Agent Skills
› mohitagw15856/pm-claude-skills
› collaboration-contract
collaboration-contract
GitHub用于启动跨团队协作项目，通过制定一页纸协作契约明确决策权、接口规范、沟通节奏及完成标准，预防团队间因规则不清导致的冲突。
Trigger Scenarios
启动跨团队合作项目
解决团队协作碰撞问题
定义跨团队工作流与沟通机制
在合作开始前建立伙伴关系
Install
npx skills add mohitagw15856/pm-claude-skills --skill collaboration-contract -g -y
SKILL.md
Frontmatter
{
"name": "collaboration-contract",
"description": "Start a cross-team project with the collaboration contract that prevents the classic collisions — who decides what, how work flows between teams, the communication channels and cadence, and what done means — agreed before the first collision instead of during it. Use when asked kick off this cross-team project right, our two teams keep colliding, define how we'll work with the other team, or set up the partnership before we start. Produces the one-page contract: decision rights, interfaces, cadence, and the done-definition."
}
Collaboration Contract Skill
Cross-team projects fail at the seams: each team runs on its own invisible rules (working-agreements are per-team; the seam has none), decision rights are assumed differently ("we thought design signed off; they thought we did"), and handoffs bounce because "done" meant different things. The collaboration contract is the seam's rulebook, one page, agreed at kickoff while everyone still likes each other: who decides what (the decision-rights table), how work crosses the boundary (the interface), how the teams talk (channel + cadence), and what done means per handoff — cheap to write on day one, expensive to reverse-engineer during the first fight.
What This Skill Produces
- The decision-rights table — the project's likely decisions × who decides / who's consulted, the ambiguous ones forced now
- The interface spec — how work moves between teams: the handoff format, the request path, the done-definition per handoff type
- The communication layer — the shared channel, the sync cadence (as light as the project allows), the escalation path with names
- The friction protocol — what happens at the first collision: the two-lead conversation before anything ascends
Required Inputs
Ask for these if not provided:
- The project and the teams — what's being built/done, which teams, their prior history (a scarred partnership needs the contract more and trusts it less — the tone adjusts)
- The likely decisions — the calls this project will force (scope, priority conflicts, quality bars, launch timing); the table pre-decides the deciders, and the awkward ones ("who wins when priorities conflict?") are exactly the ones to force now
- The handoff shapes — what actually crosses the seam (designs → build? data → analysis? approvals?); each shape gets its done-definition
- The leads — the two humans who own the seam; the contract is theirs to sign and theirs to enforce
Framework: The Contract Rules
- Force the ambiguous decisions now: the table lists the project's foreseeable calls with a decider each — and the ones both teams assume they own are the ones the kickoff must settle ("tie-breaks on priority: [name]"). Every ambiguity left standing is a scheduled fight with interest.
- Interfaces are formats plus done-definitions: each handoff type gets its shape ("designs arrive as [format] with states covered; 'done' = the checklist passes") — because bounced handoffs are almost always definition mismatches, not quality failures. The runbook-writer verify-per-step logic, applied to seams.
- Cadence as light as survivable: one shared channel, one short sync at the necessary frequency (not the reassuring one — meeting-cost-meter math doubles across two teams), and the asks flowing per async-update-format shapes between syncs. The contract names it all so nobody invents parallel channels in week two.
- The friction protocol de-escalates by design: first collision → the two leads talk within 48h, before either escalates upward ("no surprising each other's managers" is the clause that saves partnerships) → unresolved after a real attempt → the named tie-breaker. Written while calm, the protocol converts the first fight from a relationship event into a process event.
- The contract is one page and revisited once: kickoff-signed, then reviewed at the first milestone ("which clause did we violate? fix it or change it" — the working-agreements revisit logic) — a living page, not a treaty. Longer contracts don't get read; unrevisited ones drift into fiction exactly when needed.
Output Format
Collaboration Contract: [project] — [team A] × [team B]
Decision Rights
| Decision | Decides | Consulted | Notes |
|---|---|---|---|
| [The forced-ambiguity rows marked ✓ settled] |
Interfaces
[Per handoff type: format · done-definition · the request path]
Communication
[The channel · the sync (cadence + length) · between-syncs format · escalation names]
Friction Protocol
[48h two-lead talk first · no-surprise-escalation clause · the tie-breaker · milestone review date]
Quality Checks
- The both-teams-assume-they-own decisions were forced and settled
- Every handoff type carries a done-definition
- The cadence is justified by need, not reassurance
- The no-surprise-escalation clause is explicit with the 48h talk first
- The milestone review is dated
Anti-Patterns
- Do not skip the contract because the teams get along — the contract is why they'll keep getting along
- Do not leave "who wins priority conflicts" unassigned — that's the fight, pre-scheduled
- Do not accept vibe-based done — bounced handoffs are definition gaps wearing quality-complaint costumes
- Do not escalate surprises — the 48h clause is the partnership's real load-bearing wall
- Do not write three pages — one page gets signed and remembered; three get filed and violated
Version History
- 471c606 Current 2026-07-22 14:02
