Cubase-MCP-GAJ3T.MusicGen.R.../prompts/step_10.md

99 lines
5.7 KiB
Markdown

# Step 10: Installed software acceptance and truthful delivery documentation
## Objective and prerequisites
Read the amended plan and reports through `report_9.md`. The operational
Standard MIDI service was independently verified with 140 focused service/server
tests. Native AGP/SDP writers and real vendor/application acceptance remain
blocked, not passed. Complete all independent software/package/docs checks now.
This is not permission to label the original full native project complete.
Report software acceptance and native/application acceptance separately.
No scope expansion, commits, delegation or additional worker prompts.
## Owned files
- Add `tests/integration/test_installed_server.py`, or extend the existing
distribution/stdio integration surface if simpler without duplication.
- Add narrowly necessary regression tests/source fixes for defects found here.
- Update `README.md`, `docs/architecture.md`, `docs/compatibility.md`,
`docs/articulation-profiles.md` and format docs only for accurate final status.
- Correct the plan's obsolete initial "proposed/awaiting review" status to the
actual approved/partial-native status; preserve historical decisions/reports.
- Update packaging metadata only for an actual distribution defect.
- Write `prompts/report_10.md`.
Do not change prior worker reports, invent vendor profiles, fabricate native
files or application results, add more formats, or create unrequested planning
files. Preserve unrelated work.
## Installed-distribution acceptance
1. Build current wheel AND sdist from the final source.
2. Run the actual installed wheel's server outside the checkout, not accidentally
importing the editable source. Assert the imported package/submodule locations.
3. Reuse existing development dependencies without unnecessary reinstalls.
An offline --no-deps install into an isolated target plus an explicit isolated
Python probe path is acceptable when documented accurately; do not call that
a fresh full dependency certification. Avoid network access in normal tests.
4. Through real stdio, initialize, list exactly two tools and three standard-only
profiles, generate a combined explicitly requested MIDI and default/native
fallback, and verify structured outputs against published schemas.
5. Parse actual published MIDI: exact type/PPQ/tempo/meter/track layout, aligned
duration, legal/pair-complete notes, profile IDs, channels and warning/status
coverage. Verify native-disabled and unsupported techniques produce errors
with no output directories.
6. Verify protocol-clean UTF-8 stdout, separate diagnostics, working-directory
independence, clean EOF and bounded child cleanup.
7. Exercise packaged JSON/resources as real wheel contents. Preserve all current
distribution/source/profile/compiler/MIDI/native-boundary/service tests.
8. No Cubase/plugin acceptance is implied by Mido opening a file. Record those
application checks as UNRUN if applications/evidence are unavailable.
## Documentation and consistency
- Present the project as an operational Windows Standard MIDI MCP generator with
explicit unavailable-native fallback, NOT a finished native AGP/SDP exporter.
- Include verified setup, required existing output root, absolute Windows MCP
client configuration, profile discovery, concrete accepted request examples,
actual response shape and error/fallback behavior.
- Make the closed local grammar discoverable; do not imply arbitrary descriptions
or vendor techniques work. Explain normal/sustain, unsupported palm mute with
standard maps, exact keys/meters/subdivisions/defaults, and override precedence.
- Explain standard mapping identity and receiver routing, no bank/program setup
or claimed real vendor compatibility, and valid MIDI import versus native files.
- Explain output-directory naming, no overwrite, Windows publication restriction,
trusted ACL assumptions, reparse checks, cleanup behavior and race/crash limits.
- Keep native format requirements and `step_7_fix.md`/`step_8_fix.md` handoffs
explicit; required real fixture/schema/application evidence is not a pytest skip.
- Update compatibility matrix with actual tested Python/OS/dependency versions,
installed-package/software results and separately unrun applications.
- Ensure no stale "empty registry" or "generation unimplemented" statements remain
in current docs, except clearly identified historical reports/decisions.
## Required validation
- New installed-wheel stdio integration and any targeted regressions.
- Complete current pytest suite.
- Ruff lint and format checks, strict mypy and pip check.
- Actual wheel/sdist build from final code.
- Confirm owned temporary build/test processes/workspaces clean up, no background
server remains, and no secrets/manuals/factory songs were added.
Use existing tools/configuration. The known editor pytest discovery limitation
may use the verified repository-venv CLI fallback. Fix tightly coupled defects
with focused regressions; do not weaken checks or silently skip failures.
## Acceptance and report
Software delivery is verified only when actual installed-wheel stdio generation,
all automated checks and documentation consistency pass. Full Step 10 original
native/application acceptance remains BLOCKED while those inputs are missing.
When complete, write your end-of-step report to ./prompts/report_10.md including: what was implemented, test results (pass/fail with details), any issues encountered, and whether you're ready for the next step
Use exactly one plain triple-backtick fence for the entire report. Include exact
commands/counts, installed-wheel evidence, software/build status, cleanup,
precise remaining native/vendor/application blockers, and an explicit statement
that no full native-project completion is claimed.