Jeden Skill in Manus ausführen
mit einem Klick

Jeden Skill in Manus mit einem Klick ausführen

domain-playbooks

Sterne1

Forks0

Aktualisiert1. Juli 2026 um 22:41

Use after task-router classification when work touches a concrete engineering domain (frontend, backend/API, database, infrastructure/cloud, testing, security) — supplies that domain's delegation routing, verification contract, and danger list to embed in MKCREW delegation prompts.

Installation

Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.

In Manus ausführen

Quelle

rayngnpc

rayngnpc/mkcrew

GitHub-Repository öffnen Creator-Repositorys ansehen

Download

In Manus ausführen

Verwandte BerufeSOC

Basierend auf der SOC-Berufsklassifikation

Sonstige ComputerberufeInformatik- und Mathematikberufe·SOC 15-1299

SKILL.md

readonly

name	domain-playbooks
description	Use after task-router classification when work touches a concrete engineering domain (frontend, backend/API, database, infrastructure/cloud, testing, security) — supplies that domain's delegation routing, verification contract, and danger list to embed in MKCREW delegation prompts.

Domain Playbooks

The models already know the domains; what they skip under pressure is domain-specific verification and danger handling. After task-router classifies the work, open the matching playbook(s) and copy the Verify contract and Dangers list into the delegation prompt (safe-agent-delegation owns the prompt mechanics — allowed files, stop condition, job-ID echo). Playbooks compose: a full-stack feature uses several, but each milestone's delegation carries only ITS domain's contract.

Roster facts (who runs what) live in .mkcrew/team.config — verify there, never recall.

Frontend / UI

Route: spec by planner (read-only artifact; planner never writes code); implement by an Opus-tier agent; visual verification by reviewer or via browser tooling.
Verify: build passes; lint on touched files only; responsive sweep at the project's key widths; accessibility basics on touched surfaces (labels matched to control ids, contrast, focus order); zero new console errors on the changed flows.
Dangers: never claim visual correctness from code reading alone — runtime evidence required. Pre-existing lint/contrast debt gets NOTED, not silently fixed (scope).

Backend / API

Route: Opus-tier agent implements. Auth, payments, and distributed-system logic always go to the strongest implementation tier — never to a fast-tier sonnet agent for complex work.
Verify: tests for the changed behavior (failing test first when fixing a bug); the project's full test command after; exercise changed endpoints once for real (test client or curl) including at least one error path; flag any breaking change to a public contract explicitly in the reply.
Dangers: never assert API/SDK facts from memory — read the installed version or docs in-repo; orchestrator recall of API details is unreliable.

Database / Migrations

Route: Opus-tier agent, with the migration plan stated in the delegation; destructive schema changes get a reviewer pass BEFORE apply.
Verify: migration applies cleanly on a dev/copy database first; a rollback path exists and was executed once; post-migration data spot-check; the app boots and its smoke flow works against the migrated schema.
Dangers (HARD STOPS — require explicit human approval inside the delegation text): DROP/TRUNCATE/destructive ALTER on data anyone cares about; mass UPDATE/DELETE without a reviewed WHERE clause and an expected-row-count preview; pointing any migration at a database not clearly identified as dev. Dump/backup before every irreversible step and say where the backup landed.

Infrastructure / Cloud

Route: plan written by planner or main; execution by an Opus-tier agent with this danger list pasted into the delegation verbatim.
Verify: dry-run FIRST, always (terraform plan, kubectl diff, --dry-run, provider equivalents); apply only what the dry-run showed; post-apply smoke check; call out anything cost-relevant (instance sizes, regions, autoscaling, egress) in the reply even when unasked.
Dangers (HARD STOPS): deleting stateful resources; DNS changes; IAM/permission widening; production deploys; anything billable beyond trivial. No --force or --auto-approve flags in delegated commands. Credentials never appear in prompts, replies, or logs — reference their location instead.

Testing / QA

Route: reviewer or an Opus-tier agent runs sweeps and reproduces bugs; a Sonnet-tier agent writes missing tests for well-scoped cases.
Verify: bug fixes start from a failing reproduction; full suite green after; flaky tests get quarantined and reported, never deleted to go green.
Dangers: editing a test's assertions to make it pass is a reviewer-visible event, never a silent change.

Security-sensitive changes

Route: Opus-tier agent implements; the reviewer pass MUST include an explicit security dimension (injection, authz on new surfaces, secrets in code/logs).
Verify: input validation on every new external surface; authz checked on new endpoints (not just authn); no secrets in diffs or output.
Dangers: auth, payments, crypto, and PII handling are never bundled into unrelated milestones and never "quick fixes" — they get their own bounded milestone with review, every time.

Changelog

2026-06-13: Adapted for mkcrew. Replaced worker/tester roster references with Opus-tier/Sonnet-tier/reviewer. Removed Sackysocky tool references (not present on this clone). Routing and verification contracts are otherwise faithful to the Linux original.

Mehr aus diesem Repository

gleiches Repository

safe-agent-delegation

rayngnpc/mkcrew

Use before delegating coding or planning to MKCREW agents when git backup, worktree isolation, detailed prompts, allowed files, stop conditions, and verification are needed. Canonical source of truth for the planner read-only prompt, git-checkpoint commands, reply discipline, and delivery watchdog.

2026-07-011

senior-developer-loop

rayngnpc/mkcrew

Use when asked to keep developing a project, run an autonomous senior developer loop, keep improving / iterate over milestones, or coordinate the MKCREW team across reviewable milestones. This skill owns milestone coordination only; for delegation prompts and git-backup mechanics it invokes safe-agent-delegation rather than duplicating them. Do not use for one-off single edits or initial task classification (use task-router).

2026-07-011

task-router

rayngnpc/mkcrew

Use at the start of broad, ambiguous, multi-step, risky, or MKCREW team tasks to classify the request, pick skills, pick agents/tier, set the git-safety level, and emit one routing block BEFORE any implementation. Use for a single classification pass at the start; for an ongoing autonomous build loop, route to senior-developer-loop and stop. It does not replace implementation skills; it selects them.

2026-07-011

team-self-improvement

rayngnpc/mkcrew

Use after a MKCREW task when experience should become durable: the same mistake recurs, reviewer flags a preventable issue, a worker misunderstands a delegation, a check is missing/flaky/too broad, the human corrects team behavior, or a workflow succeeds repeatedly and should be reusable. Converts experience into memory, skills, scripts, tests, or lessons. Not for implementing features or routing a fresh task.

2026-07-011

name	domain-playbooks
description	Use after task-router classification when work touches a concrete engineering domain (frontend, backend/API, database, infrastructure/cloud, testing, security) — supplies that domain's delegation routing, verification contract, and danger list to embed in MKCREW delegation prompts.

Domain Playbooks

Roster facts (who runs what) live in .mkcrew/team.config — verify there, never recall.

Frontend / UI

Route: spec by planner (read-only artifact; planner never writes code); implement by an Opus-tier agent; visual verification by reviewer or via browser tooling.
Verify: build passes; lint on touched files only; responsive sweep at the project's key widths; accessibility basics on touched surfaces (labels matched to control ids, contrast, focus order); zero new console errors on the changed flows.
Dangers: never claim visual correctness from code reading alone — runtime evidence required. Pre-existing lint/contrast debt gets NOTED, not silently fixed (scope).

Backend / API

Route: Opus-tier agent implements. Auth, payments, and distributed-system logic always go to the strongest implementation tier — never to a fast-tier sonnet agent for complex work.
Verify: tests for the changed behavior (failing test first when fixing a bug); the project's full test command after; exercise changed endpoints once for real (test client or curl) including at least one error path; flag any breaking change to a public contract explicitly in the reply.
Dangers: never assert API/SDK facts from memory — read the installed version or docs in-repo; orchestrator recall of API details is unreliable.

Database / Migrations

Route: Opus-tier agent, with the migration plan stated in the delegation; destructive schema changes get a reviewer pass BEFORE apply.
Verify: migration applies cleanly on a dev/copy database first; a rollback path exists and was executed once; post-migration data spot-check; the app boots and its smoke flow works against the migrated schema.
Dangers (HARD STOPS — require explicit human approval inside the delegation text): DROP/TRUNCATE/destructive ALTER on data anyone cares about; mass UPDATE/DELETE without a reviewed WHERE clause and an expected-row-count preview; pointing any migration at a database not clearly identified as dev. Dump/backup before every irreversible step and say where the backup landed.

Infrastructure / Cloud

Route: plan written by planner or main; execution by an Opus-tier agent with this danger list pasted into the delegation verbatim.
Verify: dry-run FIRST, always (terraform plan, kubectl diff, --dry-run, provider equivalents); apply only what the dry-run showed; post-apply smoke check; call out anything cost-relevant (instance sizes, regions, autoscaling, egress) in the reply even when unasked.
Dangers (HARD STOPS): deleting stateful resources; DNS changes; IAM/permission widening; production deploys; anything billable beyond trivial. No --force or --auto-approve flags in delegated commands. Credentials never appear in prompts, replies, or logs — reference their location instead.

Testing / QA

Route: reviewer or an Opus-tier agent runs sweeps and reproduces bugs; a Sonnet-tier agent writes missing tests for well-scoped cases.
Verify: bug fixes start from a failing reproduction; full suite green after; flaky tests get quarantined and reported, never deleted to go green.
Dangers: editing a test's assertions to make it pass is a reviewer-visible event, never a silent change.

Security-sensitive changes

Route: Opus-tier agent implements; the reviewer pass MUST include an explicit security dimension (injection, authz on new surfaces, secrets in code/logs).
Verify: input validation on every new external surface; authz checked on new endpoints (not just authn); no secrets in diffs or output.
Dangers: auth, payments, crypto, and PII handling are never bundled into unrelated milestones and never "quick fixes" — they get their own bounded milestone with review, every time.

Changelog

2026-06-13: Adapted for mkcrew. Replaced worker/tester roster references with Opus-tier/Sonnet-tier/reviewer. Removed Sackysocky tool references (not present on this clone). Routing and verification contracts are otherwise faithful to the Linux original.