Skip to content

v1.3.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 16 Oct 17:08
· 27 commits to develop since this release
v1.3.0
3f13b93
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Added

  • Integrations: connect Bitbucket Cloud. A workspace admin installs the Aixle
    Flow app on the workspace and connects it from Workspace settings → Aixle
    Flow    ; agents then act as the app's bot with short-lived tokens, and
    Bitbucket reports CI as it finishes. A workspace, project or repository access
    token, or a personal API token, connects without the app. Jira and Bitbucket
    are now under one Atlassian entry in the Connect menu.
  • Repositories: attach Bitbucket repositories, and public repositories from
    bitbucket.org, and clone and push them in sessions. They stay attached when the
    repository or the workspace is renamed in Bitbucket.
  • Repositories: agents open, update, comment on, review and merge pull
    requests on GitHub, GitLab and Bitbucket with the same pull_request_* tools.
    A merge goes ahead only if the pull request is still at the commit the agent
    reviewed, a call repeated after a timeout never opens a second pull request or
    comment, and merging can be turned off per connection. A run started from a
    tracker issue puts the issue key in its branch names and pull-request titles,
    so Jira links them to the issue.
  • Triggers & gates: a bitbucket_pr_statuses_completed gate waits for every
    build status on a Bitbucket pull request's head commit — Pipelines or any
    other CI.
  • Integrations: operators who run their own Bitbucket app set
    BITBUCKET_FORGE_APP_ID and BITBUCKET_FORGE_ENVIRONMENT_ID, and
    BITBUCKET_FORGE_INSTALL_URL for the install button (see
    docs/reference/configuration.md and bitbucket-app/README.md). Without the
    first two Bitbucket connects with tokens only.
  • Tasks: a subtask's agent sees its parent epic's number, title and id, and
    board_list_tasks can list an epic's subtasks (parent_task_id).
  • Sign-in & onboarding: an admin can skip connecting an agent during
    onboarding ("Skip for now", then confirm) and still invite members and manage
    the company. Until they connect an agent in Profile settings, they cannot start
    workflows, sessions or the Builder.

Changed

  • Tasks: any project member who is not a viewer can set up a board and add,
    rename, reorder or delete its columns from the personal MCP server, and a board
    template installs its columns for them. Until now only the project owner could,
    although the board screen already let every member edit columns.
  • Sessions & Runs: a workflow run, session or Builder launch by someone with
    no agent login in the company is refused up front with "No agent credentials.
    Connect an agent in Profile settings to run workflows." It used to start and
    then end as cancelled with "Agent credential unavailable". A triggered run is
    skipped with the same reason, and a step whose agent is not connected fails
    with a message naming that agent.
  • Sessions & Runs: a workflow step's initial context summarizes only the
    steps it depends on, directly or through others, and only the latest completed
    or skipped attempt of each. Finished steps on independent branches no longer
    appear, and a step with no dependencies gets no previous-steps summary. Files
    from earlier steps are unaffected.
  • Sessions & Runs: when the generated context exceeds CONTEXT_TOKEN_BUDGET,
    every sub-step → data: line in the previous-steps summary is dropped, whatever
    its length (before, a long one was only shortened), and other notes over 100
    characters are cut. A session's context_metadata gains a rendering object
    with the final size, estimated tokens, compressed sections and whether it is
    still over budget.
  • Sessions & Runs: the Output Rules now say "Use only the MCP servers and
    tools needed for the task" instead of "Use all available MCP servers and
    tools". This is in the context of every session except the Aixle Builder's.
  • Aixle Builder: workflow guidance favors one canonical specification,
    resources scoped to each step and reviews with distinct purposes.

Fixed

  • Repositories: picking the chosen integration, repository or branch again in
    Add Repository keeps it instead of clearing it and the fields below. When the
    integration has nothing left to add, the repository field says why: every
    repository it reaches is already in the project, or it reaches none.
  • Sessions & Runs: a running step whose owner does not share active sessions
    now says its live view is private, and how the owner can share it, instead of
    showing "Session starting…" until the step ends.
  • Workflows: step, sub-step, workflow and task names that an agent wrote
    HTML-escaped ("Plan & release next batch") are stored and shown as plain
    text. Existing step, sub-step and task names are decoded by a migration.
  • Triggers & gates: a GitLab pipeline webhook no longer fails with a server
    error on the first delivery each web server receives, so a CI gate on that
    pipeline resolves when the pipeline ends instead of at the next reconcile sweep.
  • Sessions & Runs: an agent session or workflow step gets up to ten minutes to
    set itself up (skills, repositories, BMAD) before the agent starts, up from five.
    A slow package registry used to fail the step with "Unresolved exec; no
    reservation is held for this phase". The BMAD install now gives up after three
    minutes and the session goes on without it. If setup still runs out of time,
    the error says that an earlier attempt timed out or lost its worker.
  • Sign-in & onboarding: a viewer's onboarding shows a preview of one of the
    company's workflows and its first steps. It showed the generic text whenever
    that workflow had steps, and could pick an archived workflow.
  • Tasks: board_get_comments filtered by human author returns people's
    comments. The only human value it offered, user, matched nothing; it now
    accepts human (and system), and user still works as an alias.

Images

Published as ghcr.io/aixlehq/flow-<name>:1.3.0: flow-web, flow-otlp-ingest, flow-agent-base-core, flow-claude-code, flow-cursor-cli, flow-codex, flow-gemini-cli, flow-antigravity-cli, flow-grok, flow-kiro-cli. Web 1.3.0 launches the agent images of the same version.

What's Changed

  • fix: exec setup budget + cancellation, onboarding viewer preview N+1 by @artempartos in #497
  • fix(board): let every project writer set up the board and its columns by @artempartos in #501
  • feat(board): expose epic context and fix human comment retrieval by @sbarcelona11 in #496
  • feat(onboarding): let an admin skip the agent step, block launches without one by @artempartos in #502
  • feat(context): scope workflow history to dependencies and add Builder… by @sbarcelona11 in #495
  • fix(gitlab): stop params wrapping from building a gitlab gem client by @artempartos in #503
  • fix: say a step's live view is private; store agent-escaped names as plain text by @artempartos in #505
  • feat(bitbucket): add Bitbucket Cloud and pull-request tools on every code host by @artempartos in #504
  • fix(bitbucket): offer app mode without an install link and keep account ids briefly by @artempartos in #506
  • chore(release): 1.3.0 by @artempartos in #507

New Contributors

Full Changelog: v1.2.2...v1.3.0

Contributors

artempartos and sbarcelona11
Assets 2
Loading