Error codes
Mandates: systems and agents
Each code of this family: what it means, its most frequent cause, what to do and who can do it.
usecase_unresolved
- What it means
- This call was blocked: it is not tied to any mandate.
- Most frequent cause
- In this workspace, every AI call must belong to a declared and approved mandate.
- What to do
- Ask an administrator to bind this application to a mandate, or send the X-Azoth-Usecase header.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_not_approved
- What it means
- This call was blocked: its mandate is not approved yet.
- Most frequent cause
- The mandate this call belongs to is a draft, under review, or was rejected. Only approved mandates are served.
- What to do
- Another owner or administrator must approve the mandate.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_divergence
- What it means
- This call was blocked: it does not match its approved mandate.
- Most frequent cause
- The call uses a provider, data or people the mandate was not approved for, so its approval was withdrawn.
- What to do
- Ask the mandate’s accountable person to have it reviewed again.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_suspended
- What it means
- This call was blocked: its mandate is suspended.
- Most frequent cause
- The person who answered for this mandate is no longer a member of the workspace. A mandate nobody answers for is not served.
- What to do
- An owner or administrator names a new accountable person for the mandate.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_expired
- What it means
- This call was blocked: its mandate has ended.
- Most frequent cause
- The mandate this call belongs to reached its end date. An ended mandate is not served until it is renewed.
- What to do
- Its accountable person renews the mandate with a new end date.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_prohibited
- What it means
- This call was blocked: its mandate is classified as a prohibited practice.
- Most frequent cause
- The EU AI Act forbids this kind of AI use (article 5). A prohibited practice is never served, whatever the mandate’s status or the governance mode, and no exception can lift it.
- What to do
- An owner or administrator rejects the mandate, or corrects its classification if it is wrong — the change sends it back to review.
- Who can do it
- An owner or admin of the workspace · Open the screen
ai_act_class_required
- What it means
- A mandate is approved only with its EU AI Act risk class.
- Most frequent cause
- A system declared without a class, reviewed before the class was set.
- What to do
- A mandate is approved only with its EU AI Act risk class. Set it on the system first: the change sends it back to review.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_incomplete
- What it means
- A mandate is approved only once it names an accountable person and an end date.
- Most frequent cause
- A system declared before mandates existed, or one whose accountable person or end was never set.
- What to do
- A mandate is approved only with an accountable person who is a member, and an end date. Name both first.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandant_not_member
- What it means
- The person who answers for a mandate must be a member of the workspace.
- Most frequent cause
- The person was removed from the workspace, or has not accepted the invitation yet.
- What to do
- The accountable person must be a member of this workspace.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_end_in_past
- What it means
- The end date given for the mandate has already passed.
- Most frequent cause
- A date typed with the wrong year, or one carried over from an earlier mandate.
- What to do
- The end of a mandate must be in the future.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_end_too_far
- What it means
- A mandate lasts three years at most.
- Most frequent cause
- An end date more than three years away: a mandate is renewed explicitly, never set for good.
- What to do
- The end of a mandate is at most three years away: renew it explicitly.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_delegation_exceeded
- What it means
- The mandate of a sub-agent goes beyond the mandate of its parent agent.
- Most frequent cause
- A tool, a data scope, a cap or an end date that the parent’s latest mandate does not grant.
- What to do
- This mandate goes beyond the parent agent’s: a sub-agent never receives more than its parent.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_required
- What it means
- The request names no mandate.
- Most frequent cause
- Neither a system nor an agent was chosen.
- What to do
- Name one mandate: a system or an agent.
- Who can do it
- Any member of the workspace
mandate_not_found
- What it means
- The mandate named is not in this workspace.
- Most frequent cause
- A mandate deleted since, or an id from another workspace.
- What to do
- That mandate is not in this workspace.
- Who can do it
- Any member of the workspace
usecase_not_found
- What it means
- This system does not exist in this workspace.
- Most frequent cause
- A system deleted since, or an id from another workspace.
- What to do
- This mandate does not exist in this workspace.
- Who can do it
- Any member of the workspace
agent_not_found
- What it means
- No agent of this workspace has this id.
- Most frequent cause
- An agent deleted since, or an id from another workspace.
- What to do
- No agent with this id in the workspace.
- Who can do it
- Any member of the workspace
agent_archived
- What it means
- The agent is archived: it is read-only, and nothing runs for it — no execution, no tool call, no new mandate, key or red-team campaign.
- Most frequent cause
- An owner or admin archived the agent. Its identity, mandate versions, red-team campaigns and decisions are kept as evidence.
- What to do
- This agent is archived: it is read-only and runs nowhere. An owner or admin can unarchive it from its page; its mandate, keys and emergency stop then work again.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_archived
- What it means
- This call was blocked: its mandate is archived.
- Most frequent cause
- An owner or administrator archived this mandate. An archived mandate is kept with its history and its evidence, but none of its calls is served, whatever the governance mode.
- What to do
- Bind the key, agent or connector to another mandate. Or an owner or administrator unarchives it from its page: it comes back as a declaration to review again.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_not_archived
- What it means
- This mandate is not archived, so there is nothing to unarchive.
- Most frequent cause
- A second click, or the mandate unarchived from another screen.
- What to do
- This mandate is not archived: there is nothing to unarchive. Reload to see its state.
- Who can do it
- An owner or admin of the workspace · Open the screen
usecase_is_default
- What it means
- This mandate is the workspace’s default use case, so it is not archived.
- Most frequent cause
- Every call bound to no key, agent or connector falls back to the default: archived, it would refuse them all.
- What to do
- This mandate is the workspace’s default: every call bound to nothing falls back to it. Choose another default — or none — in the declarations below, then archive it.
- Who can do it
- An owner or admin of the workspace · Open the screen
redteam_campaign_running
- What it means
- A red-team campaign is still testing this mandate, so it is not archived yet.
- Most frequent cause
- A campaign queued or running against it: an archived mandate refuses every call, so the probes the campaign has left would all end in error. A campaign no longer holds the mandate once its own key has expired, a few hours after it started.
- What to do
- A red-team campaign is still testing this mandate: archived, it would refuse the probes the campaign has left to send. Wait until the campaign has finished, then archive the mandate.
- Who can do it
- An owner or admin of the workspace · Open the screen
human_reviewer_required
- What it means
- Approving or rejecting a system is a person’s act, signed in to the dashboard: an API key may declare a system, never review it.
- Most frequent cause
- Code sent the review of a system through the API, with an API key.
- What to do
- A system is reviewed by a person: an owner or admin other than the one who declared it approves or rejects it in Mandates › Systems and agents.
- Who can do it
- An owner or admin of the workspace · Open the screen
mcp_tool_changed
- What it means
- The MCP tool you reviewed is no longer the definition the catalogue holds, so your review was not applied.
- Most frequent cause
- A discovery of its server changed or removed the tool after you opened it; open it again and review the current definition.
- What to do
- This tool changed or was removed since you opened it: review it again.
- Who can do it
- An owner or admin of the workspace
identity_missing
- What it means
- An agent receives a mandate only once it has an identity: a person who answers for it.
- Most frequent cause
- No owner assigned to the agent yet.
- What to do
- Assign the agent’s owner on its page, then give it its mandate.
- Who can do it
- An owner or admin of the workspace · Open the screen
identity_revoked
- What it means
- This agent’s identity is revoked: it can no longer be changed.
- Most frequent cause
- Its owner left the workspace, or someone revoked it.
- What to do
- Create a new agent, with an identity of its own.
- Who can do it
- An owner or admin of the workspace · Open the screen
parent_not_found
- What it means
- The parent named is not another agent of this workspace.
- Most frequent cause
- The agent itself chosen as its parent, or an id from another workspace.
- What to do
- Choose another agent of this workspace as the parent.
- Who can do it
- An owner or admin of the workspace · Open the screen
parent_loop
- What it means
- This parent would make a loop of agents: an agent never delegates to one of its own ancestors.
- Most frequent cause
- The parent chosen already descends from this agent.
- What to do
- Choose a parent that is not below this agent.
- Who can do it
- An owner or admin of the workspace · Open the screen
mcp_tool_not_approved
- What it means
- The mandate names an MCP tool the catalogue has not approved.
- Most frequent cause
- A tool still waiting for its review, or removed from the catalogue.
- What to do
- Approve the tool in Mandates › MCP tools first, or leave it out of the mandate.
- Who can do it
- An owner or admin of the workspace · Open the screen
version_conflict
- What it means
- Another version of this mandate was created at the same moment, so this one was not saved.
- Most frequent cause
- Two people editing the mandate at once.
- What to do
- Reload the agent, check its latest version, then save again.
- Who can do it
- An owner or admin of the workspace · Open the screen
name_taken
- What it means
- An MCP server with this name is already in the catalogue.
- Most frequent cause
- The same server added twice.
- What to do
- Choose another name, or edit the server already there.
- Who can do it
- An owner or admin of the workspace · Open the screen
prohibited_practice_requires_exception
- What it means
- A system with a blocking flag is not approved as it stands: it is rejected, or approved with a documented exception.
- Most frequent cause
- A flag raised when the system was declared.
- What to do
- Reject the system, or approve it with a documented exception — its justification and its attachment.
- Who can do it
- An owner or admin of the workspace · Open the screen
exception_requires_justification
- What it means
- A documented exception needs a written justification of 200 characters at least.
- Most frequent cause
- A justification left empty, or too short.
- What to do
- Write why the exception is justified, in 200 characters at least, then approve again.
- Who can do it
- An owner or admin of the workspace · Open the screen
exception_requires_attachment
- What it means
- A documented exception needs the reference of the document that supports it.
- Most frequent cause
- The attachment reference left empty.
- What to do
- Give the reference of the supporting document, then approve again.
- Who can do it
- An owner or admin of the workspace · Open the screen
redteam_gate
- What it means
- A high-risk system is approved only after a recent red-team campaign that reached the workspace’s threshold, or under a waiver in force.
- Most frequent cause
- No recent full campaign, a category below the threshold, or a waiver that has ended.
- What to do
- Run a full red-team campaign on the system, or ask for a waiver, then approve again.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_models_empty
- What it means
- A mandate’s list of models names one model at least.
- Most frequent cause
- An empty list sent.
- What to do
- Choose the models the mandate may use, then save.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_models_exist
- What it means
- This mandate already has its list of models: it grows only from a comparison run that held.
- Most frequent cause
- The list created a second time.
- What to do
- To add a model, run its comparison replay, then allow the model from the replay’s result.
- Who can do it
- An owner or admin of the workspace · Open the screen
mandate_models_not_active
- What it means
- This mandate’s list of models is not in force, so nothing can be added to it.
- Most frequent cause
- The rule that holds the list was retired, or is not active.
- What to do
- Put the mandate’s model rule back in force in Rules, then try again.
- Who can do it
- An owner or admin of the workspace · Open the screen
replay_not_validated
- What it means
- A model is added to a mandate only from a comparison run that completed and passed.
- Most frequent cause
- A comparison still going, one that failed, or one not made for an alternative model.
- What to do
- Wait for the replay to complete and pass, then allow the model from its result.
- Who can do it
- An owner or admin of the workspace · Open the screen
unpriced_model
- What it means
- This model has no price, so its cost could not be capped: it is never called.
- Most frequent cause
- A model Azoth’s catalogue does not price yet.
- What to do
- Choose a priced model for the campaign.
- Who can do it
- An owner or admin of the workspace · Open the screen
no_category
- What it means
- None of the categories chosen applies to this target.
- Most frequent cause
- Categories meant for agents chosen for a system, or the reverse.
- What to do
- Choose categories that apply to the target.
- Who can do it
- An owner or admin of the workspace · Open the screen
nothing_to_replay
- What it means
- This campaign has no failed probe to run again.
- Most frequent cause
- Every probe of the campaign passed.
- What to do
- Run a full campaign instead.
- Who can do it
- An owner or admin of the workspace · Open the screen
source_running
- What it means
- The campaign whose failed probes were asked for again has not finished yet.
- Most frequent cause
- Its failed probes asked for again while the campaign still runs.
- What to do
- Wait until it has finished, then replay its failed probes.
- Who can do it
- An owner or admin of the workspace · Open the screen
library_changed
- What it means
- The probe library has changed since that campaign, so its probes cannot be run again as they were.
- Most frequent cause
- A new version of the probe library.
- What to do
- Run a full campaign with the current library.
- Who can do it
- An owner or admin of the workspace · Open the screen
already_imported
- What it means
- This file was already imported for this system.
- Most frequent cause
- The same file sent twice.
- What to do
- Import a new results file, or open the import already made.
- Who can do it
- An owner or admin of the workspace · Open the screen
not_json
- What it means
- The results file is not JSON.
- Most frequent cause
- Another file chosen, or a file cut short.
- What to do
- Export the results again as JSON, then import that file.
- Who can do it
- An owner or admin of the workspace · Open the screen
unrecognised_format
- What it means
- The file is neither a Giskard scan (to_json) nor the azoth.redteam.import/1 format.
- Most frequent cause
- A file from another tool, or exported another way.
- What to do
- Export the Giskard scan with to_json, or write the file in azoth.redteam.import/1, then import it.
- Who can do it
- An owner or admin of the workspace · Open the screen
invalid_results
- What it means
- The file’s results do not match their format: the answer’s error says where.
- Most frequent cause
- A field missing or of the wrong type in a result.
- What to do
- Correct the file where the error says, then import it again.
- Who can do it
- An owner or admin of the workspace · Open the screen
ran_in_future
- What it means
- The file says its run happened in the future.
- Most frequent cause
- A wrong ranAt date, or a clock set wrong where the run happened.
- What to do
- Correct the ranAt date in the file, then import it again.
- Who can do it
- An owner or admin of the workspace · Open the screen
agent_not_archived
- What it means
- This agent is not archived, so there is nothing to unarchive.
- Most frequent cause
- A second click, or the agent unarchived from another screen.
- What to do
- This agent is active already: there is nothing to unarchive. Reload to see its state.
- Who can do it
- An owner or admin of the workspace · Open the screen