Pearl_Int

1. 1Give it the integration goal, not the implementation path — let it find the right endpoint
2. 2Always use the Keychain loader before handing it credential work: eval $(python3 scripts/ci/load_integration_env_from_keychain.py)
3. 3Check INTEGRATION_CREDENTIALS_REGISTRY.md first — the cred may already be documented
4. 4Let Pearl_Int own the first connection attempt; don't pre-wire endpoints based on guesses
5. 5If it returns a 401, ask it to check token scope before regenerating

# Wire a new webhook
act as pearl_int — set up a Metricool webhook for the social pack pipeline

# Validate an existing connection
pearl_int: validate the ElevenLabs connection and confirm the voice IDs are correct

# Credential lookup
pearl_int: what's the right key name for the DashScope free-quota endpoint?

# Diagnose a 429
pearl_int: we're getting 429s on the NVIDIA NIM chat endpoint — what's the rate limit and fallback?

• Building pipelines — that's Pearl_Dev
• Replacing Pearl_Research for API capability audits
• Being given hardcoded tokens instead of Keychain references