Exa Reliability and Retry Design
Overview
Design bounded Exa retries, deadlines, partial-result behavior, and tested fallbacks by endpoint and failure class. Treat credentials, queries, retrieved content, generated output, spend, and destructive state as separately governed boundaries.
Prerequisites
- The target repository, environment, Exa team, product surface, and accountable owner.
- The workload's data classification, latency and freshness promise, cost ceiling, and retention policy.
- Current first-party documentation plus credentials only for a narrowly approved live check.
Current Contract
Retryability depends on status and operation. Invalid 400, auth, policy, and billing failures require repair; 429 honors Retry-After; documented transient server or overload failures can use capped jitter. Asynchronous creates and destructive calls require identity and state checks before replay.
Authentication
For normal REST work, inject EXA_API_KEY from an approved server-side secret manager and send it only as Authorization: Bearer to the configured first-party Exa API host. Team Management service keys, hosted MCP OAuth or enterprise managed authorization, and payment-protocol calls are separate trust models. Never print, commit, place in a URL, or expose a credential to an untrusted client.
Instructions
- Classify each operation as read, create, update, trigger, cancel, stop, or delete.
- Assign an end-to-end deadline and bounded attempt budget per operation.
- Retry only documented transient classes and preserve request or resource IDs.
- Define partial Contents behavior and freshness fallback explicitly.
- Choose a tested degrade path that preserves source attribution and policy.
- Exercise timeout, duplicate delivery, stale cache, overload, and recovery in tests.
Tool Discipline
Use Read, Glob, and Grep to inspect repository code, configuration, fixtures, and evidence. Use Write and Edit only for approved implementation or documentation changes. Do not call Exa, run paid research, create or alter a Monitor, Webset, Agent run, Batch, team, member, API key, budget, webhook, or deployment merely because this skill was invoked.
Approval Boundaries
Require an accountable owner before live queries involving sensitive intent, production credentials, spend or rate-limit changes, forced live crawling, generated summaries, external delivery, deployment, member or key changes, schedule creation, or destructive cancellation, stopping, deletion, or revocation. Read-only repository inspection and synthetic offline validation do not authorize live vendor actions.
Failure Modes
- Retrying a create after an ambiguous timeout can duplicate paid work.
- Serving cache-only content can violate a freshness promise unless disclosed.
- A fallback search provider changes quality, data handling, and cost boundaries.
Output
Return the operation scope, environment, team and product surface, authorization class, contract and policy decisions, deterministic validation results, content-free identifiers, status and cost counts, risks, cleanup or rollback state, and a concise pass or fail receipt. Exclude credentials, raw queries, prompts, presigned URLs, retrieved content, generated output, and customer-derived data unless separately approved.
Example
- Retry a 503 read with capped jitter, but reconcile an Agent run ID before deciding whether to submit another run.
- Finish with request or resource IDs, assertion counts, cost and terminal state, rollback or deletion status, and the decision owner; never reproduce secrets or retrieved content.
Validation
Rerun the smallest relevant deterministic test, compare actual behavior with the requested outcome and current first-party contract, verify sensitive fields are absent from evidence, and confirm deadlines, terminal state, downstream retention, and rollback before reporting success.
References
Review the dated first-party evidence map before relying on any endpoint, parameter, search type, price, limit, beta, compliance, identity, retry, or lifecycle claim.