dev-collaboration

name	dev-collaboration
description	Role-based collaboration pattern (Drafter, Reviewer, Breaker). Assigns roles to available agents via prompt parameters. Use when starting multi-agent dev work.
metadata	{"owner":"interlateral","version":"0.1"}
compatibility	Requires at least two selected agents and access to comms.md ledger and agent notification scripts. Default to Claude Code + Codex; Gemini and Antigravity are opt-in only when explicitly selected by the human.

Dev Collaboration Pattern

Use mesh-comms-core first if direct peer notification has not already been proven. This skill defines roles and sequence; mesh-comms-core defines the transport.

Default roster: assign roles to Claude Code and Codex peers. Do not launch, recruit, or assign Gemini CLI, Antigravity CLI, or Antigravity Desktop unless the human prompt explicitly names that peer.

Prerequisites

• Notification Scripts: Use the repo's agent-specific scripts under interlateral_dna/ to send direct notifications.
  • Claude Code: node interlateral_dna/cc.js send "message"
  • Codex: node interlateral_dna/codex.js send "message"
  • Gemini, only when explicitly selected: node interlateral_dna/gemini.js send "message"
  • Antigravity CLI, only when explicitly selected: node interlateral_dna/agy.js send "message"
• Ledger: Use comms.md as the shared communications ledger. Posting to comms.md alone does not wake a peer.

Roles

Role	Job	Deliverable
DRAFTER	Creates or modifies the artifact	Working draft at the specified artifact_path
REVIEWER	Reviews for correctness, clarity, completeness	List of actionable suggestions
BREAKER	Adversarial testing; finds flaws, edge cases, failure modes	List of ways the artifact could break

Any explicitly selected v0.1 peer may hold any role. In routine use, choose Claude Code and Codex; treat Gemini and Antigravity as special opt-in peers.

Prompt Requirements

Every invocation of this skill MUST include:

1. Explicit role assignment -- which agent holds which role(s).
2. artifact_path -- the file path for the artifact being created or reviewed.

Dual-Hatting Rule

All three roles must be covered. If only two agents are available, the prompt MUST explicitly assign one agent a dual-hatted role. Example:

Codex is both REVIEWER and BREAKER.

The dual-hatted agent must complete each role as a separate phase and deliver separate outputs for each.

Sequence

1. DRAFTER creates artifact at artifact_path
2. DRAFTER posts to comms.md AND notifies REVIEWER and BREAKER via their scripts
3. REVIEWER and BREAKER work in parallel (do NOT read each other's outputs)
4. REVIEWER delivers suggestions
5. BREAKER delivers failure scenarios
6. DRAFTER incorporates feedback into a revised version
7. DRAFTER reports completion to human

Timeout Rule: If REVIEWER or BREAKER has not responded within 10 minutes of notification, DRAFTER may proceed with a partial revision but must note the missing input in the report.

Role Behaviors

If You Are DRAFTER

1. Create the artifact at artifact_path.
2. When done, perform the Ledger + Notify step:
   • Post to comms.md: [YOUR_AGENT] @REVIEWER @BREAKER -- Draft ready at [artifact_path].
   • Send direct notification to each agent using the matching script in interlateral_dna/:

     node interlateral_dna/cc.js send "[YOUR_AGENT] Draft ready at [ABSOLUTE_PATH]. Please review."
     

3. Wait for both roles to complete (or for the 10-minute timeout).
4. Read their feedback and produce a revised version.
5. Add a ## Change Log section to the artifact:

   ## Change Log
   - **Fixed:** [What was fixed] (Source: @[agent])
   - **Hardened:** [What was strengthened] (Source: @[agent])
   - **Declined:** [Suggestion] -- [Reason]
   

If You Are REVIEWER

1. Wait for DRAFTER's "ready" notification.
2. Read the artifact thoroughly.
3. Provide 3-5 actionable suggestions in this format:

   SUGGESTION 1: [Title]
   What: [Specific change]
   Why: [Benefit]
   

4. Deliver via direct notification to DRAFTER and let the helper mirror the entry to comms.md.
5. Notify DRAFTER when done.

If You Are BREAKER

1. Wait for DRAFTER's "ready" notification.
2. Read the artifact with an adversarial mindset. Assume future agents will be careless, rushed, and confused.
3. Provide 3-5 failure scenarios in this format:

   FAILURE 1: [Title]
   Attack: [How someone could misuse or break this]
   Consequence: [What goes wrong]
   Prevention: [How to guard against it]
   

4. Deliver via direct notification to DRAFTER and let the helper mirror the entry to comms.md.
5. Notify DRAFTER when done.

Completion Criteria

The pattern is COMPLETE when:

• DRAFTER created the initial artifact
• REVIEWER delivered at least 3 suggestions (or timed out)
• BREAKER delivered at least 3 failure scenarios (or timed out)
• DRAFTER produced a revised version incorporating feedback
• The artifact includes a correctly formatted Change Log
• Human notified of completion

Example Prompt

Run the dev-collaboration skill.
artifact_path: plans/integration_plan.md

Roles:
  DRAFTER: claude
  REVIEWER: codex
  BREAKER: codex

Deliver reviews via direct notification plus the `comms.md` ledger.

Two-agent example (dual-hatting):

Run the dev-collaboration skill.
artifact_path: specs/api_spec.md

Roles:
  DRAFTER: claude
  REVIEWER and BREAKER: codex

Codex: complete the REVIEWER phase first, then the BREAKER phase. Deliver separate outputs for each.