File formatBYOK

.tasks.md

A task list you can assign to AI. Write agent: @lead-qualifier under a checkbox and the agent does the work — and reports back in the same file.

revops.tasks.mdplain text · git-diffable
1## @tasks
2
3- [x] Automate weekly MRR digest
4 id: t-8f2 agent: @weekly-mrr
5 result: digest posted to #revops, 3 anomalies flagged
6
7- [~] Lead routing — ICP scoring live
8 id: t-9c1 agent: @lead-qualifier
9 mcp: [filesystem, salesforce]
10 priority: high due: 2026-01-18
11
12- [-] Decide routing schema
13 id: t-4e1 agent: @lead-qualifier
14 needs: score cutoff — 70 or 80?
15
16- [ ] Churn-risk playbook v2
17 id: t-5d7 priority: medium
revops.tasks.md — rendered
Automate weekly MRR digest@weekly-mrrJan 12
Lead routing — ICP scoring live@lead-qualifierJan 18
Decide routing schema@lead-qualifierJan 17
Churn-risk playbook v2youJan 19
CRM hygiene audit@crm-syncRun
5 tasks · 3 assigned to agents · 1 needs you — still just checkboxes in the file

How it feels

Watch one task go around.

You assign, the agent asks, you answer, it reports. Nothing leaves the file.

you

lead-qualifier

Score leads on ICP fit, route ≥80 to AEs

Running
MCP Servers2 active
Jan 8
you9:00 · assign it
revops.tasks.md
- [ ] Flag churn-risk accounts
id: t-11 agent: @lead-qualifier
mcp: [salesforce]
revops.tasks.md
@lead-qualifier
prompt · acp session → lead-qualifier
Task t-11 — Flag churn-risk accounts
Scan open accounts in salesforce —
flag no-login-90d, usage drop >50%,
or open tickets. Reply in the file.
tools: [salesforce]
@lead-qualifier9:05 · asks back
revops.tasks.md
- [-] Flag churn-risk accounts
needs: last 90 days or full year?
you9:41 · answer in place
revops.tasks.md
- [~] Flag churn-risk accounts
answer: last 90 days
@lead-qualifier9:58 · reports & closes
revops.tasks.md
- [x] Flag churn-risk accounts
result: 23 accounts flagged
note: csv in reports/churn.csv

Same file, four moments — one prompt in between. The conversation is the diff.

Who picks it up

The agents behind the names.

Bring your own agent — every card runs on your backend, your credentials. Arg0 never holds keys or proxies the call. agent: resolves by name, so the file never names a vendor.

lead-qualifier

Score inbound leads on ICP fit — route ≥80 to AEs, rest to nurture

Running
MCP Servers2 active
Steps14
Jan 8

weekly-mrr

Pull subscription deltas and draft the Monday MRR digest

Done
MCP Servers3 active
Jan 5

crm-sync

Reconcile HubSpot ↔ Salesforce contacts every night

Queued
MCP Servers2 active
Jan 2
New Agent

Register your own — any name, any backend. agent: @your-agent just works.

Assign to AI

The checkbox is who's turn it is.

Four states, one character each. They don't just track progress — they pass the work between you and the agent.

[ ]Anyone's

Todo

Open. If it has an agent: line, that agent can pick it up — or you press Run.

[~]Agent's turn

Working

The agent claimed it. Progress lands as note: lines so a fresh session can resume.

[-]Your turn

Needs you

The agent hit a question and wrote it in needs:. Answer in the file, flip it back to [~].

[x]Closed

Done

The agent wrote a result: summary. Review the diff — reopen with a single character.

you write [ ]→prompt → agent→agent asks [-]→you answer → [~]→agent reports [x]

Assign it to an agent. Keep the file.

Not an issue tracker, not a chat log — a Markdown file that humans and agents work from together.