https://metaharnessscripture.com/ Practical essays on designing, evaluating and tuning meta harnesses for LLM agents: scaffolds, evals, prompts, tool policy and loop architecture. en https://metaharnessscripture.com/posts/agent-tool-authorization/ https://metaharnessscripture.com/posts/agent-tool-authorization/ Thu, 15 Oct 2026 00:00:00 GMT Design agent tool authorization around each effect, not a cached tool list. Bound delegated scopes, enforce revocation, and recheck long-running sessions. agent harness AI agent tools agent guardrails agent reliability https://metaharnessscripture.com/posts/agent-cancellation/ https://metaharnessscripture.com/posts/agent-cancellation/ Wed, 14 Oct 2026 00:00:00 GMT Design agent cancellation around durable intent, blocked dispatch, and effect reconciliation. Show what stopped, what completed, and what stays uncertain. agent harness agent reliability agentic systems tool design https://metaharnessscripture.com/posts/agent-evaluation-uncertainty/ https://metaharnessscripture.com/posts/agent-evaluation-uncertainty/ Tue, 13 Oct 2026 00:00:00 GMT Measure agent evaluation uncertainty at the task level. Separate repeated trials from evidence, compare releases in pairs, and keep inconclusive verdicts. agent evaluation agent evals release gates agent reliability https://metaharnessscripture.com/posts/agent-memory-poisoning/ https://metaharnessscripture.com/posts/agent-memory-poisoning/ Mon, 12 Oct 2026 00:00:00 GMT Prevent agent memory poisoning with write admission, source provenance, correction chains, and quarantine. Keep retrieved claims from becoming authority. agent memory agent harness agent guardrails agent reliability https://metaharnessscripture.com/posts/agent-tool-schema-versioning/ https://metaharnessscripture.com/posts/agent-tool-schema-versioning/ Thu, 08 Oct 2026 00:00:00 GMT Use agent tool schema versioning to catch silent changes in defaults, units, outputs, and authority. Test migrations and keep rollback tied to effects. agent harness AI agent tools agent reliability schema versioning https://metaharnessscripture.com/posts/agent-concurrency-control/ https://metaharnessscripture.com/posts/agent-concurrency-control/ Wed, 07 Oct 2026 00:00:00 GMT Build agent concurrency control with revision checks, leases, and fencing. Reject stale workers at the write boundary and test races before they hit users. agent harness agentic systems agent reliability concurrency control https://metaharnessscripture.com/posts/agent-evaluation-graders/ https://metaharnessscripture.com/posts/agent-evaluation-graders/ Tue, 06 Oct 2026 00:00:00 GMT Audit agent evaluation graders before trusting a passing score. Separate state checks from model judgment, calibrate false passes, and version each rubric. agent evaluation agent evals LLM judges agent reliability https://metaharnessscripture.com/posts/agent-memory-retrieval/ https://metaharnessscripture.com/posts/agent-memory-retrieval/ Mon, 05 Oct 2026 00:00:00 GMT Design agent memory retrieval around provenance, freshness, and conflicts. Verify sources before acting, and abstain when recall cannot support a claim. agent memory agent harness context engineering agent reliability https://metaharnessscripture.com/posts/production-agent-architecture/ https://metaharnessscripture.com/posts/production-agent-architecture/ Thu, 01 Oct 2026 00:00:00 GMT Build production agent architecture around queues, durable state, scoped executors, and verifiers that distinguish a finished attempt from a verified task. agent harness agentic systems agent reliability agent architecture https://metaharnessscripture.com/posts/agent-context-window-management/ https://metaharnessscripture.com/posts/agent-context-window-management/ Wed, 30 Sep 2026 00:00:00 GMT Agent context window management preserves evidence across compaction. Set budgets, retain references, and rehydrate facts before consequential actions. context engineering agent harness agent memory agent reliability https://metaharnessscripture.com/posts/agent-evaluation-dataset-design/ https://metaharnessscripture.com/posts/agent-evaluation-dataset-design/ Tue, 29 Sep 2026 00:00:00 GMT Build agent evaluation datasets around evidence and failure boundaries. Separate difficulty from noise, prevent leakage, and test shortcuts before release. agent evaluation agent evals datasets agent reliability https://metaharnessscripture.com/posts/agent-recovery-checkpoints/ https://metaharnessscripture.com/posts/agent-recovery-checkpoints/ Mon, 28 Sep 2026 00:00:00 GMT Build agent recovery checkpoints that preserve intent, reconcile uncertain effects, and revalidate authority before interrupted work resumes after a crash. agent reliability agent harness agent recovery durable execution https://metaharnessscripture.com/posts/agent-guardrail-red-teaming/ https://metaharnessscripture.com/posts/agent-guardrail-red-teaming/ Thu, 24 Sep 2026 00:00:00 GMT Test agent guardrails against forbidden effects, not reassuring replies. Build replayable attacks, verify state, and keep legitimate tasks in test suites. agent guardrails agent evals prompt injection agent reliability https://metaharnessscripture.com/posts/agent-cost-optimization/ https://metaharnessscripture.com/posts/agent-cost-optimization/ Wed, 23 Sep 2026 00:00:00 GMT AI agent cost optimization cuts repeated context and tool overhead while preserving verified outcomes. Measure savings at the task boundary, not per call. agent cost agent harness context engineering agent tools agent evaluation https://metaharnessscripture.com/posts/multi-agent-failure-modes/ https://metaharnessscripture.com/posts/multi-agent-failure-modes/ Tue, 22 Sep 2026 00:00:00 GMT Catch multi-agent failure modes at handoffs: reject stale results, resolve state conflicts, and give final writes one owner with a checked recovery path. multi-agent systems agent harness agent reliability agent evals https://metaharnessscripture.com/posts/agent-tool-call-errors/ https://metaharnessscripture.com/posts/agent-tool-call-errors/ Mon, 21 Sep 2026 00:00:00 GMT Handle agent tool call errors by separating invalid requests, execution failures, and unknown outcomes. Recover safely without repeating external effects. agent reliability agent tools agent harness agent retries https://metaharnessscripture.com/posts/multi-agent-orchestration/ https://metaharnessscripture.com/posts/multi-agent-orchestration/ Mon, 14 Sep 2026 00:00:00 GMT Design multi-agent orchestration with independent tasks, evidence-backed handoffs, and one owner of final effects; weigh gains against coordination cost. multi-agent orchestration agent harness agent reliability agent evaluation https://metaharnessscripture.com/posts/agent-framework-comparison/ https://metaharnessscripture.com/posts/agent-framework-comparison/ Wed, 09 Sep 2026 00:00:00 GMT An agent framework comparison should evaluate the harness surface: state, tools, control flow, observability, and safe promotion—not feature counts. agent frameworks agent harness agent evaluation agent reliability agentic systems https://metaharnessscripture.com/posts/agent-regression-gates/ https://metaharnessscripture.com/posts/agent-regression-gates/ Tue, 08 Sep 2026 00:00:00 GMT Agent regression gates turn known failures into release blockers, so every repaired incident becomes a durable test before the next deployment. agent regression testing agent evals release gates agent reliability meta harness https://metaharnessscripture.com/posts/sandbox-and-blast-radius/ https://metaharnessscripture.com/posts/sandbox-and-blast-radius/ Mon, 07 Sep 2026 00:00:00 GMT Agent sandboxes limit permissions, isolate side effects, and make recovery explicit so a bad run stops locally instead of becoming an incident. agent sandbox blast radius agent permissions agent safety agent harness https://metaharnessscripture.com/posts/mcp-tool-protocol-wars/ https://metaharnessscripture.com/posts/mcp-tool-protocol-wars/ Thu, 03 Sep 2026 00:00:00 GMT MCP is useful when it makes tool boundaries explicit. The real tool-protocol decision is how you preserve authority, schemas, observability, and change control. MCP tool protocols agent tools agent harness interoperability https://metaharnessscripture.com/posts/human-in-the-loop-agent-escalation/ https://metaharnessscripture.com/posts/human-in-the-loop-agent-escalation/ Wed, 02 Sep 2026 00:00:00 GMT Human-in-the-loop escalation design gives agents clear stop rules, useful handoffs, bounded authority, and a path to learn from review. human-in-the-loop agent escalation agent harness agent safety agent reliability https://metaharnessscripture.com/posts/cost-per-completed-task/ https://metaharnessscripture.com/posts/cost-per-completed-task/ Tue, 01 Sep 2026 00:00:00 GMT Cost per completed task turns agent spend into an operational metric: include retries, human review, failures, and the price of making work right. agent cost agent evaluation agent harness LLM agents agent reliability https://metaharnessscripture.com/posts/memory-architectures-long-running-agents/ https://metaharnessscripture.com/posts/memory-architectures-long-running-agents/ Mon, 31 Aug 2026 00:00:00 GMT Long-running agents need memory outside the context window. Design working, episodic, semantic, and procedural memory with retrieval and expiry. agent memory long-running agents agent harness context engineering agent reliability https://metaharnessscripture.com/posts/idempotency-for-agents/ https://metaharnessscripture.com/posts/idempotency-for-agents/ Sat, 29 Aug 2026 00:00:00 GMT Agents retry, resume, and get interrupted. Idempotency makes those reruns safe: same intent, one durable outcome, and an audit trail. agent idempotency agent reliability agent harness agent retries tool design https://metaharnessscripture.com/posts/retry-policies-diversify/ https://metaharnessscripture.com/posts/retry-policies-diversify/ Thu, 27 Aug 2026 00:00:00 GMT An agent retry with the same prompt and same context produces the same failure. Diversified retry policies turn repeated failures into forward progress. agent retry retry policy agent harness agentic systems agent reliability https://metaharnessscripture.com/posts/verifier-spectrum/ https://metaharnessscripture.com/posts/verifier-spectrum/ Mon, 24 Aug 2026 00:00:00 GMT Agent verification lives on a spectrum from cheap schema checks to expensive ground-truth comparison. The art is matching the verifier to the risk at each step. agent verification verifiers agent harness agent reliability output validation https://metaharnessscripture.com/posts/model-swap-drill/ https://metaharnessscripture.com/posts/model-swap-drill/ Sat, 22 Aug 2026 00:00:00 GMT A model swap breaks agents in predictable ways: tool-call formats drift, stop conditions change, costs spike. Here's the drill to run before you deploy. model swap agent harness llm agents meta harness agent reliability https://metaharnessscripture.com/posts/agent-observability-what-to-log/ https://metaharnessscripture.com/posts/agent-observability-what-to-log/ Fri, 21 Aug 2026 00:00:00 GMT Most agent logs are noise. The signal lives in tool calls, decision boundaries, and failure traces. Here is what to record, what to throw away, and why. agent observability agent harness agent logging agent monitoring agent reliability https://metaharnessscripture.com/posts/prompt-versioning-for-agents/ https://metaharnessscripture.com/posts/prompt-versioning-for-agents/ Wed, 19 Aug 2026 00:00:00 GMT Prompt versioning turns agent regressions from archaeology into a diff. Version, review, and roll back prompts with the discipline you apply to code. agent harness prompt engineering llm agents meta harness agent reliability https://metaharnessscripture.com/posts/guardrails-budgets-and-kill-switches/ https://metaharnessscripture.com/posts/guardrails-budgets-and-kill-switches/ Tue, 18 Aug 2026 00:00:00 GMT Agents fail sideways: runaway loops, blown budgets, tools misused at 3am. Guardrails and kill switches turn an agent from a liability into infrastructure. agent guardrails agentic systems kill switch reliability https://metaharnessscripture.com/posts/what-is-an-agent-harness/ https://metaharnessscripture.com/posts/what-is-an-agent-harness/ Mon, 17 Aug 2026 00:00:00 GMT An agent harness is the scaffolding around an LLM that makes it reliable — tools, context, loops, and guardrails. The anatomy of a well-performing agent. agent harness agentic systems llm agents architecture https://metaharnessscripture.com/posts/the-agentic-loop/ https://metaharnessscripture.com/posts/the-agentic-loop/ Sun, 16 Aug 2026 00:00:00 GMT Most agent failures are loop failures. Design stop conditions, verifiers, and retry budgets that make LLM agents finish reliably instead of wandering. agentic loop agent evaluation control flow llm agents https://metaharnessscripture.com/posts/context-engineering-beats-prompt-engineering/ https://metaharnessscripture.com/posts/context-engineering-beats-prompt-engineering/ Sat, 15 Aug 2026 00:00:00 GMT What your agent knows at inference time is decided by context engineering, not prompts. Compaction, retrieval, and window hygiene for LLM agents. context engineering llm agents memory retrieval https://metaharnessscripture.com/posts/tool-design-for-agents/ https://metaharnessscripture.com/posts/tool-design-for-agents/ Fri, 14 Aug 2026 00:00:00 GMT The tool surface is the agent's API to reality. Design LLM agent tools that are hard to misuse: small surfaces, strict schemas, scoped permissions. tool design llm agents agent tools schemas https://metaharnessscripture.com/posts/evals-for-agentic-systems/ https://metaharnessscripture.com/posts/evals-for-agentic-systems/ Thu, 13 Aug 2026 00:00:00 GMT Model benchmarks don't predict agent performance. How to build task-level evals: golden traces, scenario suites, regression gates, scoring harness design. agent evaluation benchmarks testing llm agents