Skip to content

INSTALL.md

Latest commit

 

History

History
100 lines (65 loc) · 3.79 KB

INSTALL.md

File metadata and controls

100 lines (65 loc) · 3.79 KB
Raw

Installing DFT Skills For Clouds_Coder

This repository is optimized first for the FonaTech/Clouds-Coder ecosystem and specifically for the loading behavior of Clouds_Coder, while remaining portable to Codex, Claude Code, and OpenCode.

GitHub quick link:

This repository is already laid out in a form that Clouds_Coder can discover:

  • package root: DFT_Skills/
  • native skill tree: DFT_Skills/skills/dft-workflow-orchestrator/SKILL.md

Repository-level docs:

GitHub-friendly repository adapters:

Those repository folders are intentionally visible for GitHub upload and browsing. Actual platform installs still go to runtime-native paths such as .claude/, .opencode/, ~/.codex/, and ~/.agents/.

Clouds_Coder.py can see the portable source directly, while the generated mirror is the optimized compact-loading path.

Mode 1: Direct Skills Root

Use DFT_Skills/skills as the active skills_root.

This is the cleanest portable mode if the repository itself is dedicated to this skill pack. It reads the standard source SKILL.md; use Mode 3 when Clouds-specific aliases, entrypoints, and compact metadata are required.

Mode 2: External Library Auto-Discovery

Keep DFT_Skills/ adjacent to the active skills/ directory.

Clouds_Coder.py scans sibling directories of the active skills_root and auto-discovers external libraries that contain skills/*/SKILL.md.

That means a workspace like this is valid:

workspace/
├── skills/
└── DFT_Skills/
    └── skills/
        └── dft-workflow-orchestrator/
            └── SKILL.md

Mode 3: Project-Local Native Mirror

Mirror the skill into the standard Clouds project path:

python3 DFT_Skills/skills/dft-workflow-orchestrator/scripts/sync_skill_to_platforms.py \
  --repo-root "$(pwd)" \
  --targets clouds \
  --force

This creates:

  • skills/generated/dft-workflow-orchestrator/

That path is the most predictable and recommended Clouds option when the main workspace already uses a skills/ tree. The sync command must use copy mode because it applies the platform overlay only to the generated target.

On-Demand Loading Alignment

The skill separates portable and Clouds-specific metadata:

  • source SKILL.md frontmatter contains only standard name and description
  • agents/clouds-coder.json contains aliases, triggers, attachments, entrypoints, preferred tools, and the runtime contract
  • sync_skill_to_platforms.py --targets clouds --mode copy renders that overlay into the generated Clouds copy
  • reusable resources are exposed through attachments
  • the rendered copy can enter compact mode so the runtime shows the contract and resource manifest before deeper reads

Verification

Run the built-in compatibility check:

python3 DFT_Skills/skills/dft-workflow-orchestrator/scripts/verify_clouds_compat.py

The verifier checks:

  • standard source frontmatter
  • the Clouds sidecar overlay and all entrypoints
  • rendered project-local copy generation
  • entrypoint presence
  • Clouds compact-mode loading behavior when Clouds_Coder is importable

Boundary Rules

  • keep the skill bundle itself read-only during normal workflow execution
  • keep generated structures, workflow files, run directories, logs, and analysis outputs in the user project root
  • do not write calculation outputs back into DFT_Skills/skills/...