Skip to content

Diagnostics ​

Every error and warning Mermaid Flow can show for a workflow. An error means the workflow can't run as written. A warning points at something that runs but is likely a mistake.

IdSeverityApplies toIn one line
Placing nodes
TASK_OUTSIDE_FLOWerrortask, decisionA task or decision sits outside every flow.
CONNECTOR_NOT_ROOTerrorconnectorA connector sits inside a flow.
INPUT_NOT_ROOTerrorinputAn input sits inside a flow.
FLOW_EMPTYwarningflowA flow is empty, or holds only other flows.
ID_DUPLICATEerrorany nodeTwo nodes share an id.
SHAPE_UNSUPPORTEDerrorany nodeA node uses a shape that isn't available.
Writing tasks
INSTRUCTION_MISSINGerrortaskA task has no instruction.
TASK_MISSING_LABELerrortaskA task has an empty label and no instruction.
INSTRUCTION_PLACEHOLDERerrorflow, task, decision, arrowAn instruction still holds a placeholder such as <REPO>.
Referring to things
REF_UNRESOLVEDerrorarrow, dotted linkA link points at an id no node has.
REF_OUT_OF_SCOPEerrorarrowAn arrow reaches into a flow from outside (inputs excepted).
REF_FORWARDβ€”β€”No longer raised.
Connectors
CONNECTOR_UNBOUNDerrorconnectorThe connector has no link, except to other connectors.
CONNECTOR_ID_RESERVEDerrorconnectorA connector uses the reserved id system.
CONNECTOR_BINDING_DUPLICATEwarningdotted linkThe same task and connector are linked twice.
Arrows
EDGE_KIND_INVALIDerrorarrow, dotted linkA link joins two shapes it can't join.
EDGE_LABEL_UNSUPPORTEDerrorarrowAn arrow that isn't a branch has a label.
EDGE_INSTRUCTION_UNSUPPORTEDerrorarrowAn arrow that isn't a branch has an instruction.
EDGE_REDUNDANTwarningarrowThe same arrow is drawn twice.
EDGE_SHORTCUTerrorarrowAn arrow skips a step its source already reaches.
EDGE_UNSUPPORTEDerrorany linkA link uses an operator that isn't available.
PIPE_PREFER_CHAINEDβ€”β€”No longer raised.
Decisions
DECISION_TOO_FEW_BRANCHESwarningdecisionA decision has fewer than two branches.
DECISION_BRANCH_UNGUARDEDerrorbranchA branch has no label.
DECISION_BRANCH_AMBIGUOUSwarningbranchTwo branches share a label.
DECISION_BRANCH_SHARED_TARGETerrorbranchTwo branches go to the same node.
How the workflow moves
FLOW_CONFLICTING_TRANSITIONSerrorlast step of a flow, arrowA flow and its last step both say what comes next.
CYCLE_NO_EXITerrorloopA loop, or a loop inside it, takes no decision branch out.
CYCLE_TASK_EXITwarningarrowA task's arrow leaves a loop, so its target runs every lap.
UNREACHABLE_COMPONENTwarningany nodeNothing leads to a node.
Settings
METADATA_UNKNOWN_KEYwarningany node, branchA setting the shape doesn't take.
METADATA_INVALID_VALUEerrorany nodeA setting has a value it can't take.
METADATA_REQUIREDerrorinput, connectorA required setting is missing.
METADATA_KEY_CONFLICTerrorinputTwo settings that exclude each other are both set.

Placing nodes ​

TASK_OUTSIDE_FLOW ​

error Β· task, decision Β· active

Why. A task or decision is a step of a flow, so it has to sit inside one.

Code:
mermaid
agentflow TB
  summarise["Summarise"]@{ instruction: "Summarise the ticket in two lines." }
Ctrl + Enter|

Task "summarise" is outside every flow. Move it into a flow.

Fix. Wrap the task or decision in flow id … end.

Related. CONNECTOR_NOT_ROOT, INPUT_NOT_ROOT

CONNECTOR_NOT_ROOT ​

error Β· connector Β· active

Why. A connector is shared by the whole workflow, so it can't belong to one flow.

Code:
mermaid
agentflow TB
  flow triage["Issue triage"]
    connector github["GitHub"]@{ protocol: "mcp", url: "https://api.githubcopilot.com/mcp/" }
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read -.- github
  end
Ctrl + Enter|

Connector "github" is inside flow "triage", but connectors belong at the top level of the workflow. Move it out of the flow.

Fix. Move the connector line above or below the flow, outside every flow … end. Its dotted links can stay where they are.

Related. TASK_OUTSIDE_FLOW, INPUT_NOT_ROOT

INPUT_NOT_ROOT ​

error Β· input Β· active

Why. An input starts flows from outside, so it can't sit inside one.

Code:
mermaid
agentflow TB
  flow triage["Issue triage"]
    weekday[/"Weekdays at 09:00"/]@{ trigger: "schedule", cron: "0 9 * * 1-5" }
    read["Read issues"]@{ instruction: "Summarise every open issue." }
  end
Ctrl + Enter|

Input "weekday" is inside flow "triage", but inputs belong at the top level of the workflow. Move it out of the flow.

Fix. Move the input outside every flow, and point an arrow from it to the flow: weekday --> triage.

Related. CONNECTOR_NOT_ROOT, EDGE_KIND_INVALID

FLOW_EMPTY ​

warning Β· flow Β· active

Why. A flow that holds nothing, or only other flows, has nothing of its own to run. Any other shape directly inside it counts as something to run.

Code:
mermaid
agentflow TB
  flow drafts["Drafts"]
  end
Ctrl + Enter|

Flow "drafts" is empty. Add a task to it, or remove it.

A flow that only holds other flows is flagged too:

Code:
mermaid
agentflow TB
  flow support["Support"]
    flow triage["Triage"]
      read["Read ticket"]@{ instruction: "Read the ticket." }
    end
  end
Ctrl + Enter|

Flow "support" only holds other flows, so it has nothing of its own to run. Add a task to it.

Fix.

  1. Add a task to the flow.
  2. Remove the flow.

Related. TASK_OUTSIDE_FLOW

ID_DUPLICATE ​

error Β· any node Β· active

Why. Arrows and settings find a node by its id. Two nodes with one id make every reference to it unclear.

Code:
mermaid
agentflow TB
  flow blog["Blog post"]
    draft["Draft"]@{ instruction: "Write a 600-word post." }
    draft["Edit"]@{ instruction: "Tighten the post to 400 words." }
  end
Ctrl + Enter|

"draft" is the id of 2 nodes. Give each node its own id.

Fix. Rename one of the nodes, and update the arrows that should point at it.

A line such as draft@{ instruction: "…" } only adds settings to a node, so it never counts as a second node.

Related. REF_UNRESOLVED

SHAPE_UNSUPPORTED ​

error Β· any node Β· active

Why. The node's shape is one Mermaid draws, but the workflow can't run it. While a diagram has an unavailable shape or link, the other checks are skipped, so fix this first.

The message names the node and its shape:

text
Node "<id>" uses the "<shape>" shape, which isn't available. Use "task", "flow" or "decision".

Fix. Rewrite the node as one of the available shapes: a task, decision, input, connector or flow.

Related. EDGE_UNSUPPORTED

Writing tasks ​

INSTRUCTION_MISSING ​

error Β· task Β· active

Why. A label names a step, but it doesn't say what the step should do. A task without an instruction can't run.

Code:
mermaid
agentflow TB
  flow support["Support"]
    summarise["Summarise"]
  end
Ctrl + Enter|

Task "summarise" has no instruction, so the workflow can't run it. Add an instruction that says what the step should do.

Fix. Add an instruction: summarise["Summarise"]@{ instruction: "Summarise the ticket in two lines." }. An instruction of only spaces counts as missing.

Related. TASK_MISSING_LABEL, INSTRUCTION_PLACEHOLDER

TASK_MISSING_LABEL ​

error Β· task Β· active

Why. A task whose label is empty, and that has no instruction, has nothing to do.

Code:
mermaid
agentflow TB
  flow support["Support"]
    summarise[" "]
  end
Ctrl + Enter|

Task "summarise" has no label or instruction, so it has nothing to do. Add an instruction.

Fix. Add an instruction, and a label if you want one.

Related. INSTRUCTION_MISSING

INSTRUCTION_PLACEHOLDER ​

error Β· flow, task, decision, arrow Β· active

Why. A value in capitals inside angle brackets, such as <REPO>, is a placeholder nobody filled in. The step has no real instruction to run.

Code:
mermaid
agentflow TB
  flow triage["Issue triage"]
    open["Open issue"]@{ instruction: "Open an issue in <REPO> describing the bug." }
  end
Ctrl + Enter|

The instruction on task "open" still has the placeholder "<REPO>". Replace it with a real value.

Fix. Replace each placeholder with the real value. Lowercase tokens such as <repo> are not placeholders and are left alone.

Related. INSTRUCTION_MISSING

Referring to things ​

REF_UNRESOLVED ​

error Β· arrow, dotted link Β· active

Why. A link points at an id that no node has, usually a typo. The workflow can't follow it.

Code:
mermaid
agentflow TB
  flow blog["Blog post"]
    draft["Draft"]@{ instruction: "Write a 600-word post." }
    edit["Edit"]@{ instruction: "Tighten the post to 400 words." }
    draft --> eddit
  end
Ctrl + Enter|

An arrow connects to "eddit", but there's no node with that id. Check the spelling, or add the node.

Fix.

  1. Fix the spelling of the id.
  2. Add the missing node.

Related. ID_DUPLICATE, REF_OUT_OF_SCOPE

REF_OUT_OF_SCOPE ​

error Β· arrow Β· active

Why. The nodes inside a flow belong to that flow. An arrow from outside the flow can reach the flow, but not the nodes inside it.

Code:
mermaid
agentflow TB
  flow team["Team"]
    plan["Plan"]@{ instruction: "Split the work into two parts." }
    flow writers["Writers"]
      write["Write"]@{ instruction: "Write the first part." }
    end
    plan --> write
  end
Ctrl + Enter|

"plan" can't connect to "write", because "write" is inside flow "writers". Point the arrow at flow "writers" instead.

Fix. Point the arrow at the flow the message names. An input's arrow is never flagged: an input can start any flow, even one inside another flow.

Related. REF_UNRESOLVED

REF_FORWARD ​

β€” Β· deprecated

No longer raised. You can use an id in an arrow before the line that defines the node.

Connectors ​

CONNECTOR_UNBOUND ​

error Β· connector Β· active

Why. A connector is only used through the tasks linked to it. With no link, nothing can use it. Any link counts here except one to another connector: a link from anything but a task is EDGE_KIND_INVALID instead.

Code:
mermaid
agentflow TB
  connector github["GitHub"]@{ protocol: "mcp", url: "https://api.githubcopilot.com/mcp/" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
  end
Ctrl + Enter|

Connector "github" isn't linked to any task, so nothing can use it. Link it to the task that needs it: <task> -.- github.

Fix.

  1. Link the connector to the task that needs it: read -.- github.
  2. Remove the connector.

Related. EDGE_KIND_INVALID, CONNECTOR_BINDING_DUPLICATE

CONNECTOR_ID_RESERVED ​

error Β· connector Β· active

Why. The id system belongs to the built-in tools. A connector with that id would clash with them.

Code:
mermaid
agentflow TB
  connector system["Tools"]@{ protocol: "mcp", url: "https://example.com/mcp" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read -.- system
  end
Ctrl + Enter|

The id "system" is reserved for built-in tools. Give this connector a different id.

Fix. Rename the connector, and update its dotted links. A task or flow can still use the id system.

Related. ID_DUPLICATE

CONNECTOR_BINDING_DUPLICATE ​

warning Β· dotted link Β· active

Why. A dotted link has no direction, so read -.- github and github -.- read are the same link. The second one adds nothing.

Code:
mermaid
agentflow TB
  connector github["GitHub"]@{ protocol: "mcp", url: "https://api.githubcopilot.com/mcp/" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read -.- github
    github -.- read
  end
Ctrl + Enter|

"read" is already linked to connector "github". Remove the extra dotted link.

Fix. Remove the extra dotted link. Different tasks can each link to the same connector.

Related. EDGE_REDUNDANT, CONNECTOR_UNBOUND

Arrows ​

EDGE_KIND_INVALID ​

error Β· arrow, dotted link Β· active

Why. Each link joins only certain shapes. An arrow joins flows, tasks and decisions, or an input to a flow. A dotted link joins a task and a connector. The message says which end is wrong.

Code:
mermaid
agentflow TB
  connector github["GitHub"]@{ protocol: "mcp", url: "https://api.githubcopilot.com/mcp/" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read --> github
  end
Ctrl + Enter|

Connector "github" can't be the target of an arrow. Link it with a dotted link instead: <task> -.- github.

Fix.

  1. Link a connector to a task with a dotted link: read -.- github.
  2. Link a connector to a task, not to a flow or a decision.
  3. Point an input's arrow at a flow, and don't point arrows into an input.

Related. CONNECTOR_UNBOUND, EDGE_UNSUPPORTED

EDGE_LABEL_UNSUPPORTED ​

error Β· arrow Β· active

Why. A label tells a decision which branch is which. On any other arrow, nothing reads it.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Summarise the ticket in two lines." }
    reply["Reply"]@{ instruction: "Write a polite reply that answers the ticket." }
    read -- then --> reply
  end
Ctrl + Enter|

The arrow from "read" to "reply" has a label, but only branches leaving a decision can have one. Remove the label, or add a decision.

Fix.

  1. Remove the label: read --> reply.
  2. Add a decision before the arrow, and make it a branch.

Related. EDGE_INSTRUCTION_UNSUPPORTED, DECISION_BRANCH_UNGUARDED

EDGE_INSTRUCTION_UNSUPPORTED ​

error Β· arrow Β· active

Why. An instruction on an arrow only has a place on a branch. On any other arrow, it isn't clear which step should follow it.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Summarise the ticket in two lines." }
    reply["Reply"]@{ instruction: "Write a polite reply that answers the ticket." }
    read e1@--> reply
    e1@{ instruction: "Keep the customer's name." }
  end
Ctrl + Enter|

The arrow from "read" to "reply" has an instruction, but only branches leaving a decision can have one. Move it into a task.

Fix. Move the guidance into the instruction of the next task, or add a task for it.

Related. EDGE_LABEL_UNSUPPORTED

EDGE_REDUNDANT ​

warning Β· arrow Β· active

Why. Drawing the same arrow twice adds nothing. It is usually left over from an edit.

Code:
mermaid
agentflow TB
  flow blog["Blog post"]
    draft["Draft"]@{ instruction: "Write a 600-word post." }
    edit["Edit"]@{ instruction: "Tighten the post to 400 words." }
    draft --> edit
    draft --> edit
  end
Ctrl + Enter|

There is already an arrow from "draft" to "edit". Remove the extra one.

Fix. Remove the extra arrow. a --> b and b --> a are a loop, not a repeat.

Related. DECISION_BRANCH_SHARED_TARGET, CONNECTOR_BINDING_DUPLICATE

EDGE_SHORTCUT ​

error Β· arrow Β· active

Why. A task continues along every arrow that leaves it. If it already reaches a step through another node, a direct arrow to that step hands it the same work a second time.

Code:
mermaid
agentflow TB
  flow blog["Blog post"]
    draft["Draft"]@{ instruction: "Write a 600-word post." }
    edit["Edit"]@{ instruction: "Tighten the post to 400 words." }
    publish["Publish"]@{ instruction: "Publish the edited post." }
    draft --> edit --> publish
    draft --> publish
  end
Ctrl + Enter|

The arrow from "draft" to "publish" skips a step: "draft" already reaches "publish" through "edit". Remove this arrow.

Fix. Remove the arrow that skips the step. Branches leaving a decision are never shortcuts.

Related. EDGE_REDUNDANT

EDGE_UNSUPPORTED ​

error Β· any link Β· active

Why. The link uses an operator Mermaid draws, but the workflow can't follow it. While a diagram has an unavailable shape or link, the other checks are skipped, so fix this first.

The message names the link by its two ends:

text
The arrow from "<a>" to "<b>" uses "<operator>", which isn't available. Use an arrow (-->) to connect steps, or a dotted link (-.-) to link a connector.

Fix. Use an arrow (-->) between steps, or a dotted link (-.-) between a task and a connector.

Related. SHAPE_UNSUPPORTED, EDGE_KIND_INVALID

PIPE_PREFER_CHAINED ​

β€” Β· deprecated

No longer raised. A chain such as a --> b --> c and one arrow per line mean the same.

Decisions ​

DECISION_TOO_FEW_BRANCHES ​

warning Β· decision Β· active

Why. A decision picks between branches. With fewer than two, there is nothing to choose.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Read the ticket." }
    kind{"Bug?"}
    fix["File a bug"]@{ instruction: "Open a bug report with steps to reproduce." }
    read --> kind
    kind -- yes --> fix
  end
Ctrl + Enter|

Decision "kind" has only 1 branch. Add another, because a decision needs at least 2.

A decision with no branches at all reads:

text
Decision "<id>" has no branches. Add at least 2.

Fix.

  1. Add another labelled branch.
  2. Replace the decision with a regular arrow (-->).

Related. DECISION_BRANCH_UNGUARDED

DECISION_BRANCH_UNGUARDED ​

error Β· branch Β· active

Why. A decision chooses a branch by its label. A branch with no label can't be chosen. An instruction on the branch doesn't replace the label.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Read the ticket." }
    kind{"Bug?"}
    fix["File a bug"]@{ instruction: "Open a bug report with steps to reproduce." }
    answer["Answer"]@{ instruction: "Reply using the help center." }
    read --> kind
    kind -- yes --> fix
    kind --> answer
  end
Ctrl + Enter|

The branch from "kind" to "answer" has no label, so the decision can't choose it. Add a label.

Fix. Add a label: kind -- no --> answer.

Related. DECISION_BRANCH_AMBIGUOUS, EDGE_LABEL_UNSUPPORTED

DECISION_BRANCH_AMBIGUOUS ​

warning Β· branch Β· active

Why. Two branches with the same label can't be told apart. Each one is flagged.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Read the ticket." }
    kind{"Bug?"}
    fix["File a bug"]@{ instruction: "Open a bug report with steps to reproduce." }
    answer["Answer"]@{ instruction: "Reply using the help center." }
    read --> kind
    kind -- yes --> fix
    kind -- yes --> answer
  end
Ctrl + Enter|

Decision "kind" has more than one branch labelled "yes". Give each branch its own label.

Fix. Give each branch its own label.

Related. DECISION_BRANCH_SHARED_TARGET

DECISION_BRANCH_SHARED_TARGET ​

error Β· branch Β· active

Why. Two branches that go to the same node lead to the same place, so the choice changes nothing. Each one is flagged.

Code:
mermaid
agentflow TB
  flow support["Support"]
    read["Read ticket"]@{ instruction: "Read the ticket." }
    kind{"Bug?"}
    fix["File a bug"]@{ instruction: "Open a bug report with steps to reproduce." }
    read --> kind
    kind -- yes --> fix
    kind -- maybe --> fix
  end
Ctrl + Enter|

Decision "kind" has more than one branch to "fix". Combine them into one branch.

Fix.

  1. Point one of them at a different node.
  2. Combine the branches into one. If that leaves the decision with one branch, it is DECISION_TOO_FEW_BRANCHES: replace the decision with a regular arrow (-->).

Related. DECISION_BRANCH_AMBIGUOUS, EDGE_REDUNDANT

How the workflow moves ​

FLOW_CONFLICTING_TRANSITIONS ​

error Β· last step of a flow, arrow Β· active

Why. When a flow finishes, either the flow says what comes next or its last step does. If both do, it isn't clear which arrow applies.

Code:
mermaid
agentflow TB
  flow research["Research"]
    find["Find sources"]@{ instruction: "Find three recent sources on the topic." }
  end

  flow write["Write"]
    draft["Draft"]@{ instruction: "Write a 600-word post from the sources." }
  end

  flow archive["Archive"]
    save["Save sources"]@{ instruction: "Save the source links to the archive." }
  end

  research --> write
  find --> archive
Ctrl + Enter|

Flow "research" continues to "write" when it finishes, but its last task "find" also has an arrow to "archive". Keep one of the two.

Fix.

  1. Remove the flow's arrow and keep the step's.
  2. Remove the step's arrow and keep the flow's.

Related. REF_OUT_OF_SCOPE

CYCLE_NO_EXIT ​

error Β· loop Β· active

Why. A loop ends only when a decision in it takes a branch that leads out. Without one, the loop runs forever. An arrow out of a task is not a way out, because a task continues along every arrow that leaves it. Neither is the arrow out of a flow whose last step is a decision: it is taken whatever the decision picks.

Code:
mermaid
agentflow TB
  flow editor["Editor"]
    draft["Draft"]@{ instruction: "Write a 300-word product description." }
    review["Review"]@{ instruction: "List what to improve in the draft." }
    draft --> review
    review --> draft
  end
Ctrl + Enter|

The loop through "draft" never ends, because it has no decision with a branch leading out. Add a decision with a branch that leaves the loop.

When the loop has a decision but none of its branches leads out, the message reads:

text
The loop through "<id>" never ends, because no decision in it has a branch leading out. Add a branch that leaves the loop.

A loop that can end can still hold a smaller loop that can't. Here check can leave the big loop, but draft and research go round each other without taking any of check's branches.

Code:
mermaid
agentflow TB
  flow editor["Editor"]
    draft["Draft"]@{ instruction: "Write a 300-word product description." }
    check{"Good enough?"}
    publish["Publish"]@{ instruction: "Publish the approved description." }
    research["Research"]@{ instruction: "Find one more fact for the draft." }
    draft --> check
    check -- no --> draft
    check -- yes --> publish
    draft --> research
    research --> draft
  end
Ctrl + Enter|

The loop through "draft" never ends, because it has no decision with a branch leading out. Add a decision with a branch that leaves the loop.

Fix. Make every loop take a decision branch, with one branch that goes back and one that leaves:

text
draft --> check
check -- no --> draft
check -- yes --> publish

Related. CYCLE_TASK_EXIT, DECISION_TOO_FEW_BRANCHES

CYCLE_TASK_EXIT ​

warning Β· arrow Β· active

Why. A task continues along every arrow that leaves it. When a task in a loop also points outside the loop, the target runs on every lap, not once at the end.

Code:
mermaid
agentflow TB
  flow editor["Editor"]
    draft["Draft"]@{ instruction: "Write a 300-word product description." }
    check{"Good enough?"}
    publish["Publish"]@{ instruction: "Publish the approved description." }
    notify["Notify"]@{ instruction: "Tell the team a new draft is ready." }
    draft --> check
    check -- no --> draft
    check -- yes --> publish
    draft --> notify
  end
Ctrl + Enter|

Task "draft" has an arrow leaving its loop to "notify", so "notify" runs on every lap. To run it once after the loop, start that arrow from the loop's decision.

Fix.

  1. Start the arrow from the loop's decision, as a labelled branch.
  2. Keep it, if the target should run on every lap.

Related. CYCLE_NO_EXIT

UNREACHABLE_COMPONENT ​

warning Β· any node Β· active

Why. A node that nothing leads to never runs. You won't see this warning on a diagram. Every step with no arrow into it counts as a starting point. So does the first step written in a loop that nothing leads into. The id stays in the list so that tools reading diagnostics know it.

The message has one of these forms, for a node in a flow or at the top level of the workflow:

text
Nothing in flow "<flow>" leads to "<id>", so it never runs. Add an arrow to it, or remove it.
text
Nothing in the workflow leads to "<id>", so it never runs. Add an arrow to it, or remove it.

Fix. Add an arrow to the node, or remove it.

Related. FLOW_EMPTY

Settings ​

METADATA_UNKNOWN_KEY ​

warning Β· any node, branch Β· active

Why. The shape doesn't take this setting, so it is ignored. It is usually a typo.

Code:
mermaid
agentflow TB
  flow support["Support"]
    summarise["Summarise"]@{ instruction: "Summarise the ticket in two lines.", promt: "Be brief." }
  end
Ctrl + Enter|

Task "summarise" has an unknown setting "promt". Check the spelling, or remove it.

Fix. Fix the spelling, or remove the setting. Each shape page lists the settings it takes.

Related. METADATA_INVALID_VALUE

METADATA_INVALID_VALUE ​

error Β· any node Β· active

Why. The setting is known, but its value can't be used: the wrong kind of value, a choice that isn't on the list, or text in the wrong format.

Code:
mermaid
agentflow TB
  connector github["GitHub"]@{ protocol: "http", url: "https://api.githubcopilot.com/mcp/" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read -.- github
  end
Ctrl + Enter|

The "protocol" setting on connector "github" can't be "http". Use "mcp".

The message names what is wrong. A wrong kind of value reads like this:

text
The "<setting>" setting on <shape> "<id>" should be text, not a number.

Fix. Use a value of the kind the message asks for. The shape pages list each setting's values.

Related. METADATA_UNKNOWN_KEY, METADATA_REQUIRED

METADATA_REQUIRED ​

error Β· input, connector Β· active

Why. Some settings have no default. A connector needs protocol and url, an input needs a trigger, and a schedule needs a cron or an at.

Code:
mermaid
agentflow TB
  connector github["GitHub"]@{ protocol: "mcp" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "List the open issues labelled bug." }
    read -.- github
  end
Ctrl + Enter|

Connector "github" is missing its "url" setting. Add one.

Fix. Add the setting the message names.

Related. METADATA_INVALID_VALUE, METADATA_KEY_CONFLICT

METADATA_KEY_CONFLICT ​

error Β· input Β· active

Why. A schedule runs either on repeat (cron) or once (at), not both.

Code:
mermaid
agentflow TB
  weekday[/"Weekdays at 09:00"/]@{ trigger: "schedule", cron: "0 9 * * 1-5", at: "2026-12-24T09:00" }

  flow triage["Issue triage"]
    read["Read issues"]@{ instruction: "Summarise every open issue." }
  end

  weekday --> triage
Ctrl + Enter|

Input "weekday" has both a "cron" and an "at" setting, but a schedule takes only one. Keep "cron" to repeat, or "at" to run once.

Fix.

  1. Keep cron to repeat.
  2. Keep at to run once.

Related. METADATA_REQUIRED

Opens in mermaid.ai