Cubase-MCP-GAJ3T.MusicGen.R.../prompts/step_7.md

92 lines
4.8 KiB
Markdown

# Step 7: Ample native capability assessment and safe fallback boundary
## Objective
Execute the Ample portion of the amended plan after reading reports through
`report_6.md`. Text-to-MIDI and standard mappings are accepted. The real AGP
layout is not established. Implement a truthful native-capability boundary
ready for Step 9; implement serialization only if actual evidence supports it.
The continuation decision permits independent fallback functionality while
preserving a blocked native-serialization acceptance gate. A capability check
is not a completed AGP writer. Never write renamed MIDI, invented JSON/XML, or a
preset that cannot carry the requested musical content.
## Owned files
- `src/riff_groove_builder/formats/__init__.py`
- `src/riff_groove_builder/formats/agp.py`
- `tests/unit/test_agp.py`
- `tests/integration/test_agp_capability.py`
- `docs/formats/agp.md`
- Relevant compatibility/architecture status updates.
- `prompts/report_7.md`
No commits, delegation, later prompts, SDP implementation, or MCP service wiring.
Preserve prior reports and all existing tests/code.
## Evidence already obtained
Raw https://www.amplesound.net/en/manual.asp exposes official public manuals
Main_Panel_Manual-AME.pdf, Main_Panel_Manual-AMR.pdf, Guitar_Riffer.pdf,
Guitar_Strummer.pdf and Settings_and_CPCv2.pdf.
The main-panel documents contain articulation/range labels; Riffer documents
MIDI import/export and preset Save/Load UI, not a verified native binary schema.
Their availability corrects the earlier simplified-page extraction error.
The publicly linked demo archive
https://www.amplesound.net/dl0822/ds/AME-Eclipse.zip was readable without login.
Its metadata contained MIDI, DAW projects and plugin state files, but no AGP
files. Do not re-download/copy factory music into tests or treat that absence as
proof no AGP format exists. No actual AGP layout has been observed.
Review accessible relevant official evidence as needed, but do not loop through
guessed URLs or bypass access controls. Record facts, limitations and sources.
## Implementation requirements
1. Provide a small typed capability API for lead/bass AGP support, reusable by
service.py. Validate target/part and distinguish no verified format support
from bad requests or corrupted input.
2. A shared concrete NativeCapability type in formats/__init__.py is acceptable
for reuse by SD3. Keep fields simple: part, format, available and safe reason.
No generic exporter plugin framework or speculative native parser.
3. Current standard profiles and unknown layouts must return available=false
with explicit stable NATIVE_FORMAT_UNAVAILABLE context. Do not assume
native_available metadata alone proves a codec exists.
4. Expose an explicit require/check boundary that raises the repository's named
native-unavailable error before any path is created or touched when support
is absent. Do not add a fake successful write API.
5. If lawful real fixtures/specifications actually establish a layout, implement
its correct writer and validation with evidence-based tests; otherwise
document that serialization remains unimplemented and application untested.
6. Document expected future lead/bass artifact roles/names, separate product
version handling, precise missing schema/fixture requirements and real
Eclips/Ray load/save acceptance needed.
## Required unit and integration tests
- Lead and bass capability results are typed and explicit.
- Drum/wrong-target or mutated-profile input is rejected, not misclassified.
- Standard profile/native false and codec-not-established cases remain honest.
- Require/check raises named errors with format/part/reason context.
- Capability/unavailable operations create no files, do not overwrite an existing
sentinel, and never return fake artifact paths or MIDI with AGP extensions.
- Preserve profile/compiler/MIDI/package regressions; run focused new tests,
Ruff lint/format and strict mypy.
- Real native round trips/application checks only if genuine evidence exists;
missing tests are reported UNRUN, never counted as passing compatibility.
## Acceptance and report
The software capability/fallback boundary may pass while full native AGP
acceptance remains BLOCKED. State both distinctly. No application evidence may
be fabricated. This step is substantial evidence/contract work, not a claim that
an unavailable capability is a native writer.
When complete, write your end-of-step report to ./prompts/report_7.md including: what was implemented, test results (pass/fail with details), any issues encountered, and whether you're ready for the next step
The whole report must be inside exactly one plain triple-backtick fence. Include
files/API, exact test/check outcomes, evidence conclusions, unresolved native
gate, and readiness for the next independent capability/service work.