Skip to content
Agentmart
Submit a tool

teamplay

Save implementation cost by defaulting specification-locked coding to GPT-5.6 Luna max while the current main Codex agent retains specification, integration…

Skill by youngchangjo

IndexedMIT★ 22
Write a review

Chalk Labs

Need help deploying teamplay?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/youngchangjo-teamplay

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Save implementation cost by defaulting specification-locked coding to GPT-5.6 Luna max while the current main Codex agent retains specification, integration, final review, acceptance QA, and final Gate judgment. Teamplay never selects Sol; Terra xhigh is the maximum child route and explicit or evidenced post-Luna exception only. Use when the user invokes Teamplay, asks for cost-efficient Luna implementation, requests optional Fast children, or wants bounded parallel implementation.

Use cases: testing, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Python package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/teamplay

What's inside

Skill name
teamplay
Description
Save implementation cost by defaulting specification-locked coding to GPT-5.6 Luna max while the current main Codex agent retains specification, integration, final review, acceptance QA, and final Gate judgment. Teamplay never selects Sol; Terra xhigh is the maximum child route and explicit or evidenced post-Luna exception only. Use when the user invokes Teamplay, asks for cost-efficient Luna implementation, requests optional Fast children, or wants bounded parallel implementation.
License
MIT
Instructions
282 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (22)

  • SKILL.md
  • delivery-speed.md · references/
  • evidence-contract.md · references/
  • execution-policy.md · references/
  • qa-surfaces.md · references/
  • reporting.md · references/
  • role-contracts.md · references/
  • routing.md · references/
  • runtime-identity.md · references/
  • session-continuity.md · references/
  • spec-contract.md · references/
  • inspect-agent-runtime.sh · scripts/
  • render-task-packet.py · scripts/
  • continuation-packet.md · templates/
  • final-report.md · templates/
  • lead-review.md · templates/
  • qa-packet.md · templates/
  • research-packet.md · templates/
  • review-packet.md · templates/
  • spec-brief.md · templates/
  • spec-contract.md · templates/
  • task-packet.md · templates/

SKILL.md (first lines)

# Teamplay

Teamplay is cost-first: the current main agent is always Lead and the default
implementation Coder is Luna max. The Lead spends its existing context on the
specification, integration, final review, acceptance QA, and final Gate judgment;
Luna supplies lower-cost implementation throughput. Teamplay never creates a
Sol child. Terra xhigh is the maximum child route and is used only by explicit
user choice or an evidenced post-Luna capability exception.

## Read contracts progressively

Read a detailed contract when its stage begins, not all contracts up front:

- `references/routing.md` before selecting a model or writer pool;
- `references/spec-contract.md` before locking a Brief or Full Spec;
- `references/session-continuity.md` before assigning or continuing a mutating
  Coder;
- `references/execution-policy.md` only when rendering an initial child packet;
- `references/runtime-identity.md` after spawning any child whose route or
  isolation will be claimed;
- `references/role-contracts.md` before selecting optional roles;
- `references/delivery-speed.md` for Fast, waves, early checks, or repair;
- `references/evidence-contract.md` before judging completion evidence;
- `references/qa-surfaces.md` before acceptance QA;
- `references/reporting.md` when closing the run.

Use `templates/spec-brief.md` for one bounded ordinary outcome and
`templates/spec-contract.md` for a Full Spec Lock. Use
`templates/continuation-packet.md`, `templates/lead-review.md`,
`templates/qa-packet.md`, and
`templates/final-report.md` at their named Lead stages.

## Fixed authority

- The current main conversation agent is Lead; never spawn another Lead.
- Keep the Lead's selected model, reasoning effort, and service tier unchanged.
- The Lead owns specification interpretation, route choice, integration, final
  code review, acceptance QA, final Gate judgment, conflict resolution, and
  completion verdict.
- The Lead may directly take over a stalled Coder's unchanged whole outcome
  after the lifecycle contract's wait and same-session redirect.
- Child Reviewer and QA roles are advisory only. There is no Gate child; the
  current main Lead performs the final Gate directly.
- Every Luna child uses max reasoning. Fast changes only the selected Luna
  child's service tier.
- Luna max is the default mutating Coder. Difficulty, ambiguity, criticality,
  and file count do not select an initial Sol Coder.
- Luna max has a 90% lower-bound audit floor, not a quota. Terra has zero
  allocation budget and no target share; it requires T1 or T2 for each outcome.
  Without T1 or T2, the current run is 100% Luna. Never create Terra to fill 10%.
- Sol is prohibited for every Teamplay child. Terra xhigh requires
  `T1 explicit_user_terra` or `T2 evidenced_luna_capability_blocker` from
  `references/routing.md`.

The Lead owns review, QA, and Gate because it alone retains the canonical user
conversation and locked specification, sees the integrated cross-outcome diff,
controls faithful QA surfaces and evidence-layer separation, and holds final
authority. A Coder's checks support review but cannot approve its own work. If
the Lead writes code after takeover, it reopens the pre-existing specification
checklist and keeps implementation, review, QA, and Gate as separate phases.

## Workflow

### 1. Establish the baseline

Read repository instructions and `CHANGELOG.md`, inspect branch/worktree state,
canonical product documents, relevant code, and available validation surfaces.
Preserve unrelated and user-owned changes. State the requested outcome and
observable success criteria.

### 2. Apply routing precedence

Use the exact R0-R3 order in `references/routing.md`:

1. authority;
2. specification readiness;
3. model;
4. writer pool.

Do not use a stronger model to bypass missing authority, an unresolved decision, or an
unfinished specification. A failed readiness check returns to the Lead; it does
not select Terra, and Sol is never available.

### 3. Lock the appropriate written specification

The Lead creates or confirms exactly one canonical Spec Brief or Full Spec Lock
before mutation. Record a stable spec revision and repository baseline.

The specification must be sufficient for the Lead to review each requirement
against the actual diff and execute requirement-linked QA. It locks decisions,
not keystrokes. Resolve material product ambiguity before delegation.

The ordinary Spec Brief exposes five implementation-facing sections—objective,
ownership, interfaces, constraints, and verification—behind one compact control
header. Keep stable requirement IDs and acceptance evidence, but do not add a
per-file recipe or implementation sequence. Use Full Spec Lock only when its
actual coordination, risk, authority, or evidence triggers apply.

Hard, novel, cross-cutting, security-sensitive, concurrent, or migration work
may require a Full Spec Lock, but difficulty does not change the default Coder.
The Lead resolves consequential choices first, then delegates the locked whole
implementation outcome to Luna max.

### 4. Keep one Coder session for one outcome

Apply the lifecycle contract before spawning or continuing a Coder. Define one
independently integratable outcome, derive its stable session key from the spec
ID/revision, outcome ID, route, and owned surfaces, and keep one Coder identity
through implementation, directly coupled checks, Lead feedback, and bounded
in-spec repairs. A file, layer, exact code fragment, command, checklist item,
test failure, or repair is not a new outcome.

The initial assignment gets exactly one rendered execution capsule. For an
unchanged session key, use message/reuse or resume with the compact,
capsule-free continuation packet. Use redirect once for bounded non-progress.
A same session key keeps the prior Coder identity; it does not create a new
assignment for a focused check or repair.
Closing a completed child to release a concurrency slot is resource management,
not a restart boundary. Retain its agent ID and resume it if Lead review or QA

From youngchangjo/teamplay, MIT.

Signals

Stars
22
Forks
1
Open issues
0
Last commit (this folder)
Aug 5, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from youngchangjo

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get teamplay running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying teamplay?

CallGet help
Cite this page

teamplay skill. Agent Mart. https://agentmart.dev/skills/youngchangjo-teamplay (last updated 2026-10-24)