feat(skills): remotion-to-hyperframes references (6/7)

[jrusso1020]

What

Eleven reference files under skills/remotion-to-hyperframes/references/ that the skill loads progressively (per skill-creator's design pattern) when translating a specific Remotion idiom.

File	What it covers	Lines
api-map.md	The authoritative translation table (the index loaded first)	137
timing.md	interpolate, spring (validated configs), easing, count-up, stagger	160
sequencing.md	Sequence, Series, Loop, Freeze, AbsoluteFill, Composition root	193
media.md	Audio, Video, Img, IFrame, OffthreadVideo, staticFile, asset paths	152
transitions.md	@remotion/transitions presentations → manual GSAP crossfades or HF shader-transitions	93
lottie.md	@remotion/lottie → HF lottie adapter, AE feature limitations	118
fonts.md	Google Fonts, local @font-face, system fallback noise floor	110
parameters.md	Zod schemas, defaultProps, sync vs async calculateMetadata	164
escape-hatch.md	When to bow out + the runtime interop pattern from PR #214	116
limitations.md	Known caveat patterns (volume ramps, Loop with state, custom presentations)	131
eval.md	How to run the validation harness, threshold rule of thumb, noise floor	137

Total: ~1500 lines across 11 files, every file under 200 lines (skill-creator's progressive-disclosure budget).

Why

#6 in the 7-PR stack.

The references are the skill's working memory — SKILL.md (PR 7) is intentionally lean (a 5-step workflow) and links into these files as needed. When the agent encounters a spring() it loads timing.md; when it sees <TransitionSeries> it loads transitions.md. The agent never loads everything at once — that's the whole point of the progressive-disclosure pattern.

How

Evidence-driven, not speculative:

• Every spring → back.out(N) mapping in timing.md is from the validated T2 / T3 renders (mean SSIM 0.985 / 0.953).
• Every threshold and noise-floor number in eval.md is from the calibration runs in PR 3 / PR 4.
• Every blocker rule in escape-hatch.md matches a lint_source.py rule and a T4 case.
• The pix_fmt + colorspace gotcha in eval.md (setVideoImageFormat("png") + setColorSpace("bt709")) is the empirical fix that bumped T1 from 0.958 → 0.974 during validation.

Stack

#506 (1/7) — scaffold
#507 (2/7) — eval harness
#508 (3/7) — T1 + T2 fixtures
#509 (4/7) — T3 data-driven fixture
#515 (5/7) — T4 escape-hatch fixtures
this PR (6/7) — references/*.md
7/7 — SKILL.md body + corpus orchestrator

Test plan

• All 11 references pass oxfmt --check
• Cross-references between files use relative links (e.g. [timing.md](timing.md))
• Every external link points at a real artifact (PR feat(core): add Remotion adapter for framework interoperability #214, hyperframes-vs-remotion guide, lottie/airbnb feature list, hyperframes packages)
• No reference file > 200 lines (skill-creator progressive-disclosure budget)
• Replaces the .gitkeep from PR 1's scaffold
• Final wiring into SKILL.md body — that's PR 7's job

[miguel-heygen]

The references still disagree on the lambda-only case, so the stack does not give agents one stable decision rule for when to stop versus translate the remainder.

[miguel-heygen]

This edge case conflicts with the policy above: lines 10-23 say any blocker, including r2hf/lambda-import, should recommend interop because translating blockers produces silently-wrong output. api-map.md repeats the same any-blocker rule. If lambda-only is intended to be translatable, make it a warning/special action consistently across the linter references and final skill; if it is a true blocker, remove this edge case and update T4 case 05.

[jrusso1020]

@miguel-heygen — addressed in the amended commit 662c4dbd:

The lambda-only "Edge case" section in escape-hatch.md is gone. Lambda is now in the regular "When NOT to bow out: warnings only" table, alongside delayRender, useCallback, etc:

Rule	Action
r2hf/lambda-import	drop the @remotion/lambda config; HF runs single-machine, log gap
...	...

Plus a short paragraph below the table explaining why lambda is a warning (deployment config, orthogonal to rendered composition).

api-map.md distributed-rendering table updated:

• Lambda imports → "drop the import (warning r2hf/lambda-import)"
• renderMediaOnLambda(...) → "drop the call; note in TRANSLATION_NOTES.md"
• A header paragraph above the table now says these are warnings, not blockers.

limitations.md "React patterns the skill refuses" list: removed @remotion/lambda (no longer a refused pattern), added a follow-up paragraph pointing at escape-hatch.md for the warning-level treatment.

The blocker rules table at the top of escape-hatch.md now lists 5 rules instead of 6 (lambda removed). All three references give the same any-blocker decision rule for the agent.

[miguel-heygen]

Re-reviewed the latest head. The references now consistently treat @remotion/lambda as a warning/deployment gap rather than a blocker, matching T4 and the final skill. No remaining blockers on this layer.