| slug | cross-team-coordination |
| name | Cross-Team Coordination |
| description | Sequence handoffs when work spans multiple teams. |
| version | 0.1.0 |
| metadata | {"sources":[{"mode":"port-original","author":"Jannes Stubbemann","added_in":"0.2.0"}]} |
Cross-Team Coordination
Port-original skill. Hand-authored for this Agent Company; not from upstream anthropics/financial-services. Owned by the CEO role.
When to fire
After intake-triage flagged the request as cross-team (2+ agents from different teams). The CEO uses this to plan and sequence the handoffs so specialists don't block each other or duplicate work.
Inputs
- The triage decision from
intake-triage (primary agent, secondary agents, deadline, priority).
- The work request itself.
Outputs
A coordination plan, recorded on the work item:
sequence:
- step: 1
agent: <agent-slug>
skill: <skill>
inputs: <what they need to start>
outputs: <what they produce, by when>
- step: 2
agent: <agent-slug>
...
checkpoints:
- after_step: 1
review: <what the CEO checks before unblocking step 2>
final_handoff: <agent that owns the deliverable>
How to plan a multi-team workflow
- Identify the deliverable. Who owns the output that goes to the requester? That agent runs last.
- Walk back the dependency chain. What does the deliverable need? Each input becomes a step earlier in the sequence, owned by whichever agent produces it.
- Identify shared dependencies (the same comps run feeds the pitch and the model — run it once, fan out to both consumers).
- Set checkpoints at every team boundary. The CEO reviews artifacts before unblocking the next step. This catches "this isn't what I needed" before it propagates.
- Allocate buffer for the deadline. Cross-team work always takes longer than serial estimates suggest.
Common coordination patterns
| Workflow | Sequence |
|---|
| Full pitch | market-researcher (sector) → model-builder (DCF + comps) → pitch-agent (deck) → meeting-prep-agent (briefing) |
| Earnings update with re-rating | earnings-reviewer (note + model update) → market-researcher (peer impact) → pitch-agent if a fresh client take is needed |
| Quarterly close + statement | gl-reconciler (recon) → month-end-closer (accruals + variance) → valuation-reviewer (NAV) → statement-auditor (LP statement) |
| New LP onboarding + first capital call | kyc-screener (onboarding pack) → valuation-reviewer (capital call schedule) → statement-auditor (initial statement format) |
Boundaries
- The CEO does not write the artifacts — only sequences them.
- If a checkpoint reveals a problem (wrong inputs, missing data, off-thesis output), invoke
escalation-routing rather than papering over it.
- If the workflow doesn't match a known pattern, plan it explicitly rather than improvising at runtime.