[cursor]

Regenerated examples after elevenlabs/skills#131 ([agents] update structured procedures skills).

Generate summary

0 generated · 17 verified · 2 updated · 0 failed

First pass (pnpm generate, 42m 50s): 0 generated · 14 verified · 2 updated · 3 failed

The three ElevenAgents examples failed immediately with Argument list too long. Linux MAX_ARG_STRLEN is 128KiB, and the concatenated agents skill plus references now exceed that when passed as a single argv. After feeding the prompt on stdin, the retry succeeded:

• agents/expo/quickstart — verified, no changes (6m 12s)
• agents/nextjs/guardrails — verified, no changes (6m 39s)
• agents/nextjs/quickstart — verified, no changes (4m 9s)

Verified with no changes: dubbing, music (Next.js / Python / TypeScript), sound-effects (Next.js / Python / TypeScript), speech-to-text Python and TypeScript, text-to-speech (Next.js / Python / TypeScript), voice-changer, voice-isolator, and all three agents examples.

Directories changed (git diff --stat)

• scripts/generate-examples.sh — pass the Cursor prompt on stdin so large skill files do not hit MAX_ARG_STRLEN
• speech-engine/nextjs/quickstart — allow restart after a voice error and show a disconnected status
• speech-to-text/nextjs/realtime — stop caching the single-use Scribe token, surface token/connect errors, and flatten the realtime UI

Are the changes meaningful?

Yes.

• The generate-script stdin change is the important one. Without it, agents examples cannot be regenerated now that the skill pack is larger than 128KiB.
• The speech-engine update is a small, real error-recovery fix.
• The speech-to-text realtime updates mix useful token/error handling with a typical generator UI rewrite. Worth a quick review, but they match previous regeneration PRs.
• Skills PR 131 was about structured procedures. None of the existing example prompts cover procedures, so the agents examples correctly stayed unchanged.

  

[cursor]

PR Summary

Low Risk
Dev-tooling change only; behavior is the same except large prompts no longer fail at exec time.

Overview
Fixes example regeneration when inlined skill context exceeds Linux’s 128KiB per-argument limit (Argument list too long), which was blocking agents examples after the skill pack grew.

Instead of appending the full preamble + skills + PROMPT.md to the cursor agent argv, generate-examples.sh writes the prompt to a temp file and runs the agent with that file redirected to stdin via bash -c 'exec "$@" < "$0"', still wrapped in the existing run_with_timeout flow. A trap removes the temp file on exit.

^{Reviewed by Cursor Bugbot for commit db664a6. Bugbot is set up for automated code reviews on this repo. Configure here.}