coo-agent-comms

name	coo-agent-comms
description	Sub-skill invoked by the COO. Reads agent-comms inbox results and outbox status, detects stale items, writes delegation files for code tasks, and returns structured Claude Code task data for the COO briefing.

COO Agent Comms — Cross-Agent Communication Sub-Skill

This skill is invoked by the COO during Step 0. It MUST NOT be run standalone. The COO passes control here; this skill reads agent-comms state, handles delegation writing, and returns structured data for the briefing.

---

Step 1: Read State

1. Read agent-comms/inbox/*.md — results from Claude Code work. For each file, note what was completed and which project it was in.
2. Read agent-comms/outbox/*.md — pending instructions not yet run. Check the status and priority fields.
3. Scan agent-comms/project-registry/*.md frontmatter — know which projects exist and their current status. Only read full files when you need project-specific context for a delegation.

---

Step 2: Alert Detection

Flag these as ALERT items (merged into COO alerts):

Condition	Alert
Inbox file older than 3 days unread by Cowork	ALERT: UNPROCESSED CLAUDE CODE RESULT
Urgent outbox file pending more than 2 days	ALERT: URGENT TASK NOT STARTED

---

Step 3: Delegation Writing

When the COO identifies a task requiring work inside a specific code project, this sub-skill writes the delegation file.

When to Delegate to Claude Code

• Plugin skill updates (mentilead-growthos)
• App code changes (mentilead-b2b-onboard, mentilead-orderflow)
• Website changes (mentilead-website)
• Infrastructure/dashboard work (mentilead-helm)
• Any task that needs access to a project's full codebase

File Format

Write to agent-comms/outbox/{YYYY-MM-DD}-{NN}-{slug}.md:

---
from: cowork-coo
to: claude-code
project: {folder name from project-registry}
priority: normal | urgent
created: {YYYY-MM-DD}
status: pending
---

Body Requirements

Body MUST include:

• Context — why this task matters
• Task — specific instructions
• Constraints — project gotchas from project-registry
• Expected output — result file path: ~/Documents/the-autonomy-experiment/agent-comms/inbox/{filename}-result.md

Project Context Lookup

Before writing a delegation, ALWAYS read agent-comms/project-registry/{project}.md to include the right gotchas and constraints.

Sequence Numbering

Check existing outbox files for today's date and increment the sequence number ({NN}).

---

Step 4: Return Format

Return structured data to the COO:

CLAUDE CODE TASKS block

CLAUDE CODE TASKS
  Completed (inbox):
  - [{project}] {what was done} -- {result summary}
  Pending (outbox):
  - [{project}] {task} -- {priority} -- "Run this in {project}: follow ~/Documents/the-autonomy-experiment/agent-comms/outbox/{filename}"
  New delegations:
  - [{project}] {task} -- {why}

Alert items

List of ALERT items with condition and description.

Completed items

List of inbox results processed (for COO COMPLETED section).

---

Step 5: Archive Management

After the COO logs inbox results, move processed inbox files to agent-comms/archive/ so they are not surfaced again in future briefings.

---

Logging Format

Append to the COO log under ### Claude Code Results:

### Claude Code Results
- [{project}] {what was done}: {result summary}