Security and stability release. Action required: the Fluent CLI and its MCP server now require a patched Node.js version.Breaking changes:
The CLI and MCP server now require a patched Node.js release (22.23.4+, 24.22.1+ or 26.12.1+) and exit at startup on anything older, including Node 20 and the odd-numbered 21, 23 and 25 lines. Set `FLUENT_SKIP_NODE_VERSION_CHECK=1` to bypass the check at your own risk
Security:
Updated production dependencies to patched releases, including `@modelcontextprotocol/sdk`, `axios`, `undici`, `adm-zip`, `moment` and `follow-redirects`
Upgraded `zod` to 4.x to fix GHSA-mrhx-c7xc-m3w5
Improvements:
Added `--context global` to `setting get` and `setting list`. `--context all` now includes global settings, and unknown `--context` values are rejected instead of ignored
Added paging and filters to `event list`: `--to`, `--count`, `--start`, `--root-entity-ref`, `--entity-type`, `--entity-id`, `--entity-ref` and `--event-status`. `--size` is now a deprecated alias of `--count`
Added an `env(VAR)` template function so module config values, such as webhook authentication, can be sourced from CI/CD environment variables. Only variables matching `--env-prefix` (default `FLUENT_`) are resolved, and missing or disallowed references fail with a single aggregated error
Added `--skip-bootstrap` to `profile create` and a new `profile bootstrap` command, so granting roles and applying account settings is opt-in instead of a side effect of creating a profile
MCP `graphql_execute` responses are now scrubbed of personal data before they reach the AI assistant. Setting values are exempt, so manifests and security headers can still be read, edited and saved
Added `--strict` to `blueprint install`, and asset install failures are now reported per module instead of as "installed successfully"
Added x/y progress and an ETA while loading products during `module install`
Added a warning when a later workflow fragment replaces or deletes a ruleset from an earlier module, and `workflowlog describe` now lists ruleset names and merge strategies
Improvements:
`module install``--include`/`--exclude` file entries are now scoped per asset folder and matched exactly, including sub folders. A file level `--exclude` skips only that file instead of the whole asset folder
`module install` skips empty or malformed JSON assets and empty plugin JARs with a warning instead of crashing the run
Debug logging now redacts credentials in arguments, metadata, errors and messages
`setting get` now defaults `--context` to retailer, and `event list` warns that `--retailer` does not filter and `--size` is capped at 100
Bugfixes:
Entity upserts that the platform rejects are now surfaced in the end-of-run summary and `--strict` instead of being reported as success
The workflow log is no longer written when the workflow upload is rejected, and `workflowlog insert` no longer overwrites existing entries
Network updates now remove locations that were removed from the network definition, instead of only adding
Template replacement no longer drops falsy-but-valid values such as an empty string
`module install` no longer crashes when the workflow log setting has a null value
Fixed the MCP workflow tools failing with ENOENT on Windows
Added `fluent event send` command to submit a payload to the Fluent Event API, with an opt-in --sync flag for the synchronous endpoint
Added `--skip-if-installed` to `module install`, treating an already-installed module version as a successful no-op for idempotent CI pipelines
`module install` now validates --include/--exclude entries against the module's real assets, failing fast on an unmatched --include instead of silently installing nothing
`module install` no longer swallows asset install failures; added `--strict` to exit non-zero in CI/CD pipelines when an asset fails
Fixed `profile create` hard-exiting when a read-only user can't self-grant account roles/settings; it now warns and continues
Added `--format bruno` to `fluent profile export`, writing a native .bru vars block alongside the existing Postman export
Added `fluent profile import` to create a local profile offline from a Postman environment export produced by `profile export --format postman`
Incremental release focused on making the vector search and skills introduced in 2.2.0 more reliable in day-to-day use, plus a set of authentication and session fixes.
Vector search now lets you choose explicitly whether you're searching code assets or docs content, so results aren't diluted by the other corpus. It also auto-injects your configured modules when no module scope is supplied, meaning agents get relevant, project-aware results without having to be told which modules to look in.
A selection of new skills are included to help agents when building and reasoning about GraphQL, working with the CLI, modules, and profiles. Additionally, the new `fluent-setup` skill can be used to set up a new agentic harness or ensure your current one meets our recommended best-practice guidelines.Get the new and updated skills with `fluent ai install-skills` on the latest CLI version to pick these up.
The vector search MCP tool will allow your coding agents to look up rule, util, and component source code for reference, as well as all of the content from our docs platform. The library currently contains over 400 code assets and 4,000 docs articles.
The search is based on semantic similarity, so your prompts can be matched against content even if the specific verbiage is slightly different. This allows agents to make better decisions, like using an existing rule instead of creating a duplicate that needs to be tested and maintained, or to base a new component on an existing one to ensure compatibility.Note: Vector search is a labs project that falls under the beta terms and conditions. We'd love for you to give it a go and send any feedback to help guide the direction of the product.Also included are new skills for using the vector search, updating security headers, plus enhancements to existing skills around developer workflow. Be sure to update using `fluent ai install-skills` to get the most out of the new capabilities.
Introducing the first set of Fluent Agentic Coding skills!Also included: additional tools for summarizing and working with large workflow and manifest documents.New features:
Tool `ai install-skills` supporting installation of Fluent skills across several agentic coding platforms. Supports in-project, global, and custom installation.
Bundles SKILL.md files for Mystique (component-build, component-list, manifest-configure, manifest-load-and-save), Rubix (build, test, review, understand, util-core, util-dynamic, util-test), fluent-understand, and fluent-workflow-rubix.
Tool `workflow summarize` with colour-coded domain tree output, depth limiting, and line-break formatting, plus `workflow extract` for pulling rulesets out of a workflow into a partial file.
Tool `manifest summarize` (with optional --descendants component tree), `manifest extract`, and `manifest merge` commands with supporting utilities for reading, diffing, and patching Mystique manifests.