本文へ移動
cccskills
無料GitHub で公開

negotiation

Agents with competing priorities reach consensus through structured trade-offs.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md4.0 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Negotiation

Purpose

Multiple agents, each advocating for a different priority, negotiate to reach a consensus. No agent has authority; all must agree through trade-offs and compromise.

Use mesh-comms-core first if direct peer notification has not already been proven. comms.md is the ledger, not the wake-up path.

Default roster: use Claude Code and Codex peers. Do not include Gemini CLI, Antigravity CLI, or Antigravity Desktop unless the human prompt explicitly names that peer.

Roles

RoleDescription
ADVOCATEEach agent advocates for one priority (assigned in prompt)

All advocates are equal. Each agent gets one priority to champion.

Protocol

1. POSITION STATEMENTS: Each advocate declares their priority and rationale
2. NEGOTIATION: Agents propose trade-offs, discuss, counter-propose
3. CONSENT CHECK: Any agent may call for consent check
4. ITERATE: Continue until all agents signal [DONE] or [CONSENT]

Position Statement Format

[ADVOCATE - AGENT_NAME]
PRIORITY: [What I'm advocating for]
RATIONALE: [Why this matters]
RED LINE: [What I cannot compromise on]
WILLING TO TRADE: [What I can give up]

Trade-Off Format

[PROPOSAL - AGENT_NAME]
I OFFER: [What I'm giving up]
I WANT: [What I need in return]
RATIONALE: [Why this is fair]

Consent Check

Any agent may call: [CONSENT CHECK]

All agents must respond:

  • [CONSENT] - I agree to the current proposal
  • [OBJECT: reason] - I cannot agree because...

Inputs

All inputs come from the prompt.

Optional:

  • output: File path for consensus document
  • work: File containing the decision to be made
  • max_turns: Override default (30)

Prompt Format Examples

Minimal:

Use negotiation to decide on team meeting frequency.
CC=Productivity, CX=Work-Life-Balance, CC-peer-02=Collaboration.

With output:

Use negotiation for authentication protocol design.
CC advocates Reliability, CX advocates Speed, CC-peer-02 advocates Security.
Output: projects/auth/protocol.md

Safety

  • max_turns: 30 (default)
  • Deadlock: If max_turns reached, exits with [INCOMPLETE]
  • Status logged to: direct notification plus comms.md

Example Session

[ADVOCATE - CC]
PRIORITY: Reliability
RATIONALE: Users must always receive their messages
RED LINE: No silent message loss
WILLING TO TRADE: Latency up to 500ms acceptable

[ADVOCATE - CX]
PRIORITY: Speed
RATIONALE: Real-time feel is essential for UX
RED LINE: p99 latency must stay under 200ms
WILLING TO TRADE: Can accept eventual consistency for non-critical messages

[ADVOCATE - CC-peer-02]
PRIORITY: Security
RATIONALE: Messages must be authenticated and encrypted
RED LINE: No plaintext transmission
WILLING TO TRADE: Can use faster symmetric encryption after key exchange

[PROPOSAL - CC]
I OFFER: Accept 150ms latency target instead of "as fast as possible"
I WANT: Guaranteed delivery with at-least-once semantics
RATIONALE: Small latency increase enables retry mechanism

[PROPOSAL - CX]
I OFFER: Accept at-least-once (may require dedup on client)
I WANT: Fast path for messages under 1KB (skip some validation)
RATIONALE: 90% of messages are small, can optimize the common case

[CONSENT CHECK - CC-peer-02]
Current proposal:
- 150ms p99 target
- At-least-once delivery
- Fast path for small messages
- TLS required, symmetric encryption after handshake

[CONSENT - CC]
[CONSENT - CX]
[CONSENT - CC-peer-02]

[DONE] - Consensus reached.

Final Status Format

SKILL: negotiation
STATUS: DONE
TURNS: 8
OUTPUT: projects/auth/protocol.md
PARTICIPANTS: CC (Reliability), CX (Speed), CC-peer-02 (Security)
CONSENSUS: Achieved

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Method for adding agent comments and work to a shared file without overlapping other agents. Creates isolated workspaces per agent with UTC timestamps. Use when told to add comments, reviews, or work to any file.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

Check any artifact against an explicit specification and report adherence using the add-comments workspace format. Use when validating plans, scripts, docs, or code changes against a source-of-truth spec.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

Join the Antigravity CLI (agy, Gemini 3.5 Flash) to the Interlateral mesh as a native CLI peer with its own tmux session, identity stamping, and the agy.js send helper. Use this instead of the Antigravity desktop-app CDP path.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

Agents work in parallel on the same task; judges evaluate and select the winner.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

concierge

無料

Boot the explicitly selected CLI(s) as the human's concierge: a project-aware side channel that sits ABOVE the running agent system in a shared writable terminal. It answers status questions and handles the human's ad hoc tasking while the working agents collaborate on the mesh — it is not a worker, manager, or gatekeeper, and team progress never waits for it.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

Multiple agents draft a structured document through federated co-authorship and formal voting.

日本語の概要は準備中です。原文の説明を表示しています。

dazzaji/interlateral_agents112026年9月28日 更新

dazzaji のスキルをすべて見る

このスキルの問題を報告する