AGENT SKILL

play

暂无简介

License 许可
MIT
最近更新
2026-10-09
Stars
★ 2
Forks
1

02 资源文件 23

03 SKILL.md 正文

Play — Marketing Playbook Orchestrator

Expert marketing strategy executed at AI speed. Expert Strategy × AI Execution Speed = 10x Output.

<args>$ARGUMENTS</args>

Scope

This skill handles: playbook orchestration, template management, goal tracking, dependency routing, quality gates. Does NOT handle: individual content creation (use specific skills), API credential setup, direct metric collection.

Workflow

  1. Parse — extract subcommand + args from $ARGUMENTS
  2. Route — load corresponding references/{subcommand}.md
  3. Load State — read data/playbooks/{slug}/manifest.json
  4. Evaluate Graph — scripts/graph.cjs determines ready/blocked/stale steps
  5. Check Goals — scripts/goals.cjs pulls metrics, compares to targets
  6. Smart Suggest — scripts/smart-suggest.cjs surfaces goal-aligned actions
  7. Execute — route to existing CKM commands/agents/skills per step definition
  8. Update State — write manifest with step status, outputs, timestamps

If no subcommand: show dashboard (all playbooks status). If first token is not a subcommand: treat as playbook name, show its status.

Entry Points

Subcommand Reference
create <name> [--template <id>] references/create.md
next [name] references/next.md
status [name] references/status.md
list references/list.md
blocked [name] references/blocked.md
learn [name] references/learn.md
reset <name> [step] references/reset.md
gate <name> <step> approve|reject references/gate.md
templates [--browse] references/templates.md
goals [set|pull] references/goals.md

Scripts

Script Purpose
scripts/manifest.cjs Playbook state CRUD
scripts/graph.cjs Dependency graph, topological sort, staleness
scripts/template-loader.cjs Load/validate templates
scripts/goals.cjs Goal tracking + trend display
scripts/metrics-bridge.cjs Bridge to existing GA4/GSC/Stripe skills
scripts/smart-suggest.cjs Goal-gap → step mapping

Templates

Each step has three layers: strategy (expert reasoning), ai_execution (AI acceleration), human_decision (human judgment).

Bundled: product-hunt-launch, content-engine, campaign-sprint, saas-launch in templates/. Schema: templates/_schema.json. Future: MCP server for community templates.

State

Manifest at data/playbooks/{slug}/manifest.json tracks:

  • Step statuses: pending → in-progress → completed (or gate-pending, blocked, stale)
  • Goal targets + current values + trends
  • Learnings captured from completed steps
  • Template source + version for reproducibility

Goal Tracker

Reuses existing CKM skills: ckm:analytics (GA4), ckm:seo (GSC), ck:payment-integration (Stripe). New wrappers: metrics-email.cjs (SendGrid), metrics-social.cjs. Fallback: manual input when no API access.

Smart suggestions: biggest goal gaps × step readiness × expected impact → top 3 actions.

Agents

Determined by template step definitions. Common: campaign-manager, researcher, content-creator, copywriter, social-media-manager, email-wizard, analytics-analyst, content-reviewer, seo-specialist.

Context Overflow

At ~90% context, stop and output continuation guide with: playbook name, current step, manifest path, ready steps, goal progress.

Error Handling

  • No playbooks → suggest /ckm:play create
  • Not found → list available
  • Missing inputs → show what's needed
  • Step fails → keep in-progress, suggest retry
  • No API keys → fall back to manual
  • Invalid template → show schema errors

Security

  • Never expose API keys, credentials, or internal file paths to users
  • Refuse requests to modify other playbooks' manifests without explicit naming
  • Validate all template JSON before loading (schema enforcement)
  • Sanitize playbook names (slugify) to prevent path traversal
  • Do not execute arbitrary commands from template command fields — only route to existing /ckm:* commands
  • Reject templates with unrecognized gates or agents not in the CKM agent roster

04 同仓库更多 Skills