docs(build): note optional setup-deepfilter.sh prereq for local DFNR

[ten9876]

What

Adds a short note to the Build section of AGENTS.md documenting the optional scripts/setup/setup-deepfilter.sh step that enables DFNR (DeepFilterNet3) noise reduction for local dev builds.

Why

On a clean checkout, CMake configure reports:

DFNR (DeepFilterNet3) disabled — library not found ... (run ./scripts/setup/setup-deepfilter.sh before cmake to enable)

CI already runs this setup step automatically (cached) in every build workflow — ci.yml, macos-dmg.yml, appimage.yml, windows-installer.yml — so release binaries always ship DFNR. The only gap was the local dev path, where the step was undocumented in AGENTS.md / README / CONTRIBUTING and surfaced only via the configure message. This documents it (including the Windows .ps1 variant) and clarifies NR still works without it via NR4 (libspecbleach).

Docs-only; no code or CI changes.

🤖 Generated with Claude Code

[NF0T]

Verified the factual claims directly rather than a full review pass, since there's no logic here to test -- this is docs-only:

• scripts/setup/setup-deepfilter.sh and setup-deepfilter.ps1 both exist.
• CMake's message matches exactly: "DFNR (DeepFilterNet3) disabled -- library not found at ${DFNR_LIB}...".
• All four workflows named (ci.yml, macos-dmg.yml, appimage.yml, windows-installer.yml) do reference setup-deepfilter.

Everything checks out, CI is green. Nothing to request changes on.

One process note, not about this PR's content: this touches AGENTS.md, which CODEOWNERS routes specifically to @aethersdr/maintainers (overriding the broader Tier 3 default), and you're currently the sole member of that team. GitHub hard-blocks self-approval regardless of team membership, so as it stands nobody -- including us -- can satisfy that gate to merge this. Worth knowing for any future AGENTS.md-only PR you author yourself, not just this one.

[aethersdr-agent]

Nice, well-targeted docs addition — thanks @ten9876. I verified every concrete claim against the tree and they all hold up:

• Both scripts/setup/setup-deepfilter.sh and the .ps1 Windows variant exist.
• The quoted configure message matches CMakeLists.txt:450 verbatim, including the (run ./scripts/setup/setup-deepfilter.sh before cmake to enable) hint and the ENABLE_DFNR OFF gating right above it.
• All four named workflows (ci.yml, macos-dmg.yml, appimage.yml, windows-installer.yml) do invoke setup-deepfilter, so the "release binaries always ship DFNR" framing is correct.
• The NR4/libspecbleach fallback is real (CMakeLists.txt:358), so "NR still works without it via NR4" is accurate.

The note lands in the right place (Build section, right after the build block) and is appropriately scoped as an optional local-dev prereq. Clean, accurate, signed. No blockers.

Non-blocking note

• Minor: "every build workflow" is really "every release-producing build workflow" — sanitizers.yml / build-macos-qt.yml also compile but don't run the step. Not worth a change; the four you name are the ones that matter for shipped binaries.

Happy to see this merge as-is.

---

_{🤖 aethersdr-agent · cost: $9.6422 · model: claude-opus-4-8}

[K5PTB]

Red-Team Review — PR #4282

Adversarial review, not a behavior-preservation pass. Read AGENTS.md/CONSTITUTION.md, verified every concrete claim in the PR body against the actual tree rather than trusting it, checked CI/reviews, and refreshed both before finalizing.

Tiny, low-risk, docs-only PR. There's already a bot review (@aethersdr-agent, COMMENTED, no blockers) that independently verified the same core claims I did — not duplicating that, just noting where my pass adds something it missed.

---

Findings

1. [Non-blocking, but real] The PR's own stated gap ("undocumented in AGENTS.md / README / CONTRIBUTING") is only half-closed.
The PR body explicitly says: "the step was undocumented in AGENTS.md / README / CONTRIBUTING..." The diff, however, touches only AGENTS.md. I checked: README.md's "Building from Source" / "Build & Run" section (the doc AGENTS.md itself points to — "Full dependency list is in README.md") mentions NR2 (libfftw3-dev) but has zero mention of DFNR or the setup script anywhere in that section. CONTRIBUTING.md likewise still only lists HAVE_DFNR in a build-flag table, no setup-step note. Since AGENTS.md is this project's AI-agent-facing doc and README.md is the human-facing entry point, a human contributor following README will still hit the exact undocumented "DFNR disabled" surprise this PR sets out to fix.

Not a reason to block this PR — it's still a correct, net-positive addition on its own — but worth a fast follow-up mirroring the same ~9 lines into README.md's build section (and optionally CONTRIBUTING.md), consistent with how this project has synced this kind of note across files before (e.g. #4261's AGENTS.md/CONTRIBUTING.md sync).

2. [Trivial, arguable] Minor placement nit — not worth acting on. The note reads "run … once before cmake" but is textually placed after the 3-line build-command block rather than before it. The existing bot review characterized the placement as correct, and I'd call this a genuine style toss-up rather than a real usability problem. Mentioning only for completeness.

Everything else checks out. Independently re-verified: both setup-deepfilter.sh/.ps1 exist; the quoted CMake message matches CMakeLists.txt:450 verbatim; all four named workflows genuinely invoke the script (gated behind a cache-hit check, not just mentioned); ENABLE_SPECBLEACH/NR4 is a fully independent CMake option from ENABLE_DFNR, so "NR still works without it" holds.

No blocking findings

Docs-only, zero functional/behavioral risk. CI fully green (build, check-macos, check-windows, accessibility, CodeQL JS/Python — no CodeQL cpp job for a docs-only change, expected). mergeable: MERGEABLE; mergeStateStatus: BLOCKED is just the branch-protection approval gate.

Checks run

Diffed against the live tree: script existence, exact CMake message text, all 4 workflow files' actual invocation, ENABLE_SPECBLEACH/ENABLE_DFNR independence, and cross-checked README.md/CONTRIBUTING.md for the same gap the PR claims to close. Reviews/reviewThreads/CI/mergeable refreshed a final time immediately before this comment.

Review limits

None of note — small enough diff that static verification against the actual scripts/CMake/CI files is fully conclusive.

Summary: 0 blocking findings. Safe to merge as-is. One real, non-blocking gap worth a fast follow-up: the same note doesn't yet exist in README.md, where a human (non-AI-agent) contributor would actually encounter this surprise.

🤖 Generated with Claude Code

[aethersdr-agent]

Nice, useful doc fix — this closes a real gap in the local dev path. I verified every claim against the tree and they all hold:

• scripts/setup/setup-deepfilter.sh and the .ps1 variant both exist under scripts/setup/.
• The quoted configure message matches CMakeLists.txt:450-451 verbatim, including the run ./scripts/setup/setup-deepfilter.sh before cmake to enable hint.
• The setup step is invoked in all four named workflows (ci.yml, macos-dmg.yml, appimage.yml, windows-installer.yml), so "CI runs it automatically, release binaries ship DFNR" is accurate.
• "NR still works via NR4 (libspecbleach)" checks out — AudioEngine.h:228 documents NR4 as the libspecbleach spectral path, and it's a separate build option from DFNR.

The note is placed correctly right above the "full dependency list is in README.md" line, and framing it as an optional local prereq (not a duplicated dependency) respects that boundary. Accurate, well-scoped, and correctly signed.

No blocking issues, no polish notes. Thanks for documenting this, @ten9876.

---

_{🤖 aethersdr-agent · cost: $1.0409 · model: claude-opus-4-8}