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.

fix: a sweep waits for the whole-reference observations that seed it

rponeb1c5c25ea51c
diff --git a/.agents/skills/hypit/references/reconstruction/index.md b/.agents/skills/hypit/references/reconstruction/index.md
index 39186066..a540337e 100644
--- a/.agents/skills/hypit/references/reconstruction/index.md
+++ b/.agents/skills/hypit/references/reconstruction/index.md
@@ -122,7 +122,8 @@ hypit-reference-video-tools observe_reference --reference-id <id>
```
On the `agent` observer both commands return immediately with the observations they owe, and reading
-the files below is what you do between answering them.
+the files below is what you do between answering them. Answer `prepare_reference`'s four first:
+`observe_reference` quotes them into every shot's question and refuses to run until they exist.
Then read the files below while the sweep is working, and collect its result when you are done.
diff --git a/.agents/skills/hypit/references/reconstruction/observers.md b/.agents/skills/hypit/references/reconstruction/observers.md
index c45f5c5c..a93883c0 100644
--- a/.agents/skills/hypit/references/reconstruction/observers.md
+++ b/.agents/skills/hypit/references/reconstruction/observers.md
@@ -135,7 +135,15 @@ hypit-reference-video-tools observe_reference --reference-id <id>
```
Both return `pending_observations`: a list of tasks, each with its `key`, its `instruction`, its
-`prompt` and the `image_refs` to read. Answer one, then record it:
+`prompt` and the `image_refs` to read.
+
+**Answer `prepare_reference`'s four before running `observe_reference`.** Every shot observation quotes
+the whole-reference evidence into its own question, so a sweep run before those four exist asks each
+shot with less than it should have and caches the answer. `observe_reference` refuses until they are
+recorded and names the ones it is waiting for; a narrow `--question` is exempt, being a follow-up
+rather than the sweep.
+
+Answer one, then record it:
```text
hypit-reference-video-tools record_observation --reference-id <id> --key visual:shot-002 \
diff --git a/packages/reference-video-tools/src/tools.ts b/packages/reference-video-tools/src/tools.ts
index d9333115..18f43822 100644
--- a/packages/reference-video-tools/src/tools.ts
+++ b/packages/reference-video-tools/src/tools.ts
@@ -534,6 +534,16 @@ export function createReferenceVideoTools(options: ToolOptions = {}): ReferenceV
pending_observations: pending,
};
}
+ // Every shot observation is asked with the four whole-reference observations quoted into it. The
+ // observer that answers in band has them the moment prepare_reference returns; the one that
+ // answers out of band has not written them yet, and a sweep run first would ask every shot its
+ // question with that context missing and cache the answers.
+ if (observer === "agent") {
+ const owed = WHOLE_REFERENCE_KEYS.filter((key) => state[key]?.status !== "complete");
+ assert(owed.length === 0,
+ `answer the whole-reference observations first, with record_observation: ${owed.join(", ")}. `
+ + "Every shot observation quotes them, so a sweep run before they exist asks each shot with less than it should have.");
+ }
const selectedIds = new Set(selected.map((shot) => shot.shot_id));
const boundaryRights = state.shots.filter((shot) => {
const left = state.shots.find((candidate) => candidate.index === shot.index - 1);