co-authored-by: Mart22052006 <[email protected]>

* fix: honor detected Windows-1254 text encoding (#16149)

* fix: enforce flow permissions for event access (#16150)

* fix: enforce flow permissions for event access

* test: cover shared flow event permissions

* fix: run knowledge base rollback off the event loop (#16148)

* fix: run knowledge base rollback off the event loop

* fix: preserve queued knowledge base rollback on cancellation

* fix: preserve numeric column types in Parser templates (#16146)

* fix: preserve numeric column types in Parser templates

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

* fix: retain Parser missing-value formatting

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

* fix: preserve Parser missing values without slicing every row

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

* fix: honor MCP flow visibility for project resources (#16127)

* fix: honor MCP flow visibility for project resources

* test: document MCP resource visibility cases

* ci: pin PR frontend tests to Ubuntu 24.04

* ci: align reusable and manual runner defaults

* test: document legacy MCP visibility and isolate CI runner changes

* fix: validate class pattern attributes (#16126)

* fix: validate class pattern attributes

* fix: require explicit attributes in class patterns

* ci: pin PR frontend tests to Ubuntu 24.04

* ci: align reusable and manual runner defaults

* fix: explain saved flow class pattern restrictions

* docs: explain saved flow class pattern migration

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix: enforce outbound request policy for OpenAPI agents (#16125)

* fix: enforce outbound request policy for OpenAPI agents

* test: ensure OpenAPI request protection runs in CI

* ci: pin PR frontend tests to Ubuntu 24.04

* ci: align reusable and manual runner defaults

* fix: bound OpenAPI request timeouts and isolate CI runner changes

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix: validate MCP shell wrapper arguments consistently (#16124)

* fix: validate MCP shell wrapper arguments consistently

* test: align MCP runtime rejection with shell policy

* fix: exclude keyword assignment mode from MCP shell wrappers

* ci: pin PR frontend tests to Ubuntu 24.04

* ci: align reusable and manual runner defaults

* fix: restrict MCP shell modes and explain inline command requirements

* docs: document MCP shell configuration migration

* test: align MCP shell fixtures with restricted modes

* fix: authenticate project MCP clients in auto-login mode (#16186)

* fix: allow first ingestion into a new OpenSearch index (#16171)

* fix: prevent Docker latest tag rollback

* fix: prevent Docker latest tag rollback

* fix: patch release 1.12.10 image vulnerabilities (#16201)

* fix: patch Brotli and image dependencies for 1.12.10

* fix: protect Brotli decoder precedence across interpreters

* fix(lfx): apply Split Text Text Key to lists of Data records (#16164)

* fix(lfx): apply Split Text's Text Key to lists of Data records

Messages keep their own text, and records keep their own key under the default Text Key.

* fix(lfx): keep each chunk as Split Text's output text

A leftover text field in the source record no longer replaces the chunk.

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

---------

Co-authored-by: Olayinka Adelakun <[email protected]>
Co-authored-by: Olayinka Adelakun <[email protected]>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

* fix: patch bundled npm glob and minimatch for 1.12.10 (#16205)

* chore(release): bump version to 1.12.10

* fix(deps): update FastMCP to 3.4.10

* fix: harden release 1.13.2 image dependencies (#16196)

* fix: harden release 1.13.2 image dependencies

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

* fix: enforce patched release dependency floors

* docs: explain release security guards

* fix: protect Brotli decoder precedence across interpreters

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

* fix(lfx): report input length and format in Batch Run metadata (#16152)

* fix(lfx): report the input length in Batch Run metadata

* fix(lfx): add input_format to Batch Run metadata and clarify tooltips

Tests cover TOML mode and failure paths; regenerates the component index and starter project.

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

* [autofix.ci] apply automated fixes

---------

Co-authored-by: Raphael Fakhri <[email protected]>
Co-authored-by: Olayinka Adelakun <[email protected]>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: Olayinka Adelakun <[email protected]>

* fix: preserve integer IDs in Parser Stringify tables (#16214)

* fix: use namespace-scoped Docker Hub tag API

* fix(schema): stop sending "None" as the default of table columns without one (#16144)

fix(schema): keep an unset column default unset instead of stringifying it

Column.set_formatter_from_type ran str() on a None default, so a column with no default reached the frontend as the string "None" and the table editor prefilled new rows with it. Return early when the default is None. Declared defaults keep their current normalization.

Ported from #15225. Adds tests for an explicit formatter and for the TableInput template path.

Co-authored-by: Lengshuang <[email protected]>

* fix: validate forwarded prefixes as URL paths (#16242)

* fix: scope frozen vertex results to the build caller (#16239)

* fix: scope frozen vertex results to the build caller

* fix: forward build callers through flow execution

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

* test(security): cover reflective namespace laundering (#16243)

* test(auth): cover trusted proxy credential sources (#16238)

* test(auth): cover trusted proxy credential sources

* test(auth): verify WebSocket query token forwarding

* fix(decisions): follow the error action for oversized inputs from every provider

An input over a decision model's token limit stopped the run even with
Evaluation Error Action set to Route to Else, because Langflow only
recognised HTTP 413 or a 400 whose error.code is max_tokens_exceeded or
context_length_exceeded. None of the three providers answers that way:

- OpenAI: 400, code null, "Decision input exceeds the token limit."
- TypeSafe: 400 {"detail": {"error_type": "max_tokens_exceeded"}}
- Perplexity: 400, code "400", "Decision question requires N tokens
  with context, exceeding maximum context length (M)"

TypeSafe's detail.error_type is now read as a code. OpenAI and Perplexity
name no code, so their own fixed sentence is matched, and only as the
whole message: a message that quotes the input can't pass for one.

The error reference also showed Perplexity's code when it only repeated
the status ("HTTP 400, 400"); it now names the provider's type instead
("HTTP 400, invalid_request").

Fixes LE-3117 F1 and F2.

* fix(decisions): say on the node that Decision Model needs a Context

A Decision Model with Context empty failed at run time with "Enter a
Context, or connect a Structured Context.", but nothing on the node said
Context was needed: Decision Model and Question carry the required
marker, Context did not.

Context can't be marked required: a connected Structured Context
replaces it, and the editor's required-field check can't see that
connection, so it would block valid flows. The field now says so where
the user looks instead: its placeholder reads "Required, or Structured
Context", and its tooltip ends with "Required unless a Structured
Context is connected." The tooltip keeps its purpose first because it
is also the argument description an Agent reads in Tool Mode.

Fixes LE-3117 F3.

* fix(decisions): stop using a decision model switched off in Model Providers

Switching a decision model off only hid it from the picker. A field that
already named it kept it, marked Switched off, and the flow still sent
its messages to that provider. Templates opened with a switched-off
jev-latest selected.

A decision model sends the flow's messages to its provider, so:

- The editor clears a saved decision model that is switched off: its
  provider is set up and lists the model, but the field no longer offers
  it. It never picks another model in its place, so messages never go to
  a provider the user didn't choose. A model whose provider has no key
  yet, or that policy no longer offers, is kept and explained as before.
- A run refuses a model the user switched off: "gpt-6-luna is switched
  off in Model Providers. Turn it on there, or pick another Decision
  Model." Only an explicit switch-off counts, so a key sent with the
  request still works without setup, and standalone lfx, which keeps no
  settings, is unchanged.
- An empty field now asks "Pick a Decision Model." instead of "Pick one".

Fixes LE-3117 F4.

* fix(i18n): translate the Smart Router and Decision Model fully

With the interface in any language but English, the Smart Router showed
its name, Input, Routes and Include Else Output translated, and its
description, Routing Mode, Decision Model, Decision Type, Decision
Question, Minimum Probability, Evaluation Error Action and every
decision-mode field in English. The Decision Model had no translations
at all.

- Adds the 55 missing Smart Router and Decision Model strings to de,
  es, fr, ja, pt and zh-Hans, and 63 to ko, until the Globalization
  Pipeline sync supplies them.
- Relabels Routes, Include Else Output and Additional Instructions where
  the old label disagreed with the field names the new tooltips use
  (e.g. ja 経路 -> ルート, fr Itinéraires -> Routes).
- "Open table" stayed English on every table field: each TableInput
  sends that English default, which overrode the translated label. The
  default now gives way to the translation; a component's own text is
  kept. The translations themselves read as a restaurant table in six
  languages ("Tisch reservieren", "Mesa libre") and now say Open table.

Dropdown option values (Decision Model, Choice, Stop with error) are
stored values, not labels, and stay as they are.

Fixes LE-3117 F5.

* fix(frontend): create one flow on a repeated New Flow click (#16232)

* fix(agents): keep parallel tool results on their own run (#16233)

* fix(api): recover milestones a slow event stream dropped (#16234)

* fix(components): send If-Else message on default route (#16275)

At Max Iterations the router switches to the default route, but the
output methods still picked their message from the original condition,
so the default branch ran with an empty message. Both outputs now resolve
the active route after the iteration helper runs.

* fix(decisions): cover dynamic translations and isolate switch tests

* fix: make model provider sections collapsible (LE-3358) (#16281)

fix: make model provider sections collapsible

* fix(security): reject frozen and ZIP loader imports (#16282)

* fix(security): reject frozen and ZIP loader imports

* test(security): keep stdlib boundary fixtures importable

* fix: apply connector URL policy before connecting to NextPlaid (#16283)

* fix: apply connector URL policy before connecting to NextPlaid

* fix: refuse unvalidated NextPlaid SDK redirects

* fix: require owner credentials for every webhook run (#16284)

* fix: require owner credentials for every webhook run

* docs: qualify the webhook authentication change by version

* fix(authz): refuse unavailable configured plugins (#16286)

* fix(serve): scope durable workflow operations to caller (#16287)

* fix(security): reject unsafe library execution sinks (#16289)

* fix(security): reject unsafe library execution sinks

* fix(security): reject mutable template sandbox constructors

* fix(security): block Jinja extension constructor re-export

* fix: enforce credential destinations during provider validation (#16290)

* fix: enforce credential destinations during provider validation

* fix: validate gateway destinations before OpenAI key checks

* fix: keep public API Request destinations fixed (#16291)

* fix: keep public API Request destinations fixed

* fix: reject public API Request tool destinations

* fix(deps): dependency upgrades for 1.13.2 (#16169)

* fix(deps): dependency upgrades for 1.13.2

* fix(deps): require adm-zip 0.6.1 in frontend

* fix(deps): require markdown 3.11.2 on Python 3.11+

* chore(deps): drop brotli constraints already in release-1.13.2

* fix(deps): remove ineffective markdown security bump

* fix(deps): require OAuthLib 4.0.1 for device-code validation

* fix(deps): upgrade AG2 to 1.1.1 in the optional extra

* fix(deps): use Tornado 6.6 on Python 3.11 and newer

* fix(models): clear rejected Bedrock sampling fields before retrying (#16298)

* fix: revoke refresh sessions on logout (#16293)

* fix(auth): revoke native refresh tokens on logout

* fix(auth): consume refresh tokens by canonical identity

* fix: coordinate browser refresh token exchanges

* fix: revoke login sessions without consuming refresh tokens

* docs: clarify legacy refresh session migration

* fix: require a fresh sign-in for legacy refresh tokens

* fix: prevent refresh from recreating expired login sessions

* fix: scope refresh sessions to active account owners

* test: cover logout revocation storage failures

* ci: collect flattened single-shard Playwright reports

* fix(tracing): isolate concurrent Phoenix workflows (#16294)

* fix(tracing): isolate Phoenix LangChain spans across concurrent workflows

LangChainInstrumentor is a process-wide singleton, so instrumenting it per workflow routed every concurrent workflow's spans to whichever provider registered first, and uninstrumenting on one workflow's teardown removed instrumentation while other workflows were still running. Build a per-workflow OpenInference LangChain callback bound to the tracer's own provider and hand it to the run through get_langchain_callback instead.

Fixes #15555

* test(tracing): assert duplicate Phoenix spans across projects

* test(tracing): reject unexpected Phoenix projects

* fix(tracing): preserve workflow callbacks across component execution

* [autofix.ci] apply automated fixes

* [autofix.ci] apply automated fixes (attempt 2/3)

* fix(packaging): version Phoenix callback bundle fixes

* fix(packaging): align Enterprise profile with bundle versions

* fix(tracing): repair isolated callback CI checks

* docs(tracing): address CodeRabbit review feedback

---------

Co-authored-by: yu1128-L <[email protected]>
Co-authored-by: TT <12365@566>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix(serialization): sanitize nonfinite NumPy and pandas floats (#16297)

* fix(serialization): sanitize nonfinite NumPy and pandas floats

* fix(serialization): sanitize NumPy object arrays

* fix(deps): use patched Zod in the docs toolchain (#16299)

fix(deps): update Puppeteer to use patched Zod in docs

* fix: recheck execution identity and tool approval at runtime (#16292)

* fix(security): recheck runtime tool authorization

* test: use native users in component persistence integration tests

* test: account for lifecycle reads in workflow query budgets

* fix: preserve local and synchronous tool execution

* fix: declare synchronous validation in the auth service protocol

* fix(security): require admin custom-code authoring by default (#16288)

* fix(security): require admin custom-code authoring by default

* fix: preserve projects under the default code policy

* test: scope trusted code fixtures under the administrator default

* test: update remaining trusted authoring and execution fixtures

* test: bind an administrator for the standalone REPL tracing test

The tracing callback test added in #16294 runs the Python REPL tool with no
bound principal. Under the administrator-only custom-code default that call is
rejected, so bind an explicit administrator for the test.

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix(lfx): keep non-string text values in Message (#16218)

Message.model_post_init treated every non-string text value as a stream,
so a number, dict or list was removed from data and parked in _text_stream.
message.text still showed the value, but data["text"], the JSON output,
to_lc_message() and Type Convert all saw an empty string.

Only iterators and async iterators are streams now. Other non-string values
are converted to text.

* fix(api): rename a flow that loses a concurrent create for its name (#16302)

Flow creation picks a free name with a read and only then inserts, so two
POST /api/v1/flows/ calls for the same name from one user both read "free"
and the loser hit the unique constraint, surfacing as 400 "Name must be
unique" for a name the caller never insisted on. The same window exists
for the endpoint name.

create_flow now starts the create over when the insert loses that race,
so the loser gets the next free suffix exactly as a sequential create
does. The retry rolls the transaction back rather than using a savepoint:
on SQLite a savepoint that opens the transaction commits on release, and
on PostgreSQL the failed insert dooms the transaction. A colliding
caller-supplied id is not retried, and callers that pass
fail_on_name_conflict / fail_on_endpoint_conflict are untouched.

Because the retry rolls back, MCP credentials carried by the flow are now
extracted once and staged inside each attempt, as update_flow and
upsert_flow already do, so a retried create cannot reference a variable
that was never stored.

* fix(graph): give list inputs the same value order in every process (#16285)

Graph._build_edges collected edges in a set. Edge hashes come from
strings, so the set's order followed PYTHONHASHSEED, and a list input
fed by several components got its values in a different order after a
restart or on another instance.

Collect the edges in a dict instead. It drops duplicates the same way
and keeps the flow's edge order.

Fixes #16110

* fix: stop Filter Values crashing on ordinary text filters (#16202)

Filter Values read `.text` on its Filter Key and Filter Value inputs. In a
graph run these are plain strings, so every run raised AttributeError,
including runs with the default values.

Read both inputs through a small helper: use `get_text()` for Data and
Message, map None to "", and convert other values to str. A connected
Data whose text is a number now matches too.

Regenerate the component index entry and add focused tests.

Fixes #15729

Co-authored-by: Eric Hare <[email protected]>

* fix(lfx): keep top-level JSON arrays in JSON Cleaner (#16206)

* fix(lfx): keep top-level JSON arrays in JSON Cleaner

JSON Cleaner always cut its input from the first "{" to the last "}".
A top-level array lost its brackets: '[{"a": 1}]' came back as a single
object, '["x", "y"]' raised "Missing '{' or '}'", and enabling Validate
JSON on '[{"a": 1}, {"b": 2}]' failed with "Extra data".

Take the outermost JSON value instead: use the array span when it
encloses the object span (or there is no object), otherwise the object
span as before.

* fix(lfx): skip bracketed labels before JSON arrays

---------

Co-authored-by: Eric Hare <[email protected]>

* fix: reset model provider scroll on selection (#16314)

* docs: clarify 1.13.2 upgrades and project authentication (#16318)

* docs: fix upgrades from 1.11.x full installations

* docs: explain project MCP and A2A authentication defaults

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

* fix(amazon): address Bedrock Converse QA failures (#16326)

* fix(amazon): address Bedrock Converse QA failures

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

* feat(frontend): let a deployment block the new-project and Deploy buttons with a reason (#16328)

* fix(settings): allow custom General sections and profile controls (#16324)

* fix: bump 1.14 bundles for merged release fixes

* fix: bump Amazon bundle for 1.13.2 Bedrock fixes

* feat: create connections inline and inspect trigger payload fields (#16332)

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* feat: enable tool mode for dedicated integration actions (#16315)

* feat: enable tool mode for dedicated integration actions

* feat: enable tool mode for Telegram actions

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

* fix: gate Valkey vector store on Python 3.15

* fix: align tool mode bundle release metadata

* test: synchronize Redis fixtures after maintenance admission

* Revert "test: synchronize Redis fixtures after maintenance admission"

This reverts commit f641273.

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix: enforce deployment variable readiness at runtime (#16334)

* fix: enforce deployment variable readiness at runtime

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

* fix: address deployment variable review findings

* [autofix.ci] apply automated fixes

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix(api): accept a new session_id on the advanced run endpoint (#16295)

* fix(api): accept a new session_id on the advanced run endpoint

* fix(api): skip the cached-graph policy check on a cache miss

* docs(api): describe session_id on the advanced run endpoint

* fix(frontend): mark sidebar triggers and drop the grip icon (#16278)

* fix(api): apply run tweaks after resuming a paused run (#16296)

* fix(api): apply run tweaks after resuming a paused run

* docs(api): scope the resume tweak comment to background runs

* fix(frontend): name playground chat attachments (#16277)

* fix(frontend): name playground chat attachments

* fix(frontend): name shared-playground files and delete buttons

* feat(triggers): enforce governance and report delivery health (#16333)

* feat(triggers): enforce governance and report delivery health

* fix: gate Valkey vector store on Python 3.15

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

* fix(triggers): defer governance imports during service bootstrap

* chore: auto-bake note keys and regenerate backend locales/en.json [skip ci]

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>

* fix: preserve Gunicorn compatibility cap in release installs

* fix: merge trigger governance and refresh session migrations

* fix: refresh Smart Router starter project code

* fix(ci): align release tests with current behavior

---------

Co-authored-by: Eric Hare <[email protected]>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: Mart22052006 <[email protected]>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: erichare <[email protected]>
Co-authored-by: olayinkaadelakun <[email protected]>
Co-authored-by: Olayinka Adelakun <[email protected]>
Co-authored-by: Olayinka Adelakun <[email protected]>
Co-authored-by: Raphael Fakhri <[email protected]>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Viktor Avelino <[email protected]>
Co-authored-by: Lengshuang <[email protected]>
Co-authored-by: Cristhian Zanforlin Lousa <[email protected]>
Co-authored-by: Adam-Aghili <[email protected]>
Co-authored-by: yu1128-L <[email protected]>
Co-authored-by: TT <12365@566>
Co-authored-by: Mohammad Hijjawi <[email protected]>