Report problems from Slack

Mention @Volcano in a Slack thread that describes a problem, and LavaFlow reads the report, picks the project, and runs its usual diagnosis, fix, review and pull request. It answers in the thread, and answers again with the outcome. You can also ask it for a change, such as new content or functionality; see Request a change.

Slack reports need the project's lavaflow.slack capability. Reporting from Slack is rolling out environment by environment, so check it first. It is available where the project's lavaflow capability is and Slack reports have launched in the environment:

curl "https://api.volcano.dev/projects/$PROJECT_ID/capabilities" \
  -H "Authorization: Bearer $VOLCANO_TOKEN"

Set up

Set up Slack in the dashboard; the CLI does not manage it yet.

1. Install. As the project's owner, open LavaFlow settings and choose Add to Slack. Slack asks you to install the Volcano app in your workspace. The installation becomes a Slack connector, which everyone connected to that workspace can bind channels with. Only you, as the one who installed it, can remove it from Volcano.
2. Connect. Everyone who will report problems opens account settings and chooses Connect Slack, while signed in to the Volcano account that owns the projects. Slack confirms who they are with Sign in with Slack, and the connection is tied to the Volcano session they connected in. Several people can connect to one account, and one Slack account can connect to several Volcano accounts.
3. Add Volcano to a channel. In Slack, invite the app to each channel that will report problems, for example with /invite @Volcano.
4. Bind. In LavaFlow settings, the owner picks which channels report problems for which projects. The picker lists only channels the app and the Slack account you connected in this Volcano session are both in, and you can bind only such a channel, so on an account several people share, each picks from their own channels. After signing in to Volcano again, connect Slack again to pick channels. Archived, Slack Connect and shared channels can't report. Binding a channel the app isn't in answers 400: invite the app, then bind again. A channel reports for nothing until it is bound.

Report a problem

In a bound channel, reply to the message that describes the problem with @Volcano, or mention @Volcano in that message itself. LavaFlow picks the project:

• One bound project: it uses that project.
• Several, and your reply names one: the reply after the mention must be exactly one project's Volcano name, such as @Volcano checkout-api. Quotes, backticks and trailing punctuation are ignored. Anything else gets a question listing the channel's projects, each with a link to its repository.
• Several, and you only mention @Volcano: it uses the project whose name the original message contains. Otherwise LavaFlow's AI picks one of the channel's projects, or asks.

Only Volcano project names select a project, not repository names. When the original message names a bound project LavaFlow can't work on now, it asks instead of picking another. To answer a question, mention @Volcano again in the same thread with only the project's name. If you described the problem in replies rather than in the original message, the answer keeps them, up to your ten newest in the thread. If they don't all fit a report's limit of 8,200 characters, it keeps the newest that fit, and the report LavaFlow works from notes how many it left out.

Where problem clarification has launched, LavaFlow checks whether it has enough detail to investigate before starting diagnosis. If something essential is missing, it asks about it in the same thread with one of a fixed set of questions: what happened, where it happened, what you expected instead, how to reproduce it, or which of two conflicting replies is right. The question never quotes the thread or anything LavaFlow's AI wrote. Anyone in that thread can answer by mentioning @Volcano with the missing details, even if they have not connected their own Slack account to Volcano. The question says that a reply reaches LavaFlow's AI and how long Volcano keeps it, and links to Data and retention. LavaFlow assesses the replies for relevant evidence, asks again if needed, and continues the original report once it has enough information. Ordinary replies without a mention are not read.

Contributors do not choose another project or grant new permissions, and their replies don't change what the original reporter reported. LavaFlow lists the reporter's own replies apart from everyone else's, and reads the others as context: they can clarify facts, but they can't add to, narrow, or replace the reporter's report. The original reporter must remain connected and authorized, and the original channel binding must remain valid. The report keeps up to the ten newest clarification replies within its 8,200-character limit. Other people's replies are left out first; the original description is shortened only to make room for the reporter's own replies. Waiting for an answer does not start a run. Once diagnosis starts, later replies no longer change that report. A later mention in the thread starts a new report, which reads the thread again: for example, a request for a change after the run's result, or a retry after LavaFlow refused the report at a limit.

LavaFlow then answers in the thread with the project it picked and a link to its repository, and posts the outcome as a new reply. It never edits an earlier reply, so the thread keeps each step of the report. A run from Slack goes as far as opening a fix pull request, even where the project lets LavaFlow merge, and so does a run someone starts again on the report's issue: merging a fix someone reported in Slack stays a person's step. The outcome is the pull request's link, or a fixed reason it opened none, and a link to the run in Volcano. The run link opens the report's project in the same environment as the Slack app and still requires access to that project. It never holds the diagnosis: everyone in the channel reads the answer, including people who can't see the project's runs. Read the diagnosis in Volcano, and follow the run like any other; see Follow runs. Each report becomes an issue on the project, titled Problem reported in Slack, and counts toward the project's open issues until its fix pull request merges, which resolves it. A project at its limit of open issues is refused in the thread, and so is one that has used its LavaFlow runs for the day; that reply says when to mention Volcano again.

While LavaFlow is paused, it keeps a report for up to 10 days after it arrives, and picks it up if LavaFlow resumes in that time. A reporter without an authorized project can still be refused during a run-execution pause. During a publication maintenance pause, outgoing notices also wait, including pause and refusal notices; the pause does not use the report's retry allowance. Authorization and retention still apply. A reply to a question LavaFlow is waiting on still joins that report during a pause. A capped or unauthorized report ends without a channel notice. Its text is cleared unless it was already kept as a clarification; previously clarified text keeps its existing retention period.

Answers and fix pull request links are posted in the channel, where everyone in it can see them.

Request a change

The person who mentions @Volcano can ask for a change rather than report a defect, such as @Volcano add hello.md containing "hello world". Only what that person wrote can ask for one: the original message when they wrote it, the reply that mentions @Volcano, and their replies LavaFlow asked about. An original message someone else wrote is read only as a description of the problem, so mentioning @Volcano under another person's feature request doesn't ask for it; write the request in your own reply instead. Replies to a clarification question, the reporter's included, only describe the problem; to ask for a change after the run's result, mention @Volcano again.

Before any run starts, LavaFlow reads what you wrote, and only that, to tell what you're asking for:

• When your words name the change, such as a file or feature to add, the run builds it.
• When they describe a problem, ask LavaFlow to look at or fix one, or only point at the thread, such as a bare @Volcano or a project's name, the run diagnoses the thread like any report. If it finds nothing to fix, the reply tells you to describe the change you want in your own message.
• When they ask for new content or functionality without naming it, such as @Volcano do this under someone else's feature request, LavaFlow starts no run and asks you to say what to change in your own message.

LavaFlow builds what the request asks for and nothing beyond it, and its reviewers hold anything else the change adds. The request can't change what LavaFlow may do: its permissions, the files it may not change, its review, and when it opens a pull request stay the same, and the run stops at a pull request for a person to review and merge. The pull request says it proposes a requested change, not a fix, and the run's diagnosis.requested is true. When the request is unclear, the result already exists, or the change lies outside the repository, LavaFlow opens nothing and says so in the thread.

Who can report

A mention starts work only when all of these hold:

• Volcano's app is installed in the workspace.
• The channel is a regular channel in that workspace. Slack Connect channels and direct messages can't report problems; Volcano ignores mentions there.
• The person mentioning @Volcano has connected their Slack account to the Volcano account that owns the project.
• The project's owner bound the channel to the project.
• LavaFlow can work on the project: it is turned on, offered, and included in the owner's plan.

A person who hasn't connected gets a reply with a link to connect, except in a thread where a report is waiting for an answer to LavaFlow's clarification question: there their mention is read as an answer to it instead. A connected person in a channel bound to none of their projects gets a reply with a link to LavaFlow settings, where the owner binds it. Each person hears each refusal once a day in a thread, and a report limit's refusal once an hour: mentioning @Volcano again in the same thread for the same reason posts nothing new, while someone else, or a new reason, gets an answer.

Report limits

A clarification answer still counts against its contributor's hourly mention limit. It can join an existing report even when the workspace's waiting queue is full, because it does not create another waiting report. A report waiting for an answer doesn't count toward the workspace's waiting reports either.

A report over a limit starts no work and reads nothing from Slack. The runs reports start count toward the project's LavaFlow run limits like any other.

Each run checks these again as it goes. Disconnecting a Slack account, unbinding a channel, or removing the connector (uninstalling the app) takes effect at the next check, and cancels the report's queued work.

Enterprise Grid org-wide installs are not supported.

What LavaFlow reads

Text files reach LavaFlow's AI along with the message. Logs often hold credentials, tokens or personal data; remove them before you report, or leave the file out.

Other attachments, such as PDFs, videos, WebP images and binary files, are left out, and so is anything past these limits. The report still goes ahead, and the answer in the thread says how many attachments were left out.

Data and retention

Ten days after a report's run finishes, after LavaFlow answers without starting one, or 10 days after it arrived if it is still open, as when LavaFlow is paused, whichever comes first, Volcano deletes the report's text, its screenshots and files, and who reported it. A run still waiting for a deleted report stops. An issue still open then becomes ignored, so it stops counting toward the project's open issues without being counted as fixed. If its fix pull request merges later, the issue resolves. Deleting a message in Slack does not delete the report sooner. Deleting the project deletes its reports at once.

Replies that answer a clarification question are part of the report: they reach LavaFlow's AI with it, from whoever wrote them, and Volcano deletes them with it.

The run itself is kept like any other LavaFlow run, and its diagnosis may quote the report.

API

The dashboard uses these routes. Where Slack reports are not available in the environment, installing, connecting, the channel picker and binding a channel answer 404 with code feature_unavailable. Listing and removing connections, connectors and bindings work either way, so nothing you set up is stranded.

Starting an installation or connection, listing channels, and binding a channel accept the dashboard's current sign-in session. Browser requests include credentials so the callback can verify its cookie. Only the configured dashboard origin receives credentialed cross-origin access. A session from another Volcano project cannot authorize these operations.

Bound Slack channels keep a project's LavaFlow settings readable while LavaFlow is unavailable, even if the project has no saved settings or runs. The owner can still remove those bindings; this does not enable new reports.

Find the workspace's connector:

curl "https://api.volcano.dev/connectors" \
  -H "Authorization: Bearer $VOLCANO_TOKEN"

{
  "connectors": [
    {
      "id": "3e7a9c51-2b84-4f0d-9c6e-8d1f2a7b5e40",
      "type": "slack",
      "name": "Acme",
      "slack": { "team_id": "T0123ABCD" },
      "removable": true,
      "created_at": "2026-11-04T18:02:11Z"
    }
  ]
}

Bind a channel with it, as the project's owner:

curl -X POST "https://api.volcano.dev/lavaflow/connector-bindings" \
  -H "Authorization: Bearer $VOLCANO_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"project_id": "'"$PROJECT_ID"'", "connector_id": "'"$CONNECTOR_ID"'", "slack": {"channel_id": "C0456EFGH"}}'

{
  "id": "9b0d4e2a-3f1c-4b7e-8a6d-2c5f7e9a1b34",
  "project_id": "5c2e8f14-7a3b-4d9e-b1f6-0a8c3e5d7b92",
  "connector_type": "slack",
  "connector_id": "3e7a9c51-2b84-4f0d-9c6e-8d1f2a7b5e40",
  "slack": { "team_id": "T0123ABCD", "team_name": "Acme", "channel_id": "C0456EFGH" },
  "created_at": "2026-11-04T18:20:31Z"
}

A new binding answers 201. Binding again answers 200 with the same binding, so a retry is safe. A binding whose workspace no longer has the app installed has no connector_id, and reports again once the app is installed there again. Unbind with the binding's id:

curl -X DELETE "https://api.volcano.dev/lavaflow/connector-bindings/$BINDING_ID" \
  -H "Authorization: Bearer $VOLCANO_TOKEN"

Unbinding and reading the bindings work even while lavaflow.slack is unavailable. Any credential that can read the project reads its bindings.

Removing a connector uninstalls the app from the workspace for everyone who uses it. Only the person who installed it can; anyone else gets 403 and can uninstall the app in Slack instead. If Slack does not answer, the connector stays and the request answers 502, so try again. The workspace's bindings stay, and report again if the app is installed there again.

Agents

The MCP endpoint reads a project's bound channels with the get_lavaflow_connector_bindings tool. Binding a channel is the owner's decision, so no tool changes the bindings.