A Miadi-native Copilot plugin that turns a user’s desire (plus target directories) into a ready-to-run orchestration bash script.
How it works:
copilot_prepare_orchestration runs a fast bash scan of target dirs, kits, and all agents (~2 seconds, no AI)scan-context.md + prompt.md into a PDE working folder at .pde/<yyMMddHHmm>--<uuid>/@.pde/.../prompt.md — direct forge, no background subagentsorchestration-<slug>-<date>.sh into the PDE foldersource /workspace/repos/jgwill/miadi-orchestration-kit/copilot/miadi-wave-forge-kit/scripts/copilot_prepare_orchestration.sh
copilot_prepare_orchestration "refactor Ava Decomposer Studio and its ava-*js packages" \
/workspace/repos/avadisabelle/Ava-Decomposer-Studio \
/workspace/repos/avadisabelle/ava-langchainjs
If the session ends without writing the script, retry with:
copilot --add-dir '.pde/<ts>--<uuid>' -p '@.pde/<ts>--<uuid>/prompt.md'
Skills (scan-and-plan, review-wave-plan, forge-wave-script, resume-wave-prep) remain available for interactive Copilot sessions when you want fine-grained control:
copilot \
--model claude-sonnet-4.6 \
--plugin-dir /workspace/repos/jgwill/miadi-orchestration-kit/copilot/miadi-wave-forge-kit \
--add-dir /workspace/repos/avadisabelle/Ava-Decomposer-Studio \
-p "Use scan-and-plan then review-wave-plan then forge-wave-script for: refactor Ava Decomposer Studio"
Note: the skill chain runs background subagents sequentially — prefer copilot_prepare_orchestration for unattended use.
| Agent | Description |
|---|---|
Orchestration Context Scanner |
Scans target directories and plugin sources. Produces a structured Scan Report (no recommendations — pure discovery). |
Orchestration Wave Planner |
Takes user desire + Scan Report. Selects agents and kits, designs numbered phases, produces a WavePlan document. |
Wave Script Forge |
Takes a WavePlan. Writes a complete executable bash script with a comprehensive self-contained PROMPT and the final copilot command. |
| Skill | Description |
|---|---|
scan-and-plan |
Orchestrates Scanner → Planner flow. Validates scan completeness before planning. Outputs WavePlan. |
review-wave-plan |
Adversarial review of a WavePlan: checks agent validity, path existence, desire alignment, issue/review/synthesis phase presence. Returns ADVANCE or REVISE. |
forge-wave-script |
Invokes Wave Script Forge on a validated WavePlan (requires ADVANCE verdict), verifies the script with bash -n + preflight check, reports path and first 50 lines. |
resume-wave-prep |
Resumes an interrupted wave-forge session. Locates the most recent .pde/ folder and retries the forge from the preserved prompt.md. |
$ copilot_prepare_orchestration "add integration tests to ava-langchainjs" \
/workspace/repos/avadisabelle/ava-langchainjs
[wave-forge] PDE working folder : /workspace/wikis/.pde/2605281015--<uuid>/
[wave-forge] Scanning directories...
[wave-forge] Scan context : scan-context.md (287 lines)
[wave-forge] Prompt : prompt.md (195 lines)
[wave-forge] Launching copilot... (direct forge — no subagents)
────────────────────────────────────────────────────────────────────────────────
copilot writes orchestration-add-integration-tests-ava-260528.sh directly
────────────────────────────────────────────────────────────────────────────────
[wave-forge] ✓ Script created: .pde/2605281015--<uuid>/orchestration-add-integration-tests-ava-260528.sh
$ bash .pde/2605281015--<uuid>/orchestration-add-integration-tests-ava-260528.sh
If copilot ends the session before writing the script, the PDE folder is intact — retry:
copilot --add-dir '.pde/<ts>--<uuid>' -p '@.pde/<ts>--<uuid>/prompt.md'
miadi-wave-forge-kit/
├── .github/plugin/plugin.json
├── agents/
│ ├── orchestration-context-scanner.md
│ ├── orchestration-wave-planner.md
│ └── wave-script-forge.md
├── skills/
│ ├── scan-and-plan/SKILL.md
│ ├── forge-wave-script/SKILL.md
│ └── review-wave-plan/SKILL.md
├── scripts/
│ └── copilot_prepare_orchestration.sh
└── README.md