Build the harness that can explain a failure.
Explore execution boundaries, safe retries, independent grading and self-improvement with review gates. Runnable examples use synthetic state and never deploy a candidate automatically.
Your model is not your runtime: where the harness belongs
Read the design, run the counterexample, and inspect our implementation.
Harness engineeringThe retry that charged twice: durable checkpoints for agents
Read the design, run the counterexample, and inspect our implementation.
Harness engineeringA self-improving harness needs a promotion gate it cannot edit
Read the design, run the counterexample, and inspect our implementation.
Harness engineeringTrace the failure without turning the trace into the truth
Read the design, run the counterexample, and inspect our implementation.
Agent definitions, teams and adaptive routing
Ten additional guides with runnable Python exercises, expected results and counterexamples. Each exercise includes an expected result and a counterexample.
Define an agent before designing its harness
Separate the agent specification, execution policy and independent success criteria.
Agent definitions and routingADK agent, runner and harness: who owns what?
Place the framework lifecycle inside application-owned identity, budgets and delivery checks.
Agent definitions and routingDesign an agent team without creating a delegation loop
Use typed handoffs, capability checks and a shared budget to keep specialist teams accountable.
Agent definitions and routingBuild a small Markov agent router from transition counts
Calculate interpretable route scores, apply smoothing and preserve an explicit fallback.
Agent definitions and routingChoose routing thresholds using error costs and abstention
Measure coverage and routing mistakes before promoting a faster decision path.
Agent definitions and routingPrepare routing JSONL without teaching the router its own mistakes
Preserve label provenance and separate task families before evaluating routing updates.
Agent definitions and routingDetect agent stalls using evidence of progress
Combine task-specific progress signals with global turn, time and redirect limits.
Agent definitions and routingPrevent agent abuse at the tool boundary
Enforce resource scope, exact-action approval and independent outcome verification.
Agent definitions and routingChoose models using task evidence and hard constraints
Filter incompatible candidates before comparing measured quality, latency and cost.
Agent definitions and routingRelease an agent and its harness as one evaluated unit
Bind approvals to the complete execution configuration and preserve rollback evidence.
Run the same grader locally
git clone https://github.com/Sandeep-BA/cassandra-mcp.git
cd cassandra-mcp
node scripts/run_harness.cjs promotion examples/harness-promotion.json
npm testThe CLI and browser share our original grading code. No key, model download, provider request or cloud bill is involved. This is a teaching harness, not an autonomous production deployment system.