docs(learnings): silenceTimeoutSeconds min=10 + structuredOutput.name 40-char limit + gitignore pnpm-lock

Two field-level Vapi API validation constraints that fail at push time
and aren't documented in the gitops engine README. Surfaced during a
gitops-mudflap iteration where both blocked a deploy mid-push:

- silenceTimeoutSeconds: assistant config minimum is 10s; setting < 10
  fails with "must not be less than 10". Tradeoff implications documented
  for the "end call immediately" pattern.
- structuredOutput.name: 1-40 character limit on inline scenario
  evaluations and standalone structured outputs. Long descriptive names
  like assistant_left_voicemail_and_ended_call_promptly (48) silently
  exceed the limit. Examples of compact alternatives included.

Also gitignoring pnpm-lock.yaml — repo uses npm with package-lock.json,
but stray pnpm-lock.yaml files appear when contributors accidentally run
pnpm install. Same paper cut affects every customer fork.

Author: dhruva-vapi

.gitignore

@@ -1,5 +1,8 @@
 # Dependencies
 node_modules/
+# This repo uses npm (package-lock.json). Stray pnpm-lock.yaml files
+# can appear if someone accidentally runs `pnpm install` — never commit them.
+pnpm-lock.yaml
 
 # Environment files (secrets - never commit these!)
 # Covers dev/stg/prod and any org slug (e.g. .env.roofr-production)

docs/learnings/assistants.md

@@ -409,6 +409,17 @@ If a hook references a `toolId` that doesn't exist, Vapi logs a warning and cont
 
 `customer.speech.timeout` (hook) and `silenceTimeoutSeconds` (assistant) are separate mechanisms. The hook fires an action; the timeout ends the call. Configure them independently.
 
+### `silenceTimeoutSeconds` minimum is 10
+
+The Vapi API enforces a hard minimum of **10 seconds** on `silenceTimeoutSeconds`. Setting this field to anything less than 10 (e.g., `5` or `8`) will fail at push time with:
+
+```
+PATCH /assistant/<id> → 400
+silenceTimeoutSeconds must not be less than 10
+```
+
+The minimum is not documented in the gitops engine README and is only surfaced when you POST/PATCH. If you need an "end the call almost immediately" pattern (e.g., a voicemail-leaver that should fire `end_call` right after delivering its request-start message), drive that behavior through the assistant's prompt + `firstMessageMode: assistant-speaks-first-with-model-generated-message` + `endCallFunctionEnabled: true` so the model fires the end-call tool on activation. The 10-second timeout is then just a safety net, not the primary exit path.
+
 ### Available hook events
 
 - `call.ending` — call is about to end

docs/learnings/structured-outputs.md

@@ -21,6 +21,17 @@ Primitive schemas (`string`, `boolean`, `number`) are automatically wrapped in a
 
 The result is unwrapped before being returned. This is transparent but can cause confusion if you inspect raw API calls.
 
+### `structuredOutput.name` is limited to 1-40 characters
+
+The Vapi API enforces a hard 40-character maximum on the `name` field of any structured output (including inline structured outputs used inside scenario evaluations). Names longer than 40 chars fail at push time with:
+
+```
+POST /eval/simulation/scenario → 400
+evaluations.5.structuredOutput.Name must be between 1 and 40 characters
+```
+
+Long, descriptive evaluator names like `assistant_left_voicemail_and_ended_call_promptly` (48 chars) or `assistant_detected_hostile_recording_and_ended_call` (51 chars) will silently exceed the limit until you POST. Keep names compact (`assistant_ended_call_after_message`, `assistant_handled_hostile_recording`) and put the descriptive nuance in the `description` field, which has no length cap. The constraint applies to the field on every structured output type — both standalone resources and inline evaluations within scenarios.
+
 ---
 
 ## assistant_ids Must Be UUIDs