What this does
Finds new/untriaged PSA tickets, classifies each by priority from its
content, updates the record, moves it to In Progress, and posts a
one-line summary to chat. Can run as a one-off right now, or be installed
as an hourly Claude-managed scheduled routine.
Source
Generalizes the ticket-dispatcher agent (autotask plugin, Archetype
B — "the canonical archetype-B fit") plus the equivalent
service-desk-ops agents the catalog lists for halopsa and
connectwise-psa — the same three PSAs psa-tool-map curates tool
names for.
Prerequisites
| Requirement |
Notes |
| A connected PSA |
Autotask, HaloPSA, or ConnectWise PSA (curated — see psa-tool-map for tool names; another connected PSA works via guided discovery). Must be able to read and update tickets — priority and status. |
| A connected chat-delivery target |
Slack (shipped) — see chat-delivery-adapters. Teams is not yet available. |
| Scheduled routines access (only if installing as a routine) |
Via the /schedule capability; managed at claude.ai/code/routines. |
Detecting the customer's stack
- Check which curated PSA is connected. More than one → ask which to
use. None → tell the human a PSA connection is needed, don't guess.
- Fetch
skill://advanced-workflows/psa-tool-map/SKILL.md for that
vendor's exact tool names, status/priority discovery approach, and
gotchas.
- Fetch
skill://advanced-workflows/chat-delivery-adapters/SKILL.md for
the connected delivery target's write tool and limits.
Run it now
- Using the PSA's "list new/untriaged tickets" tool, find tickets in the
New (or equivalent untriaged) status.
- For each, get its detail (title, description, requester).
- Classify PRIORITY using the rubric below.
- Call the PSA's "update ticket" tool once per ticket: set priority AND
advance status to In Progress in the same call — the status change is
the idempotency guard, so a second run won't reprocess it.
- Post a one-line summary per ticket to the connected chat-delivery
target.
- If an update call fails, leave the ticket in its original status and
post a "needs a human" message instead — never skip silently.
Priority rubric:
- Critical: server/site down, multiple users affected, or a security
incident.
- High: a single user fully blocked from working.
- Medium: service degraded but the user can still work.
- Low: requests, questions, cosmetic issues.
- When torn between two, pick the lower-severity one.
Set up as a recurring routine
- Confirm the PSA connector works (a lightweight test/ping call from the
tool-map skill, if one exists for the connected vendor).
- Discover the tenant's status and priority IDs once, by name — per
the tool-map skill's guidance for the connected PSA. Do not bake
list_* discovery calls into the routine itself.
- Confirm the chat-delivery connector and destination channel.
- Create a Claude-managed scheduled routine named "Ticket Triage Agent":
- Schedule: hourly (
0 * * * *) — Claude-managed routines reject any
cadence faster than hourly.
- Attach the PSA connector with
permitted_tools populated with
exactly the tools this workflow needs (list, get-detail, update) —
an attached connector with an empty permitted_tools list runs with
no tools and silently does nothing.
- Attach the chat-delivery connector with its
permitted_tools.
- Routine prompt: the "Run it now" steps above, with the discovered
status/priority IDs baked in by name (never re-discovered per run).
- Create one test ticket, trigger a manual run, and verify: the ticket
moved to In Progress + reprioritized, one chat message posted, and a
second run does not reprocess it.
Known gotchas
- One-hour minimum cadence. Claude-managed routines reject any cron
faster than hourly.
permitted_tools must be populated per connector, or the routine
runs with no tools and no error.
- Routine tool calls prompt for permission the first time each new
tool is touched — pre-allow them or expect an approval prompt on first
run.
- Don't call list/discovery tools every run — enumerate once at build
time; per-run discovery calls are slow enough to hit the tool timeout.
How it works
Idempotency is structural
The agent only picks up tickets in the untriaged status, and triaging a
ticket moves it out of that status — the transition itself is the
"already handled" marker. No database or timestamp window is needed.
Failures escalate to a human
An unclassifiable ticket or a failed update call leaves the ticket
untouched and posts a "needs a human" message — never silently skipped,
never mis-prioritized without a person being told.
Extending it
This classifies priority only, deliberately — enumerating queue/category
lists per run is what triggers slow discovery calls. To add queue or
category routing, bake those lists into the routine prompt the same way
status/priority IDs are baked in.
1---2name: ticket-triage-agent3description: >4 Build (or run once) a ticket-triage agent that classifies new PSA5 tickets by priority, advances them to In Progress, and posts a summary6 to the connected chat tool. Works with any curated PSA — Autotask,7 HaloPSA, or ConnectWise PSA.8when_to_use: >-9 Use when a customer wants new support tickets automatically triaged and10 prioritized. Use when: ticket triage, ticket dispatch, priority11 classification, PSA automation.12metadata:13 source: https://skillmd.com/skills/wyre-ai/ticket-triage-agent14---1516## What this does1718Finds new/untriaged PSA tickets, classifies each by priority from its19content, updates the record, moves it to In Progress, and posts a20one-line summary to chat. Can run as a one-off right now, or be installed21as an hourly Claude-managed scheduled routine.2223## Source2425Generalizes the `ticket-dispatcher` agent (`autotask` plugin, Archetype26B — "the canonical archetype-B fit") plus the equivalent27`service-desk-ops` agents the catalog lists for `halopsa` and28`connectwise-psa` — the same three PSAs `psa-tool-map` curates tool29names for.3031## Prerequisites3233| Requirement | Notes |34|-------------|-------|35| **A connected PSA** | Autotask, HaloPSA, or ConnectWise PSA (curated — see `psa-tool-map` for tool names; another connected PSA works via guided discovery). Must be able to read and update tickets — priority and status. |36| **A connected chat-delivery target** | Slack (shipped) — see `chat-delivery-adapters`. Teams is not yet available. |37| **Scheduled routines access** (only if installing as a routine) | Via the `/schedule` capability; managed at claude.ai/code/routines. |3839## Detecting the customer's stack40411. Check which curated PSA is connected. More than one → ask which to42 use. None → tell the human a PSA connection is needed, don't guess.432. Fetch `skill://advanced-workflows/psa-tool-map/SKILL.md` for that44 vendor's exact tool names, status/priority discovery approach, and45 gotchas.463. Fetch `skill://advanced-workflows/chat-delivery-adapters/SKILL.md` for47 the connected delivery target's write tool and limits.4849## Run it now50511. Using the PSA's "list new/untriaged tickets" tool, find tickets in the52 New (or equivalent untriaged) status.532. For each, get its detail (title, description, requester).543. Classify PRIORITY using the rubric below.554. Call the PSA's "update ticket" tool once per ticket: set priority AND56 advance status to In Progress in the same call — the status change is57 the idempotency guard, so a second run won't reprocess it.585. Post a one-line summary per ticket to the connected chat-delivery59 target.606. If an update call fails, leave the ticket in its original status and61 post a "needs a human" message instead — never skip silently.6263Priority rubric:64- **Critical**: server/site down, multiple users affected, or a security65 incident.66- **High**: a single user fully blocked from working.67- **Medium**: service degraded but the user can still work.68- **Low**: requests, questions, cosmetic issues.69- When torn between two, pick the lower-severity one.7071## Set up as a recurring routine72731. Confirm the PSA connector works (a lightweight test/ping call from the74 tool-map skill, if one exists for the connected vendor).752. Discover the tenant's status and priority IDs **once**, by name — per76 the tool-map skill's guidance for the connected PSA. Do not bake77 `list_*` discovery calls into the routine itself.783. Confirm the chat-delivery connector and destination channel.794. Create a Claude-managed scheduled routine named "Ticket Triage Agent":80 - Schedule: hourly (`0 * * * *`) — Claude-managed routines reject any81 cadence faster than hourly.82 - Attach the PSA connector with `permitted_tools` populated with83 exactly the tools this workflow needs (list, get-detail, update) —84 an attached connector with an empty `permitted_tools` list runs with85 no tools and silently does nothing.86 - Attach the chat-delivery connector with its `permitted_tools`.87 - Routine prompt: the "Run it now" steps above, with the discovered88 status/priority IDs baked in by name (never re-discovered per run).895. Create one test ticket, trigger a manual run, and verify: the ticket90 moved to In Progress + reprioritized, one chat message posted, and a91 second run does not reprocess it.9293## Known gotchas9495- **One-hour minimum cadence.** Claude-managed routines reject any cron96 faster than hourly.97- **`permitted_tools` must be populated per connector**, or the routine98 runs with no tools and no error.99- **Routine tool calls prompt for permission** the first time each new100 tool is touched — pre-allow them or expect an approval prompt on first101 run.102- **Don't call list/discovery tools every run** — enumerate once at build103 time; per-run discovery calls are slow enough to hit the tool timeout.104105## How it works106107### Idempotency is structural108The agent only picks up tickets in the untriaged status, and triaging a109ticket moves it out of that status — the transition itself is the110"already handled" marker. No database or timestamp window is needed.111112### Failures escalate to a human113An unclassifiable ticket or a failed update call leaves the ticket114untouched and posts a "needs a human" message — never silently skipped,115never mis-prioritized without a person being told.116117## Extending it118119This classifies priority only, deliberately — enumerating queue/category120lists per run is what triggers slow discovery calls. To add queue or121category routing, bake those lists into the routine prompt the same way122status/priority IDs are baked in.