inbox — tm_inbox, rendered as two lists
tm_inbox is read-only and derives both lists from task.json itself, the same way tm_board
does. This skill calls it and prints the Output Template below; it makes no decisions.
Process
- No task named →
tm_inbox({}) — every waiting_human card and every auto-decided pin, across
every task this session can see. A task named (E-xxxxxxxx or a full task_id) →
tm_inbox({task_id}) — just that one.
- Render
cards first — these are actually parked, waiting on a person right now — then
decided underneath, clearly separated: a decided entry never blocks anything, it is a
record of what the engine chose instead of asking.
- Three kinds of card park, told apart by
stage/human_gate, and each wants a different
submit payload: a pinned author stage (stage is implement/draft/cases) wants the
work done; an ask card (stage: "ask") wants one decision per questions[] entry
(decisions: [{question, chose}]); a human_gate: true card wants an accept/reject
({accept: true|false, reason?, gaps?}) — say which kind a card is before naming what to do
with it.
- A card's
key is a TASK ticket (E-xxxxxxxx/Pn/subgoalId) for a package's own child-run
node, same as ever. A run/task-level judging node has no subgoal to key off (setgoal,
plan, critique, gate:goal, or — at the manager layer — shape/critique/accept/
integrate/gate/gate:goal), so its key's last segment is that node's own node_id
instead (e.g. E-xxxxxxxx/P1/critique, or E-xxxxxxxx/TASK/critique for the manager's own
graph, TASK standing in for "no package" rather than a real STORY id) — pass it to submit
exactly as given either way.
- Print only what the reply carries.
cards[].acceptance and briefing_path are the full
brief — name the path, don't paste its contents into the conversation.
- Empty
cards and empty decided is a normal answer ("nothing waiting"), not an error.
Output Template
## Waiting on a human
| key | title | who | since |
|---------------------|--------------------------|--------|-------|
| E-a1b2c3d4/P2/U1 | ask: which retry policy | shkim | 10:32 |
| E-a1b2c3d4/P3/U2 | implement: payment retry | — | 09:58 |
| E-a1b2c3d4/TASK/critique | human_gate: critique | — | 10:40 |
E-a1b2c3d4/P2/U1
acceptance:
- <criterion>
- <criterion>
briefing: <briefing_path>
## Auto-decided (not interactive)
| key | title | who | reason |
|---------------------|--------------------------|-----|--------------------------------------|
| E-a1b2c3d4/P4/U3 | assignee: human pin | — | run not interactive; dispatched to codex |
who is — when nobody has claimed the card yet (take claims it). Render one card's
acceptance/briefing_path block for each row in cards, not just the first. Omit either
section header entirely when its list is empty, rather than printing it empty.
What Claude Does / What You Do
| Claude |
You |
Calls tm_inbox with the optional task_id and renders both lists above — nothing more |
Name a task (E-xxxxxxxx) to filter to one EPIC, or nothing for every run this session can see |
Never opens a briefing_path itself; names it for the human to open |
Take a card's key with take; a decided entry can be objected to the same way (pins it to you, reroutes to waiting_human on its next attempt) |
Related Skills
take — claim a card's key from this list for yourself
submit — record your answer once you've done the work
board — every EPIC, or one EPIC's STORY kanban
ticket — one ticket's full detail, EPIC or STORY
1---2name: inbox3description: >-4 Use when the user wants to see what teams work is waiting on a human via tm_inbox — every5 waiting_human card, plus what the engine auto-decided instead of asking. Triggers: "내 할 일6 뭐 있어", "나 기다리는 카드 있어?", "what's waiting on me", "show my inbox".7scenarios:8 - "What teams work is waiting on a human right now, across every run"9 - "Show me the inbox for just this one EPIC"10 - "지금 사람 대기 중인 카드가 뭐가 있는지 보여줘"11 - "이 EPIC에서 자동으로 결정돼버린 카드도 같이 보여줘"12compatibility:13 required:14 - teams15related:16 - take17 - submit18 - board19 - ticket20metadata:21 source: https://skillmd.com/skills/newkayak12/inbox22---2324# inbox — tm_inbox, rendered as two lists2526`tm_inbox` is read-only and derives both lists from `task.json` itself, the same way `tm_board`27does. This skill calls it and prints the Output Template below; it makes no decisions.2829## Process30311. No task named → `tm_inbox({})` — every waiting_human card and every auto-decided pin, across32 every task this session can see. A task named (`E-xxxxxxxx` or a full `task_id`) →33 `tm_inbox({task_id})` — just that one.342. Render `cards` first — these are actually parked, waiting on a person right now — then35 `decided` underneath, clearly separated: a `decided` entry never blocks anything, it is a36 record of what the engine chose instead of asking.373. Three kinds of card park, told apart by `stage`/`human_gate`, and each wants a different38 `submit` payload: a pinned author stage (`stage` is `implement`/`draft`/`cases`) wants the39 work done; an `ask` card (`stage: "ask"`) wants one decision per `questions[]` entry40 (`decisions: [{question, chose}]`); a `human_gate: true` card wants an accept/reject41 (`{accept: true|false, reason?, gaps?}`) — say which kind a card is before naming what to do42 with it.434. A card's `key` is a TASK ticket (`E-xxxxxxxx/Pn/subgoalId`) for a package's own child-run44 node, same as ever. A run/task-level judging node has no subgoal to key off (`setgoal`,45 `plan`, `critique`, `gate:goal`, or — at the manager layer — `shape`/`critique`/`accept`/46 `integrate`/`gate`/`gate:goal`), so its key's last segment is that node's own `node_id`47 instead (e.g. `E-xxxxxxxx/P1/critique`, or `E-xxxxxxxx/TASK/critique` for the manager's own48 graph, `TASK` standing in for "no package" rather than a real STORY id) — pass it to `submit`49 exactly as given either way.505. Print only what the reply carries. `cards[].acceptance` and `briefing_path` are the full51 brief — name the path, don't paste its contents into the conversation.526. Empty `cards` and empty `decided` is a normal answer ("nothing waiting"), not an error.5354## Output Template5556```57## Waiting on a human5859| key | title | who | since |60|---------------------|--------------------------|--------|-------|61| E-a1b2c3d4/P2/U1 | ask: which retry policy | shkim | 10:32 |62| E-a1b2c3d4/P3/U2 | implement: payment retry | — | 09:58 |63| E-a1b2c3d4/TASK/critique | human_gate: critique | — | 10:40 |6465E-a1b2c3d4/P2/U166 acceptance:67 - <criterion>68 - <criterion>69 briefing: <briefing_path>7071## Auto-decided (not interactive)7273| key | title | who | reason |74|---------------------|--------------------------|-----|--------------------------------------|75| E-a1b2c3d4/P4/U3 | assignee: human pin | — | run not interactive; dispatched to codex |76```7778`who` is `—` when nobody has claimed the card yet (`take` claims it). Render one card's79`acceptance`/`briefing_path` block for each row in `cards`, not just the first. Omit either80section header entirely when its list is empty, rather than printing it empty.8182## What Claude Does / What You Do8384| Claude | You |85|--------|-----|86| Calls `tm_inbox` with the optional `task_id` and renders both lists above — nothing more | Name a task (`E-xxxxxxxx`) to filter to one EPIC, or nothing for every run this session can see |87| Never opens a `briefing_path` itself; names it for the human to open | Take a card's `key` with `take`; a `decided` entry can be objected to the same way (pins it to you, reroutes to waiting_human on its next attempt) |8889## Related Skills9091- `take` — claim a card's key from this list for yourself92- `submit` — record your answer once you've done the work93- `board` — every EPIC, or one EPIC's STORY kanban94- `ticket` — one ticket's full detail, EPIC or STORY