Context Management & Reliability
Domain 5 is where architecture meets reliability. As Claude agents run longer and coordinate more subagents, the biggest risks stop being model capability and start being **context hygiene**: facts that vanish under summarisation, errors that get silently swallowed, and citations that dissolve during synthesis. This 15% slice of the CCAR-F blueprint tests whether you can keep a stateless API coherent across long conversations, decide when an agent should hand off to a human, and make failures visible instead of invisible. The six lessons move from single-conversation context windows to multi-agent pipelines. You will learn structural fixes (case-facts blocks, prompt caching, scratchpad files, structured error objects) that beat prompt-level pleading every time. Expect scenario questions that reward the smallest proportionate change — usually a system-prompt or data-layout fix — over expensive architectural rewrites.
6 lessons in this domain
- 5.1
Context Window Management
Long conversations quietly destroy transactional data. Learn the structural patterns — persistent fact blocks, tool trimming, and caching — that keep Claude coherent.
7 min read - 5.2
Escalation & Ambiguity
Miscalibrated escalation wrecks first-contact resolution. Learn the three valid triggers, the two unreliable ones, and how to handle ambiguous customer matches safely.
6 min read - 5.3
Error Propagation
Failures in subagents decide whether a system recovers or fails silently. Learn structured error context, the two deadly anti-patterns, and access-failure vs empty-result.
7 min read - 5.4
Codebase Exploration
During big codebase exploration Claude drifts from exact file paths to vague 'typical patterns'. Learn why, and the scratchpad, subagent, and /compact fixes.
6 min read - 5.5
Human Review Calibration
97% overall accuracy can hide catastrophic failures on specific segments. Learn to validate by document type, calibrate confidence, and prioritise reviewer capacity.
6 min read - 5.6
Information Provenance
Attribution dies during summarisation unless you preserve it. Learn claim-source mappings, conflict handling, temporal awareness, and format-by-content-type.
6 min read