Stripe pack: resolve a customer case
The resolve_customer_case workflow: the case solved in the help desk and the customer told within a day; every check, its parameters and what each failure looks like.
Last changed Markdown versionllms.txt
One of the five workflows of the Stripe Customer Operations pack, version 2.3.0.
Contract scop_resolve_customer_case, action type resolve_customer_case, settlement window 24h, on the Builder
plan and above. This page is generated from the pack's own files (the manifest, the contract, the recorder mapping and
the fixture cases as the engine judges them), so it says what the installed pack does.
What the agent does
The agent closes a support case in the help desk and tells the customer. Money that moves to resolve a case is a separate action under its own contract.
In the pack's words: The case reads solved in the help desk and the customer was told, within a day; not solved twice, not escalated after; an approved resolution kind.
The subject
The case, by the subject the help desk's connector gives it: zendesk_ticket_{id}, intercom_conv_{id},
sf_case_{id} or hubspot_ticket_{id}.
The action record
What POST /v1/actions carries for this workflow, whether it comes from the recorder, an SDK or your own code. A
required parameter is one a check reads: an action that would pass without it is UNKNOWN on the policy lens with
param_missing:<name>. Amounts are in major units (40.0 for forty dollars).
| Parameter | Required | What reads it |
|---|---|---|
support_case_id |
no | recorded with the action; no check reads it |
resolution_kind |
yes | not one of information, refund, credit, cancellation, escalation fails the policy lens |
customer_reference |
no | recorded with the action; no check reads it |
The action of the pack's happy-path fixture, as a request body:
{
"action_id": "resolve:zendesk_ticket_901",
"contract_id": "scop_resolve_customer_case",
"subject_id": "zendesk_ticket_901",
"subject_kind": "ticket",
"params": {
"support_case_id": "zendesk_ticket_901",
"resolution_kind": "information",
"customer_reference": "cus_1"
}
}
Tool calls the recorder maps to it
| Tool key | Subject read from | Action id | Note |
|---|---|---|---|
zendesk.update_ticket |
zendesk_ticket_{arguments.ticket_id} |
resolve:zendesk_ticket_{arguments.ticket_id} |
served when the help-desk source is zendesk |
intercom.close_conversation |
intercom_conv_{arguments.conversation_id} |
resolve:intercom_conv_{arguments.conversation_id} |
served when the help-desk source is intercom |
salesforce.update_case |
sf_case_{arguments.case_id} |
resolve:sf_case_{arguments.case_id} |
served when the help-desk source is salesforce |
hubspot.update_ticket |
hubspot_ticket_{arguments.ticket_id} |
resolve:hubspot_ticket_{arguments.ticket_id} |
served when the help-desk source is hubspot |
The tool keys and argument names are this pack's convention, not a verified list of any vendor's MCP tools; rename
them to what your server exposes (recorder setup). Calls that render the same action
id are one action when they agree on every parameter both carry (the reason apart, which the rules list under
describes), and the first record stands.
A rule fires on every call of its tool: zendesk.update_ticket, salesforce.update_case and hubspot.update_ticket
are recorded as a resolution whenever they are called, whatever the update was. The published recorders (0.1.1 on
PyPI, 0.1.2 on npm) do not read the subject templates these rules use and forward the calls unrecorded.
What CommitLayer reads, and from which system
Every check of the contract, in the engine's own description of it, with the reason code it reports when it goes
against the action. within times are measured from the action's requested_at.
Limits on what was approved (policy lens)
| Check | Read from | When it goes wrong | Reason code |
|---|---|---|---|
| resolution_kind not one of information, refund, credit, cancellation, escalation | the action itself | The resolution kind is not one the agent may apply | param_not_allowed:resolution_kind |
What must be seen afterwards (outcome and policy lens)
| Check | Read from | Trust needed | When it goes wrong | Reason code |
|---|---|---|---|---|
| zendesk.status_change observed within 24h of the action where to=solved, from a source rated shared or better | zendesk |
shared | The case does not read solved in the help desk | expected_effect_missing:zendesk.status_change, effect_below_trust:zendesk.status_change |
| messaging.customer_notified observed within 24h of the action, from a source rated shared or better | messaging |
shared | No customer notification was recorded | expected_effect_missing:messaging.customer_notified, effect_below_trust:messaging.customer_notified |
What must not be seen (policy lens)
| Check | Read from | When it goes wrong | Reason code |
|---|---|---|---|
| more than 1 zendesk.status_change effect(s) observed within 24h | zendesk |
The case was solved more than once in the window: it was reopened and closed again | duplicate_effect:zendesk.status_change |
| zendesk.escalation observed within 24h of the action | zendesk |
The case was escalated after the agent marked it resolved | effect_observed:zendesk.escalation |
What undoes it later (outcome and policy lens)
| Check | Read from | When it goes wrong | Reason code |
|---|---|---|---|
| zendesk.reopen observed within 14d of the action, undoing it | zendesk |
The case was reopened | reversed:zendesk.reopen |
How the case-resolved check reads a help desk. It looks for a status_change whose to attribute is the resolved
status (solved by default; Intercom closes a conversation, so set closed for it), which is how the Zendesk,
Intercom, Salesforce and HubSpot connectors record a status change. A help desk you push from sends status_change
with the same from and to attributes. Version 2.0.0 read a status attribute that no connector writes, so a case
solved in a connected help desk read expected_effect_missing:<help desk>.status_change. See connector
requirements.
What makes it unverifiable
When zendesk has never reported in the workspace, both lenses are UNKNOWN with missing_fact_source: there is
nothing to judge the action against.
What settles it
Pending while the help desk has not synced past the day's window; until then the view shows a provisional verdict, which reads FAIL for an effect not seen yet. It settles as FAIL once the help desk is current and the case is not solved or the customer was not told.
A verdict is pending until then, and the pending view shows the provisional verdict of the evidence so far. A case
reopened within 14 days re-verifies the action and fails it.
Parameters and their defaults
Set in the install wizard, or in configuration on the pack's configure, dry-run and install routes as
{"resolve_customer_case": {"<key>": <value>}}. A value that makes a check weaker than the default (a higher limit, a
longer window) is listed as a warning before anything is saved.
| Key | What it sets | Default | Type | Places in the contract |
|---|---|---|---|---|
window |
How long the help desk has to show the case solved | 24h |
duration | 1 |
helpdesk_source |
The system of record for support cases | zendesk |
source | 5 |
case_resolved_status |
The helpdesk status that means the case is resolved | solved |
string | 2 |
What each failure looks like
The pack's 11 fixture cases for this workflow, with the verdict the engine returns for each under the default parameters (a case written for one combination of systems under that configuration): the outcome lens (did it happen), the policy lens (was it allowed and done as approved), the status, and the reason codes of both lenses. These are the cases the dry run replays in your workspace. What each code means and what to do about it is on the failure interpretation page.
| Case | What is staged | Outcome | Policy | Status | Reason codes |
|---|---|---|---|---|---|
happy_path |
The case reads solved and the customer was told | PASS | PASS | settled | effect_observed:zendesk.status_change, effect_observed:messaging.customer_notified |
closed_without_telling_the_customer |
The agent marked the case solved and sent nothing | FAIL | FAIL | settled | verified_condition_not_met, expected_effect_missing:messaging.customer_notified |
told_but_never_closed |
The customer got an answer; the case stayed open | FAIL | FAIL | settled | verified_condition_not_met, expected_effect_missing:zendesk.status_change |
solved_twice |
Solved, reopened by the customer, solved again inside the window | FAIL | FAIL | settled | reversed:zendesk.reopen, duplicate_effect:zendesk.status_change |
escalated_after_solving |
Marked solved, then escalated to a human team | PASS | FAIL | settled | effect_observed:zendesk.status_change, effect_observed:messaging.customer_notified, effect_observed:zendesk.escalation |
resolution_kind_not_allowed |
The agent resolved the case with a kind of outcome it may not apply | PASS | FAIL | settled | effect_observed:zendesk.status_change, effect_observed:messaging.customer_notified, param_not_allowed:resolution_kind |
only_the_agent_says_so |
The help-desk facts were pushed by the agent's own key, not read from the help desk | FAIL | FAIL | settled | verified_condition_not_met, effect_below_trust:zendesk.status_change |
retried_tool_call |
The update call timed out and was retried: one action, not two | PASS | PASS | settled | effect_observed:zendesk.status_change, effect_observed:messaging.customer_notified |
delayed_evidence |
The help desk has not synced past the day's window yet: nothing is asserted until it does | FAIL | FAIL | pending | verified_condition_not_met, expected_effect_missing:zendesk.status_change |
source_unavailable |
No help-desk connection at all | UNKNOWN | UNKNOWN | settled | missing_fact_source |
reopened_later |
The customer came back a week later | PASS | PASS | settled | effect_observed:zendesk.status_change, effect_observed:messaging.customer_notified |
| the same case after the later record arrives | FAIL | FAIL | settled | reversed:zendesk.reopen |