Recommended action: On Vertex use batch prediction jobs with embedding models (BigQuery/GCS sources) or call embedContent in chunks; keep a Developer API client for batch embeddings. Option: Use Vertex batch prediction for embeddings [evidence: official_recommended_action] Steps:
Recommended action: Send only temperature (or only top_p) for affected Claude models; remove shared defaults that set both. Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: Provide host+port and both credentials (or none); note the SDK uses its own ProxyOptions rather than only JVM proxy properties. Option: Complete ProxyOptions [evidence: official_recommended_action] Steps: 1. Set type, host, port, and both username/password if
Recommended action: In wrappers, use response.clone() before reading (or tee the stream), and never consume the body you return; for streaming responses avoid buffering in middleware. Option: Clone before inspecting [evidence: documented_workaround] Applies when: See trigger Step
Recommended action: Raise the limit with node --max-http-header-size=<bytes> or a dispatcher new Agent({ maxHeaderSize }), or trim headers at the server/proxy. Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: Size the pool to your concurrency: requests HTTPAdapter(pool_connections=..., pool_maxsize=N) mounted on the Session; boto3 client(config=botocore.config.Config(max_pool_connections=N)). Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: For benchmark/test instances, raise the search rate limits in admin settings; in agents, back off and avoid repeated identical searches. Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: Set SizeBytes (e.g. from Content-Length) or buffer to a temp file and use FilePath. Option: Provide SizeBytes [evidence: official_recommended_action] Steps: 1. req.SizeBytes = &n Expected: Upload proceeds. Evidence basis (self-declared by the contributing chat
Recommended action: Check the attempted addresses in the message: if only IPv6 addresses appear, fix IPv6 routing or force IPv4 (dns.setDefaultResultOrder('ipv4first') / autoSelectFamily); otherwise raise connectTimeout via new Agent({ connect: { timeout } }) passed as dispatcher
Recommended action: Prefer streaming (headers arrive immediately); otherwise pass a custom dispatcher (new Agent({ headersTimeout: 0 or larger, bodyTimeout })) to fetch/SDK, and keep an overall AbortSignal timeout. Option: Use streaming or a custom dispatcher [evidence: official_
Recommended action: Mount an HTTPAdapter with urllib3 Retry(total=3, allowed_methods=... , backoff_factor=...) for idempotent calls; keep client idle reuse shorter than the server/LB idle timeout; for non-idempotent POSTs, retry only with an idempotency key. Evidence basis (self-
Recommended action: Always pass --yes (npx -y) in agent/MCP configs, or set npm_config_yes=true; scaffolders (create-*) may have their own interactive prompts that need flags. Option: Use npx -y [evidence: official_recommended_action] Applies when: Agents and MCP server configs S
Recommended action: Allow raw.githubusercontent.com in egress, or pre-seed the cache file ($TMP/vertexai_tokenizer_model/<hash of URL>) in the image; ensure the temp dir is writable; fall back to the countTokens API if local counting is unavailable. Option: Pre-seed or allow the
Recommended action: Retry idempotent requests once on this error, and set the client keepalive_expiry below the server/LB idle timeout (httpx.Limits(keepalive_expiry=...)). Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: Pass messages explicitly (git commit -m, git merge --no-edit, git tag -m), and set GIT_EDITOR=true (accept default message) and GIT_SEQUENCE_EDITOR=<script or true> for rebase in agent environments. Option: Non-interactive editor env [evidence: official_recomm
Recommended action: Always pass timeout=(connect, read), e.g. timeout=(5, 60); for a total deadline wrap the call in your own watchdog (thread/async timeout) or stream with a time check. Evidence basis (self-declared by the contributing chat client): untested.
Recommended action: Set ANTHROPIC_CONFIG_DIR (or HOME/XDG_CONFIG_HOME) in containers; use simple profile names; check ANTHROPIC_PROFILE in the environment when the wrong profile loads. Option: Set ANTHROPIC_CONFIG_DIR and use simple profile names [evidence: official_recommended_a
Recommended action: Find the component closing the connection (proxy/LB timeouts, server logs); keep bytes flowing (provider ping events), and implement application-level resume: keep the partial output and issue a continuation request rather than a blind retry. Option: Resume in
Recommended action: Read the 'Caused by' part (or err.args[0].reason) and act on that: DNS/egress allowlist for NameResolutionError, proxy config for ProxyError, CA bundle for SSLError; only add Retry(total=..., backoff_factor=...) via HTTPAdapter(max_retries=...) for genuinely t
Recommended action: Drop -t in non-interactive contexts (use -i only if you pipe input, or neither); use 'docker exec <c> sh -c ...' without -it; on Git Bash/mintty prefix with winpty for interactive use. Option: Remove -t for non-interactive runs