docs: delete .agent-plan.md + fix AGENTS.md M3a/M8a/#74 mixer-API drift

[shaypal5]

Summary

Two related cleanups to the same drift class:

1. Delete .agent-plan.md — the file fundamentally failed the duplication test (milestone status, open issues, repo pointers, and architectural invariants are all already authoritative elsewhere). The first commit on this branch tried to fix the file by rewriting it; self-review caught that the rewrite reproduced the "if details disagree, the canonical source wins" disclaimer antipattern that PR docs(spec): #84 sync §3.1 steps 1–4 with post-M14 preprocessing #94's review just removed from docs/spec.md §3.1. Replaced with deletion in commit d1b733a.

2. Fix AGENTS.md mixer-API drift — the "TTS" section claimed SceneMixer.mix_sequential took a 4-tuple. The actual API is the Segment dataclass with six named fields. AGENTS.md was at least 2 milestones (M8a, then fix(mixer): #65 Lombard spectral tilt at I4–I5 #74) stale on this rule. Surfaced during the PR docs: delete .agent-plan.md + fix AGENTS.md M3a/M8a/#74 mixer-API drift #96 review when the original .agent-plan.md (5-tuple claim) and AGENTS.md (4-tuple claim) disagreed — turns out they were both wrong; the design-doc tracker (fix(mixer): #65 Lombard spectral tilt at I4–I5 #74 row, line 219) shows the dataclass move neither caught.

Why delete .agent-plan.md (not rewrite)

The file was a duplicate of state held authoritatively elsewhere:

Section	Already in
Milestone status	docs/audio_generation_v3_design.md → Implementation Tracker
Open issues / threads	GitHub Issues; splendor brief --agent-context
Repo orientation pointers	AGENTS.md
Architectural invariants	AGENTS.md (FLOAT subtype line 72, MixedScene leading-pad shift line 37, validate_audio() peak rule line 85, full #78 loudness contract line 35, full #87 effective-prosody cap line 73)

The original got 15 days stale before being noticed; the rewrite would have rotted faster (more state, more surface area). splendor brief --agent-context "<goal>" is the right orientation entrypoint per the feedback_use_splendor_first memory rule established 2026-05-06; the design-doc tracker is the right milestone source; AGENTS.md is the right load-bearing-rules source.

Why nothing was added to AGENTS.md beyond the M8a fix

The "additional invariants" I'd planned to fold in are mixer-internal details (audible_* timeline label use, MixMode no-audio-deps placement, _peak_limit vs _normalize_peak naming). These don't pass the cross-cutting-agent-rules bar — they belong in mixer/label-writer module docstrings if anywhere, not in AGENTS.md.

Changes

File	Change
.agent-plan.md	Deleted (75 lines). Duplicated authoritative state held elsewhere.
.claude/skills/open-feature-pr.md	Step 2 updated: drop the .agent-plan.md fallback for milestone-ID inference; point at the design-doc tracker as authoritative.
state/manifests/sources/src-9d9759e5ad…json	Deleted by splendor source forget --apply.
wiki/sources/src-9d9759e5ad…md	Deleted by splendor source forget --apply.
wiki/index.md	One-line entry for the deleted source removed by splendor source forget --apply.
AGENTS.md	"TTS" M3a bullet updated: 4-tuple (wav_bytes, pause_s, speaker_id, rms_target_dbfs) → current Segment dataclass (six named fields, with the M8a + #74 history that produced it).

Net diff: 2 insertions / 396 deletions.

What's NOT in this PR (and why)

splendor ingest --changed triggered AGENTS.md and docs/spec.md re-ingest (16 state/wiki/planning files of churn). Reverted from this PR — splendor maintenance after this AGENTS.md edit is real follow-up work, but mixing it into a doc fix muddies the diff. Drift will surface on the next splendor brief and get cleaned in a dedicated maintenance pass (per the operation-delegation memory rule).

Self-review trail visible in branch history

• 4361497 — initial rewrite of .agent-plan.md (Option B-shaped attempt; deferred to canonical sources via the antipattern disclaimer).
• d1b733a — supersedes 4361497 with deletion (Option A); includes splendor source forget cleanup.
• b20eaac — adds the AGENTS.md M3a/M8a/fix(mixer): #65 Lombard spectral tilt at I4–I5 #74 fix surfaced during the same review.

The three commits are kept (not amended) so the review trail is visible.

Test plan

• pre-commit — all hooks pass.
• grep -rn "\.agent-plan\.md" outside state//wiki//planning/ — only remaining reference is in .claude/settings.local.json (one-shot allowlist for an earlier gh pr create call, not structural).
• AGENTS.md mixer-API claim cross-checked against synthbanshee/tts/mixer.py:93-117 (Segment dataclass definition).
• No code changes — nothing functional to test.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Updates .agent-plan.md to reflect the current Audio Generation V3 milestone status through M16 and re-orient active work around the ongoing M17 automated evaluation / ASR regression thread.

Changes:

• Refreshes milestone completion list and “headline capabilities” to match the current merged state (M1–M16, with M12 gated; M17 in flight).
• Replaces the previously-stale “Active / next task” section with the current #87/#91 rate-floor-lift thread and related follow-ups.
• Adds/updates “Open threads”, context pointers, and CI/workflow notes (including Tier-3 ASR local-only guidance).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

[github-actions]

pr-agent-context report:

This is a refreshed snapshot of the current PR state.

This run includes unresolved review comments on PR #96 in repository https://github.com/DataHackIL/SynthBanshee

For each unresolved review comment, recommend one of: resolve as irrelevant, accept and implement
the recommended solution, open a separate issue and resolve as out-of-scope for this PR, accept and
implement a different solution, or resolve as already treated by the code.

After I reply with my decision per item, implement the accepted actions, resolve the corresponding
PR comments, and push all of these changes in a single commit.

# Copilot Comments

## COPILOT-1
Location: .agent-plan.md
URL: https://github.com/DataHackIL/SynthBanshee/pull/96#discussion_r3199242302
Status: outdated
Root author: copilot-pull-request-reviewer

Comment:
    This line references an evaluation harness in `synthbanshee/eval/`, but that package does not exist in the repo right now. The Phase A work appears to live in `scripts/m17_phase_a_validation.py`, and the productionized Whisper check is in `synthbanshee/package/asr_sanity.py` (via `qa-report --asr`). Please update this pointer so new readers don’t go looking for a non-existent module.

## COPILOT-2
Location: .agent-plan.md
URL: https://github.com/DataHackIL/SynthBanshee/pull/96#discussion_r3199242348
Status: outdated
Root author: copilot-pull-request-reviewer

Comment:
    `SceneMixer.mix_sequential()` is documented here as taking 5-tuples, but the current API takes `segments: list[Segment]` (see `synthbanshee/tts/mixer.py`). Please update this invariant to match the current Segment-based API so agents don’t build against the old signature.

## COPILOT-3
Location: .agent-plan.md
URL: https://github.com/DataHackIL/SynthBanshee/pull/96#discussion_r3199242368
Status: outdated
Root author: copilot-pull-request-reviewer

Comment:
    This pointer is incorrect: there is no `configs/taxonomy.yaml` in the repo, and the documented source of truth is `synthbanshee/data/taxonomy.yaml` (also referenced in `AGENTS.md`). Please update the path so the context link works.

## COPILOT-4
Location: AGENTS.md
URL: https://github.com/DataHackIL/SynthBanshee/pull/96#discussion_r3200243377
Status: outdated
Root author: copilot-pull-request-reviewer

Comment:
    AGENTS.md now says “#74 added `intensity` for the I4/I5 Lombard high-shelf”, but the Lombard spectral tilt is consistently tracked as #65 in both the design doc (`docs/audio_generation_v3_design.md` §4.2c) and the mixer implementation/docstrings (`synthbanshee/tts/mixer.py`). Consider updating the reference here to #65 as well, so issue/PR cross-links stay consistent across the repo.

Run metadata:

Tool ref: v4
Tool version: 4.0.21
Trigger: schedule
Workflow run: 25498096473 attempt 1
Comment timestamp: 2026-05-07T13:15:19.833732+00:00
PR head commit: 8df8eb56e13b554624a7b876bfd75a0f71e54af1