Monitor and auto-retry failed tool calls in agent workflows

Wraps external tool calls with a reliability layer: before every call it validates that required parameters are present and correctly typed, checks the tool isn't already marked unreliable from prior failures, and confirms request size is within known limits. When a call fails, it classifies the failure into one of nine categories and applies a matched recovery: truncated JSON is re-fetched with pagination or smaller chunks, an API timeout gets one retry with a simpler request then decomposition, a 429 rate limit gets exponential backoff up to 3 retries, expired auth is flagged for user intervention, a mid-chain break resumes from the last successful checkpoint rather than restarting, an error disguised as a 200 success response is detected and treated as a real failure, a schema mismatch triggers auto-coercion with a warning if it's lossy, a network failure gets jittered retries up to 2 attempts, and anything unclassified is logged with full context and escalated to the user. For multi-step tool chains it maintains checkpoints, so a failure at step 4 of 7 resumes from step 4 instead of starting over. It also tracks failure patterns per tool, and after 3 or more failures of the same type on the same tool it marks that tool unreliable and suggests alternatives.

Use it when tool calls return truncated or malformed JSON, when APIs timeout or rate-limit an agent mid-task, when a multi-step chain breaks partway through, when automatic retry logic is needed without writing it by hand, or for any agent workflow that depends on external tool reliability. Best practice is to let it wrap all external tool calls automatically, review its reliability reports to spot flaky tools, and use checkpoint recovery for long chains; do not disable its retry logic for rate-limited APIs, and do not ignore repeated failure warnings about a specific tool.

Input is the outgoing tool call plus its parameters; output is either the successful result or a classified, recovered response following the failure-type-specific action above. Install:

npx skills add christopherlhammer11-ai/tool-use-guardian

Distributed through the Genesis Agent Marketplace and installed via npx skills add; pairs with a related long-context memory skill, also free from the same marketplace.

Anyone building agent workflows that call external tools or APIs and need automatic retry, chain-checkpoint recovery, and per-tool reliability tracking instead of hand-writing that resilience logic themselves.