skill-synthesis

name	skill-synthesis
description	Compose multiple installed skills into one coordinated execution stack with ordered packets, minimal context load, and deterministic handoff artifacts. Use when tasks span multiple domains (for example HYDRA + browser + training + deploy) and you need a single combined workflow instead of invoking skills one-by-one.

Skill Synthesis

Overview

Use this skill to fuse multiple installed skills into one execution loop with clear ordering, packet boundaries, and artifacts.

Quick Start

1. Build a stack from a task prompt.

python C:\Users\issda\.codex\skills\skill-synthesis\scripts\compose_skill_stack.py --task "Build gamma funnels and deploy" --top 8

2. Run the resulting packets in order from highest leverage to lowest risk.

Workflow

1) Select Minimal Stack

• Pick one primary skill (owns outcome).
• Pick up to four support skills (execution lanes).
• Do not load the whole skill catalog unless explicitly requested.

2) Build Ordered Packets

• Packet A: discovery/research
• Packet B: implementation
• Packet C: validation/smoke
• Packet D: publish/deploy
• Packet E: evidence + vault notes

3) Execute with Context Discipline

• Keep live context to only the active packet.
• Move long notes to artifacts or vault docs.
• Keep source links and proofs in packet output.

4) Output Contract

• stack plan JSON
• ordered packet list
• execution report (what ran, what passed, what is blocked)

Built-in Stack Profiles

hydra-library-wing

• hydra-agent-relay-synthesis
• aetherbrowser-arxiv-nav
• aetherbrowser-github-nav
• hugging-face-model-trainer
• notion

Use for deep research + dataset handoff + multi-agent synthesis.

revenue-gamma-funnel

• living-codex-browser-builder
• article-posting-ops
• scbe-shopify-money-flow
• aetherbrowser-shopify-nav
• vercel-deploy

Use for landing pages, conversion flow, and web deployment.

platform-release

• development-flow-loop
• playwright
• scbe-connector-health-check
• vercel-deploy

Use for implementation -> smoke -> deploy.

Rules

• Favor official docs and primary sources for unstable facts.
• Keep stack size small unless user explicitly asks for full-spectrum mode.
• Prefer deterministic scripts and repeatable packets over ad-hoc chat instructions.
• Record evidence paths for every packet.

Resources

scripts/

• compose_skill_stack.py: Generate ranked skill stacks and packet plans from a task description.

references/

• stack-profiles.md: Reusable profile templates and packet recipes.