菜单
Create the MAPS Deploy phase artifact for an agent or multi-agent system. Use when preparing release plans, runtime configuration, secrets, identities, approvals, rollout strategy, rollback plans, and post-deploy verification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | deploy-agent |
| description | Create the MAPS Deploy phase artifact for an agent or multi-agent system. Use when preparing release plans, runtime configuration, secrets, identities, approvals, rollout strategy, rollback plans, and post-deploy verification. |
Current version: 0.1.0.
Follow semantic versioning for this skill:
When changing this skill, update Current version and add a Changelog entry with the date, version, and short summary of behavior changed.
Use this skill to release an evaluated agent into a real environment with controls.
At the start of every project run, look for project-foundation.md. If it exists, read Persistent Memory Contract and use its configured notes, sources, memory, RAG, and sync rules as the project defaults. If .maps/foundation-preferences.json exists, use it as the structured preference source for automation.
When this skill creates durable knowledge, write it through the shared MAPS memory helper. The helper gives this skill its own named note under the configured notes root, mirrors that note into the configured RAG location when one exists, appends MAPS Skill Run Log, and records a RAG reindex manifest.
At the end of the run, call the helper after creating the primary output artifact:
python "$CODEX_HOME/skills/foundation/scripts/maps_memory.py" complete-run --project . --skill /deploy-agent --phase A6 --output "<primary artifact path>" --summary-file "<primary artifact path>" --memory-updates "<notes, sources, memory, or RAG updates>"
If the helper is unavailable, manually append the timestamp, skill, phase, output path, memory updates, and short note to project-foundation.md, then update this skill's named note in <notesRoot>/maps-runs/.
Before writing a deployment plan, ask for any missing answers. Do not assume production targets, secrets, rollout policy, or approval gates.
Ask exactly one question at a time. Do not present the user with a multi-question form, checklist, or table to fill out. Use the questions below as the internal interview sequence: ask the next most important missing question, wait for the answer, then continue.
Ask:
If release readiness is unclear, ask whether to stop, create a pre-release checklist, or produce a draft plan only.
When the skill is complete, tell the user explicitly. Do not end with only files changed or raw output.
Report:
/observe-agent after deployment, or /evaluate-agent again if release gates fail.If the skill is blocked, say what answer, artifact, access, approval, or tool is needed before the next skill can run.
Return:
Use templates/deployment-plan.md from the MAPS repo when working inside this repository.
MAPS Phase 3 Build skill that merges base /build-agent, incremental implementation, and test-driven development into one implementation workflow.
Run MAPS Phase 6 Deploy by wrapping /deploy-agent with GitHub Actions environments and Cloudflare deployment automation. Use when Codex needs GitHub to deploy a Cloudflare Pages, Workers, or Agents project; replace manual Wrangler deploys; configure workflow gates, encrypted secrets, smoke tests, release evidence, and rollback records.
Create the MAPS Design phase artifact for an agent or multi-agent system. Use when defining roles, workflows, handoffs, memory, context, guardrails, approvals, escalation paths, test-first proof requirements, and system boundaries from an agent brief.
Run enhanced MAPS M8 Experience Design for websites, SaaS products, dashboards, landing pages, and agent-facing product surfaces. Use when Codex needs to wrap /design-experience with website design references, component-system choices, accessibility primitives, UX/service-design reasoning, responsive layout, page hierarchy, and production-ready UI handoffs.
Run MAPS M8 Experience Design for multi-agent products, or optional A2 Experience Design for single-agent products without an M layer. Use when Codex needs to turn product intent, orchestration behavior, agent behavior, and product constraints into user journeys, surfaces, information architecture, interaction states, accessibility requirements, and system-evaluation handoffs.
Run MAPS Phase 5 Evaluate by wrapping the base Evaluate process with LangSmith, Inspect AI, and Phoenix. Use when Codex needs to create eval suites, trace-backed eval runs, safety/tool-use checks, regression evidence, eval reports, or Phase 7 observability handoff for an equipped agent.