feat: add cloud render mode for zero-install video generation - #4

varg-ai/references/gateway-api.md (3)

343-351: missing language specifier on sse example block

same deal - the sse event format could use a language hint for better rendering.

quick fix

-```
+```text
 event: status
 data: {"job_id":"...","status":"rendering","started_at":"..."}

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@varg-ai/references/gateway-api.md` around lines 343 - 351, Update the SSE
example fences in gateway-api.md to include a language specifier (e.g., change
the triple-backtick blocks that show "event: status / data: {...}" to use
```text) so the SSE payloads render correctly; locate the example blocks that
contain the "event: status" lines and replace their opening backticks with
```text for each block.

---

372-376: missing language specifier on headers block

quick fix

-```
+```text
 X-RateLimit-Limit: 10
 X-RateLimit-Remaining: 9
 X-RateLimit-Reset: 2026-03-13T12:01:00.000Z

</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

Verify each finding against the current code and only fix it if needed.

In @varg-ai/references/gateway-api.md around lines 372 - 376, The fenced block
showing example headers is missing a language specifier; update the code fence
around the headers example in gateway-api.md to use a plain-text specifier
(e.g., change totext) so the headers block is rendered correctly as
plain text—locate the headers block with the lines "X-RateLimit-Limit: 10",
"X-RateLimit-Remaining: 9", "X-RateLimit-Reset: 2026-03-13T12:01:00.000Z" and
add the text language tag to the opening fence.

</details>

---

`253-255`: **missing language specifier on code block**

static analysis flagged this - adding a language identifier helps syntax highlighting and linters.

<details>
<summary>quick fix</summary>

```diff
-```
+```text
 https://render.varg.ai
 ```
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

Verify each finding against the current code and only fix it if needed.

In @varg-ai/references/gateway-api.md around lines 253 - 255, The fenced code
block containing the URL "https://render.varg.ai" is missing a language
specifier; update that fenced block in gateway-api.md by adding a language
identifier (e.g., "text") after the opening backticks so the block becomes
text ... to satisfy linters and enable proper syntax highlighting.

</details>

</blockquote></details>
<details>
<summary>varg-ai/scripts/setup.sh (2)</summary><blockquote>

`41-46`: **edge case: curl output parsing can break on malformed responses**

if the api returns an empty response or connection times out, `$RESPONSE` could be just "error" and `tail -n1` / `sed '$d'` might behave unexpectedly. the script handles this reasonably with the else branch on line 60, but the body extraction on line 46 could produce weird output.

meow, consider adding a timeout to curl and checking for the "error" case explicitly:

<details>
<summary>optional improvement</summary>

```diff
-    RESPONSE=$(curl -s -w "\n%{http_code}" \
+    RESPONSE=$(curl -s --max-time 10 -w "\n%{http_code}" \
       -H "Authorization: Bearer $VARG_API_KEY" \
       "https://api.varg.ai/v1/balance" 2>/dev/null || echo "error")
 
     HTTP_CODE=$(echo "$RESPONSE" | tail -n1)
+    if [ "$HTTP_CODE" = "error" ]; then
+      echo "  $(yellow '[WARN]') Could not reach gateway"
+    elif [ "$HTTP_CODE" = "200" ]; then
-    BODY=$(echo "$RESPONSE" | sed '$d')
-
-    if [ "$HTTP_CODE" = "200" ]; then
+      BODY=$(echo "$RESPONSE" | sed '$d')
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against the current code and only fix it if needed.

In `@varg-ai/scripts/setup.sh` around lines 41 - 46, The current parsing of
RESPONSE into HTTP_CODE and BODY can break if curl returns "error" or an empty
string; modify the setup to pass a timeout to curl (e.g., --max-time) and after
capturing RESPONSE (the variable in this diff) first check if RESPONSE equals
"error" or is empty before running HTTP_CODE=$(echo "$RESPONSE" | tail -n1) and
BODY=$(echo "$RESPONSE" | sed '$d'); if it's an error/empty response, set
HTTP_CODE and BODY to sensible defaults (or branch to the existing error
handling) to avoid unreliable tail/sed behavior while keeping the rest of the
script flow unchanged.
```

</details>

---

`50-56`: **`bc` dependency not guaranteed**

the script uses `bc` to convert cents to dollars, but `bc` isn't available on all systems (e.g., minimal docker images, some cloud shells). the fallback `|| echo "?"` handles it, but might confuse users seeing `$?` in output.

consider simplifying to just show cents, or document `bc` as optional:

<details>
<summary>suggested tweak</summary>

```diff
       if [ -n "$BALANCE" ]; then
-        DOLLARS=$(echo "scale=2; $BALANCE / 100" | bc 2>/dev/null || echo "?")
-        echo "  $(green '[OK]') Gateway connected. Balance: $BALANCE credits (\$$DOLLARS)"
+        echo "  $(green '[OK]') Gateway connected. Balance: $BALANCE credits"
       else
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against the current code and only fix it if needed.

In `@varg-ai/scripts/setup.sh` around lines 50 - 56, The script uses bc to compute
DOLLARS from BALANCE which may not exist on minimal systems; update the
BALANCE/DOLLARS handling so it first checks for bc (command -v bc) and only runs
the bc conversion when present, otherwise set DOLLARS to a clear fallback like
"N/A" or display only cents; adjust the echo lines that reference BALANCE and
DOLLARS (the variables BALANCE and DOLLARS and the green invocation) so the
message is unambiguous in both cases.
```

</details>

</blockquote></details>
<details>
<summary>varg-ai/references/templates.md (1)</summary><blockquote>

`51-65`: **polling loop could run forever on stuck jobs**

if the api returns a status that's neither "completed" nor "failed" indefinitely (or job times out server-side but status stays "rendering"), this loop never exits.

consider adding a max iteration count or timeout:

<details>
<summary>optional safeguard</summary>

```diff
 # Poll until status is "completed" or "failed"
+MAX_POLLS=90  # ~15 minutes at 10s intervals
+POLL_COUNT=0
 while true; do
+  POLL_COUNT=$((POLL_COUNT + 1))
+  if [ "$POLL_COUNT" -gt "$MAX_POLLS" ]; then
+    echo "Timeout waiting for job"
+    break
+  fi
   RESULT=$(curl -s "https://render.varg.ai/api/render/jobs/$JOB_ID" \
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against the current code and only fix it if needed.

In `@varg-ai/references/templates.md` around lines 51 - 65, The polling while true
loop that uses RESULT and STATUS can hang indefinitely if the job never reaches
"completed" or "failed"; modify the script to add a timeout safeguard by
introducing a max attempts or deadline variable (e.g., MAX_ATTEMPTS or END_TIME)
and a counter (ATTEMPT) or timestamp check inside the loop,
incrementing/checking it each iteration and exiting the loop with an error
message and non-zero status when the limit is reached; also ensure the exit
prints a helpful diagnostic (job id, last STATUS/RESULT) so callers can detect
timeout.
```

</details>

</blockquote></details>

</blockquote></details>

<details>
<summary>🤖 Prompt for all review comments with AI agents</summary>

Verify each finding against the current code and only fix it if needed.

Inline comments:
In @varg-ai/references/gateway-api.md:

• Around line 319-331: The status list under the render API currently omits
  "queued"; update the documented Status values to include "queued" alongside
  "rendering", "completed", and "failed" in the section containing Status values
  and the failure example, and ensure any related examples (the submit response
  example and the main gateway API references) consistently reflect "queued" where
  appropriate (search for the Status values block and the submit response example
  to locate the exact spots to change).

In @varg-ai/references/templates.md:

• Around line 38-42: The example that extracts JOB_ID uses jq but doesn't
  mention the dependency; update the templates.md example (the JOB_ID extraction
  using the curl + jq pipeline) to either add a short note that jq is required or
  provide an alternate fallback extraction (e.g., using basic shell tools) for
  environments without jq; reference the JOB_ID variable and the curl ... | jq -r
  '.job_id' pipeline so reviewers can locate and modify that snippet.

In @varg-ai/scripts/setup.sh:

• Around line 93-98: The ffprobe check only prints status and doesn't influence
  the HAS_FFMPEG/mode logic, so update the setup logic to treat ffprobe presence
  alongside ffmpeg: add a new boolean (or update existing HAS_FFMPEG) to reflect
  that both ffmpeg and ffprobe are required for local render mode, use the ffprobe
  check block to set that flag (e.g., set HAS_FFMPEG=false or a new
  HAS_FFMPEG_AND_FFPROBE=false when ffprobe is missing), and then use that
  combined flag in any mode determination or messaging; alternatively, if you
  prefer to keep separate flags, introduce HAS_FFPROBE and ensure mode logic
  checks both HAS_FFMPEG and HAS_FFPROBE and the echo message clearly states that
  both are required when one is missing.

In @varg-ai/SKILL.md:

• Around line 89-98: The cloud-render example mixes JSX compositional syntax
  with the documented pattern of using function calls for media (composition uses
  JSX but media should be created via functions), so update the example to use the
  Video(...) function-call form instead of JSX: replace the JSX element <Video ... /> with the corresponding Video(...) invocation and ensure model creation
  still uses fal.videoModel("kling-v3"); confirm the example keeps Render and Clip
  as composition wrappers while Video is shown as the media function to match the
  documented pattern.

---

Nitpick comments:
In @varg-ai/references/gateway-api.md:

• Around line 343-351: Update the SSE example fences in gateway-api.md to
  include a language specifier (e.g., change the triple-backtick blocks that show
  "event: status / data: {...}" to use text) so the SSE payloads render correctly; locate the example blocks that contain the "event: status" lines and replace their opening backticks with text for each block.
• Around line 372-376: The fenced block showing example headers is missing a
  language specifier; update the code fence around the headers example in
  gateway-api.md to use a plain-text specifier (e.g., change totext) so
  the headers block is rendered correctly as plain text—locate the headers block
  with the lines "X-RateLimit-Limit: 10", "X-RateLimit-Remaining: 9",
  "X-RateLimit-Reset: 2026-03-13T12:01:00.000Z" and add the text language tag to
  the opening fence.
• Around line 253-255: The fenced code block containing the URL
  "https://render.varg.ai" is missing a language specifier; update that fenced
  block in gateway-api.md by adding a language identifier (e.g., "text") after the
  opening backticks so the block becomes text ... to satisfy linters and
  enable proper syntax highlighting.

In @varg-ai/references/templates.md:

• Around line 51-65: The polling while true loop that uses RESULT and STATUS can
  hang indefinitely if the job never reaches "completed" or "failed"; modify the
  script to add a timeout safeguard by introducing a max attempts or deadline
  variable (e.g., MAX_ATTEMPTS or END_TIME) and a counter (ATTEMPT) or timestamp
  check inside the loop, incrementing/checking it each iteration and exiting the
  loop with an error message and non-zero status when the limit is reached; also
  ensure the exit prints a helpful diagnostic (job id, last STATUS/RESULT) so
  callers can detect timeout.

In @varg-ai/scripts/setup.sh:

• Around line 41-46: The current parsing of RESPONSE into HTTP_CODE and BODY can
  break if curl returns "error" or an empty string; modify the setup to pass a
  timeout to curl (e.g., --max-time) and after capturing RESPONSE (the variable in
  this diff) first check if RESPONSE equals "error" or is empty before running
  HTTP_CODE=$(echo "$RESPONSE" | tail -n1) and BODY=$(echo "$RESPONSE" | sed
  '$d'); if it's an error/empty response, set HTTP_CODE and BODY to sensible
  defaults (or branch to the existing error handling) to avoid unreliable tail/sed
  behavior while keeping the rest of the script flow unchanged.
• Around line 50-56: The script uses bc to compute DOLLARS from BALANCE which
  may not exist on minimal systems; update the BALANCE/DOLLARS handling so it
  first checks for bc (command -v bc) and only runs the bc conversion when
  present, otherwise set DOLLARS to a clear fallback like "N/A" or display only
  cents; adjust the echo lines that reference BALANCE and DOLLARS (the variables
  BALANCE and DOLLARS and the green invocation) so the message is unambiguous in
  both cases.

</details>

<details>
<summary>🪄 Autofix (Beta)</summary>

Fix all unresolved CodeRabbit comments on this PR:

- [ ] <!-- {"checkboxId": "4b0d0e0a-96d7-4f10-b296-3a18ea78f0b9"} --> Push a commit to this branch (recommended)
- [ ] <!-- {"checkboxId": "ff5b1114-7d8c-49e6-8ac1-43f82af23a33"} --> Create a new PR with the fixes

</details>

---

<details>
<summary>ℹ️ Review info</summary>

<details>
<summary>⚙️ Run configuration</summary>

**Configuration used**: Organization UI

**Review profile**: CHILL

**Plan**: Pro

**Run ID**: `7c58a480-1503-4686-b30c-bd160921eba8`

</details>

<details>
<summary>📥 Commits</summary>

Reviewing files that changed from the base of the PR and between 55ba125c3a916dc175961eb773488966bd243422 and ec99f6caa9b36b690bedf83292e3c84373262ec6.

</details>

<details>
<summary>📒 Files selected for processing (4)</summary>

* `varg-ai/SKILL.md`
* `varg-ai/references/gateway-api.md`
* `varg-ai/references/templates.md`
* `varg-ai/scripts/setup.sh`

</details>

</details>

<!-- This is an auto-generated comment by CodeRabbit for review status -->

varg-ai/references/templates.md (1)

5-5: transformation note could mention elevenlabs explicitly

line 5 says replace varg.imageModel → fal.imageModel and varg.videoModel → fal.videoModel "etc." but doesn't mention that speech/music also change providers in cloud mode (elevenlabs.speechModel vs varg.speechModel).

the "etc." covers it technically, but being explicit helps avoid confusion meow

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@varg-ai/references/templates.md` at line 5, The docs note replacing
varg.imageModel → fal.imageModel and varg.videoModel → fal.videoModel for cloud
mode but omit explicit mention of speech/music provider changes; update the
sentence to explicitly list speech/music replacements as well (e.g.,
varg.speechModel → elevenlabs.speechModel and any musicModel equivalents) so
readers know to switch varg.speechModel to elevenlabs.speechModel in cloud mode
alongside varg.imageModel→fal.imageModel and varg.videoModel→fal.videoModel.

varg-ai/SKILL.md (1)

42-42: "exactly the same" is slightly redundant but works for docs

static analysis flags "EXACTLY the same" as wordy. technically true since "exactly" implies sameness. could simplify to "exactly as-is" or "character-for-character identical" or "unchanged prompts must match exactly".

but for docs emphasizing a critical cache rule, the redundancy adds clarity. totally optional meow

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@varg-ai/SKILL.md` at line 42, Update the wording in the "3. **Cache is
sacred**" rule: replace the phrase "EXACTLY the same" with a less redundant but
clear alternative such as "character-for-character identical" or "exactly as-is"
so the line reads e.g. "When iterating, keep unchanged prompts
character-for-character identical." Keep the rest of the sentence and emphasis
intact to preserve the rule's intent.

varg-ai/references/cloud-render.md (1)

102-130: poll example assumes manual job_id extraction

the submit step (105-109) gets a response but doesn't show extracting job_id into a variable. then the poll step (128) just says "JOB_ID" expecting users to replace it manually.

this workflow break means users can't just copy-paste the commands in sequence. maybe show capturing the job_id in step 2?

something like:

RESPONSE=$(curl -s -X POST https://render.varg.ai/api/render \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"code\": $(cat video.tsx | jq -Rs .)}")

JOB_ID=$(echo "$RESPONSE" | jq -r '.job_id')
echo "submitted job: $JOB_ID"

then step 3 can just use $JOB_ID meow

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@varg-ai/references/cloud-render.md` around lines 102 - 130, The submit/poll
example is missing automatic extraction of the returned job_id; modify the Step
2 POST example to capture the full response into a shell variable (e.g.,
RESPONSE) and then extract JOB_ID using jq -r '.job_id' into a JOB_ID variable
(and echo it), and update the Step 3 poll example to use "$JOB_ID" instead of a
literal JOB_ID so users can copy/paste the sequence without manual replacement.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes