End-to-end tests through the full HTTP router + real database:
TestCommentTriggerOnComment (6 subtests):
- Top-level comment without mentions → triggers agent
- Top-level comment mentioning only others → suppresses trigger
- Top-level comment mentioning assignee → triggers agent
- Reply to agent thread without mentions → triggers agent
- Reply to member thread without mentions → suppresses trigger (Bohan's bug)
- Reply to member thread mentioning assignee → triggers agent
TestCommentTriggerOnAssignNoStatusGate:
- Assigning agent to in_progress issue → triggers (no todo restriction)
TestCommentTriggerOnMentionNoStatusGate:
- @mentioning agent on done issue → triggers (no status gate)
TestCommentTriggerCoalescing:
- Rapid-fire comments → only 1 task created (dedup)
on_assign: Remove the todo-only restriction. Assignment is an explicit
human action — if someone assigns an agent to a done/in_progress issue,
they want the agent triggered (e.g. to fix a problem found after close).
on_mention: Remove the done/cancelled check. @mention is an explicit
action and should work on any issue status. The agent can reopen the
issue if needed.
- Add thread-aware on_comment suppression: when a member replies in a
thread started by another member without @mentioning the assignee
agent, the on_comment trigger is now suppressed. This fixes the bug
where member-to-member conversations incorrectly triggered the
assigned agent.
- Add terminal status check to on_mention: enqueueMentionedAgentTasks
now skips done/cancelled issues, consistent with on_comment behavior.
- Write explicit default triggers on agent creation: new agents get
[on_assign, on_comment, on_mention] all enabled, instead of relying
on null/empty = all enabled. Existing agents with empty triggers
still work via backwards-compat fallback in agentHasTriggerEnabled.
- Consolidate trigger check logic into shared agentHasTriggerEnabled
helper, fixing inconsistency where empty [] was handled differently
by isAgentTriggerEnabled (returned false) vs isAgentMentionTriggerEnabled
(returned true).
- Add documentation comments explaining the intentional status gate
difference: on_assign fires only for todo (start new work), while
on_comment fires for any non-terminal status (conversational).
Rewrite README.md from a developer-focused quick start to a user-facing
project overview covering features, cloud vs self-host, CLI usage, and
architecture. Add SELF_HOSTING.md with complete deployment instructions
including configuration reference, database setup, reverse proxy examples
(Caddy/nginx), agent daemon setup, and upgrade procedures.
Same issue as inbox: without a key, React reuses the AgentDetail
instance when switching agents. This causes activeTab and SettingsTab
form values (name, description, visibility, maxTasks) to persist
from the previously selected agent.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the small one-line sonner toast with a richer card showing:
- Green checkmark with "Issue created" title
- Status icon + issue identifier + title on second line
- Clickable "View issue" link to navigate to the new issue
When a task is executed by a mentioned agent (not the assigned one),
the live card now resolves the agent name from activeTask.agent_id
instead of using the assignee-based agentName prop.
When a comment @mentions anyone but not the assignee agent, the
assignee's on_comment trigger is now suppressed. This prevents the
assignee agent from being re-triggered when users share results with
colleagues or ask other agents for help.
The rule: @mention is an intent signal — if you're talking to someone
else, the assignee agent should not respond.
When a user @mentions an agent in any issue's comment, the system now
enqueues a task for that agent. The agent reads the issue context and
replies to the triggering comment thread.
Changes:
- Add shared util.ParseMentions for mention parsing (used by both
comment handler and notification listeners)
- Add EnqueueTaskForMention to TaskService for explicit agent targeting
- Add on_mention trigger type support in agent trigger config
- Add HasPendingTaskForIssueAndAgent SQL query for per-agent dedup
- Add enqueueMentionedAgentTasks in CreateComment handler
Safety: prevents self-trigger (agent mentioning itself), dedup with
assignee on_comment trigger, terminal issue status check, and per-agent
pending task dedup.
Without a key, React reuses the IssueDetail component instance when
switching between inbox items. This causes stale internal state
(e.g. TaskRunHistory tasks) from the previous issue to persist,
showing execution history from a different issue.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Load root .env in next.config.ts so REMOTE_API_URL is available
- Default fallback remains localhost:8080 (no impact on existing setups)
- Add REMOTE_API_URL to .env.example with documentation
Comment body was using RichTextEditor in read-only mode which doesn't
support IssueMentionCard rendering. Switch to Markdown component so
mention://issue/ links render as rich cards with status + title.
Instead of regex-parsing markdown content to find attachment URLs
(fragile), the frontend now tracks uploaded attachment IDs and sends
them with the comment creation request. The backend links them by ID.
Frontend: upload returns attachment ID, comment/reply inputs collect
IDs during editing session, pass as attachment_ids on submit.
Backend: CreateComment accepts attachment_ids, links by ID+issue scope.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
After creating an issue, a success toast appears showing the issue
identifier (e.g. MUL-72) with a "View issue" action button that
navigates to the issue detail page. Similar to Linear's behavior.
- Add Delete/DeleteKeys/KeyFromURL methods to S3Storage
- DeleteAttachment handler now removes the S3 object after DB delete
- DeleteComment collects attachment URLs before CASCADE, then cleans S3
- DeleteIssue collects all attachment URLs (issue + comment level) before CASCADE, then cleans S3
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Tiptap Mention extension's createInlineMarkdownSpec serializes
mentions as shortcodes [@ id="..." label="..."] — the .extend()
renderMarkdown override may not reliably take effect.
Added a robust fallback: post-process the editor's markdown output
by replacing shortcodes with [@Label](mention://type/id) using the
Tiptap JSON document for type info. Also preprocess stored shortcodes
in the Markdown renderer for backward compatibility.
The API at multica-api.copilothub.ai sets CloudFront signed cookies
with Domain=.copilothub.ai, but fetch() defaults to credentials:
'same-origin'. Since the frontend (multica-app.copilothub.ai) and API
are cross-origin, the browser silently drops Set-Cookie headers without
credentials: 'include'.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix mention markdown serialization: use renderMarkdown (tiptap/markdown 3.x API)
instead of addStorage.markdown.serialize which was silently ignored
- Add IssueMentionCard component showing status icon + identifier + title
- Update Markdown renderer to use card for mention://issue/ links
The migration from tiptap-markdown to @tiptap/markdown (38e92040)
changed the getMarkdown API location. The old package stored it at
editor.storage.markdown.getMarkdown(), but @tiptap/markdown adds it
directly as editor.getMarkdown(). This caused getEditorMarkdown() to
always return "", preventing description (and any markdown content)
from being saved when creating or editing issues/comments.
Merge main to pick up 028_task_trigger_comment migration. Renumber
daemon_token migration to 029. Fix execenv tests that expected CLI hints
in issue_context.md after they were moved to CLAUDE.md.
Issue daemon auth tokens (mdt_) on pairing session claim, bound to
workspace_id + daemon_id with 1-year expiry. Add DaemonAuth middleware
that validates these tokens and falls back to JWT/PAT for backward
compatibility. Apply middleware to all daemon routes except pairing
endpoints.
* feat(cli): add `multica update` command
Detects whether multica was installed via Homebrew (by resolving the
binary symlink and checking if it lives under a Homebrew prefix).
- Brew installs: runs `brew upgrade multica` automatically.
- Non-brew installs: prints instructions for installing via brew or
downloading from GitHub releases.
- Checks latest version from the GitHub releases API and skips
the update if already up to date.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(cli): use fully-qualified tap name in brew upgrade
Use `brew upgrade multica-ai/tap/multica` instead of `brew upgrade multica`
to avoid any potential name collision with core formulae.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The migration from tiptap-markdown to @tiptap/markdown in 38e92040
broke comment creation. The old package stored getMarkdown() on
editor.storage.markdown, but the official @tiptap/markdown extension
adds it directly to the editor instance (editor.getMarkdown()).
This caused getEditorMarkdown() to always return "", making the
submit button permanently disabled and preventing any comments.
Also fix stale submitting ref in useIssueTimeline dependency array.
- Fire getMe() and listWorkspaces() in parallel instead of serially,
saving one network round-trip (~200ms on cloud)
- Render dashboard sidebar shell immediately once user is authenticated,
show loading indicator in content area while workspace hydrates
Closes MUL-41
Show an archive icon on hover for each inbox list item, allowing
users to archive a single message directly from the list without
needing to open the detail panel first.
The parent comment's header (avatar, name, time, context menu) is now
the collapsible trigger itself. Only the body, reactions, replies, and
reply input collapse — the header is always visible. This removes the
duplicate author info that appeared when expanded.
renderIssueContext() now includes a "Quick Start" section with the
`multica issue get` command so agents know how to fetch issue details.
Fixes the TestPrepareDirectoryMode and TestWriteContextFiles failures.
Each comment card now has a clickable header with a chevron toggle.
When collapsed, shows author, timestamp, and a content preview.
When expanded, shows the full comment body, replies, and reply input.
Wrap the replies section in a Collapsible component so users can
collapse/expand replies on a comment thread. The parent comment and
reply input remain always visible. A chevron trigger shows the reply
count (e.g. "3 replies") and rotates on open. Default state is expanded
to preserve existing behavior.
- URL param: ?id= → ?issue= (keyed by issue, not notification)
- Multiple notifications for same issue now share selection state
- Archive correctly clears selection when archived item's issue matches
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
