| name | optics-scope |
| description | Scope an optical-simulation request honestly — map "can this bench simulate X?" to a tier ((a) reproducible now / (b) via wave.py single-plane FFT-PSF / (c) needs an engine we lack), then cite the oracle-verified case or name the missing engine + the external tool to use. Use when asked "can it simulate …", "is X in scope", "what can/can't this engine do", "do you support FDTD / BPM / Monte-Carlo / hologram reconstruction / turbulence / g²", or "/optics-scope". Anti-overclaim only: this adds ZERO simulation capability.
|
optics-scope: route the request, never overclaim
This skill answers "can this bench simulate X?" — it does not simulate anything new. Its job is to
place a request in the right tier and back the answer with proof: either the verified case that shows we
do it, or the engine we deliberately lack plus the external tool you'd use. The source of truth is the
vendored case index docs/OPTICS_SCOPE.md, mirrored machine-readably in capabilities()['scope_map'].
Workflow
- Read the source of truth — open
docs/OPTICS_SCOPE.md (the in-repo case index: tier
definitions, the request→tier table, the oracle-verified whitelist). If the file is absent, fall back
to capabilities()['scope_map']. Do not answer from memory — the numbers and the missing-engine
list live in the file, not in your head.
- Classify the request into a tier — using the table:
- (a) reproducible now — ray/align, Gaussian-
q (w(z)/M²/resonator), thin-lens/prism/grating,
Fresnel/Malus, dispersion/Abbe, thermal lensing (opt-in estimate). The live engine already does it.
- (b) via
wave.py — a single focal-plane / Fraunhofer PSF |FFT(pupil·exp(i·2πW))|²: Airy,
Strehl, MTF, encircled energy, Maréchal penalty. One plane only — no inter-plane propagation.
- (c) needs an engine we lack — full-wave FDTD/RCWA, multi-plane Fresnel/angular-spectrum/BPM,
split-step NLSE/solitons, Monte-Carlo tissue transport, digital-hologram reconstruction,
multi-screen turbulence, quantum statistics (g²/HOM/squeezing).
- Back the answer with proof — for (a)/(b), cite the matching row and, where relevant, the
oracle-verified case from the whitelist (it's asserted in
tests/test_validation.py). For (c), state
plainly that we don't have the engine and name the canonical external tool from the file
(Meep/Lumerical/Tidy3D, POPPY/diffractio, gnlse, MCML, QuTiP) — "use X for this".
- Report the scope verdict — tier + the cited case or the named missing engine. If the request
spans tiers (e.g. "show the PSF and propagate it through turbulence"), split it: (b) for the PSF,
(c) for the turbulence propagation. Don't average a mixed request into one tier.
Disciplines
- Never overclaim a (c) capability. The modal AO loop here is a 15-Zernike caricature, not
multi-screen turbulence; a single FFT-PSF is not multi-plane propagation. Name the gap; don't paper a
(c) request over with an (a)/(b) overlay.
- Only the whitelist counts as verified. Numbers in the Obsidian catalog are book-cited, NOT
oracle-verified — never quote them as verified. A value becomes quotable only after it's wired into
tests/test_validation.py and passes the oracle (the file is the count; don't hard-code a number).
- Degrade gracefully. Vault unreachable headless → use
docs/OPTICS_SCOPE.md. Doc absent →
capabilities()['scope_map']. Cite which source you used.
- This skill routes and cites — it does not invent a
can_simulate() helper or claim a parity test that
doesn't exist. The proof is the verified case in the suite, not a fabricated check.
