Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

docs: each gate names what is wrong, so neither repair is a guess

rpone1b49cbf95e3e
diff --git a/.agents/skills/hypit/references/reference-video/final-sources.md b/.agents/skills/hypit/references/reference-video/final-sources.md
index 45390fe6..0f2b5d22 100644
--- a/.agents/skills/hypit/references/reference-video/final-sources.md
+++ b/.agents/skills/hypit/references/reference-video/final-sources.md
@@ -45,12 +45,22 @@ node --import tsx .agents/skills/hypit/scripts/preview-check.mjs path/to/main.sv
```
It reports every track that failed to build, and every track waiting on a Provider it cannot reach,
-and exits non-zero when either exists. A track that cannot be built is not done, and this is not the
+and exits non-zero when either exists. The failures it reports are the Producer's own messages — the
+track name, the Producer, and why it failed, down to the offending value (a schedule frame outside a
+window, a media edge whose artifact is not an image). This is not a guessing problem: the error says
+what is wrong, and you repair that. A track that cannot be built is not done, and this is not the
loop: the two-attempt ceiling governs how *well* a buildable element is tuned to the reference; it
does not govern whether the element builds at all. Every failure this check reports must be repaired
until the check passes — the author should never open the Playground and find that something they
were delivered cannot be seen. Repair as many times as the failure needs, then re-run the check.
+What this check cannot see is equally important: a track that *builds* but looks wrong — a typeface
+that does not match, a colour that is off, a shape that is misplaced — reports no error here, because
+nothing failed. That is the loop's work, under `reconstruction-loop.md`, where Gemini compares the
+rendered element against the reference and names the visible differences. Build failures are
+repaired here until they pass; appearance differences are resolved there, bounded by the two-attempt
+ceiling.
+
Fix package resolution and package implementation before repairing source use. Continue until all
three files are accepted. Do not create `check_svml_project` or another wrapper.
diff --git a/.agents/skills/hypit/references/reference-video/reconstruction-loop.md b/.agents/skills/hypit/references/reference-video/reconstruction-loop.md
index 0304b3dc..b612773b 100644
--- a/.agents/skills/hypit/references/reference-video/reconstruction-loop.md
+++ b/.agents/skills/hypit/references/reference-video/reconstruction-loop.md
@@ -10,6 +10,13 @@ element *builds at all* is a different gate with a different rule: `preview-chec
repaired until they pass without any attempt ceiling. A build failure is not a difference to weigh;
it is work that is not done.
+The difference between the two gates is what each reports. `preview-check` names the build failure —
+the Producer and why it failed, down to the offending value — so its repairs are not guessing. This
+loop names the appearance difference: `compare_reconstruction` sends the rendered element and the
+reference frame to Gemini and returns what differs in words, and a repair aims at a difference the
+comparison named. Neither gate expects the agent to guess; each one tells you what is wrong, and you
+repair that.
+
## The reference side is finished
The reference video does not change, and its observations are cached. Re-observing at a fixed