Context Management & Reliability

CCAR-F · 41 questions

• A long 311 session summarizes aggressively and starts losing dollar amounts and case IDs. What should be preserved separately?
• Verbose work-order tool dumps are bloating the 311 agent context. What should happen before accumulation?
• Aggregated subagent research for a policy brief buries key findings in the middle of a long prompt. What placement helps?
• Downstream context budgets are tight after several research subagents. What should subagents return?
• A multi-issue housing session jumbles IDs, amounts, and statuses across topics. What context design helps?
• A 311 resident explicitly asks to speak with a human. What should the agent do?
• Which triggers should a municipal agent treat as appropriate escalation conditions?
• A city architect wants to route hard cases using sentiment scores and the model's self-reported confidence. What judgment applies?
• A clerk lookup returns multiple residents with similar names. What should the agent do?
• How should a utility replacement agent learn when to resolve a standard meter swap versus escalate exceptions?
• A GIS search subagent fails during a zoning research workflow. What should it return to the coordinator?
• A valid GIS parcel query returns no matching parcels. How should the subagent label the outcome?
• A permit-data subagent hits a transient timeout. What error-propagation pattern is appropriate?
• During multi-source civic synthesis, some topics lack support because sources failed. What should synthesis do?
• Which multi-agent error patterns should a municipal architect reject?
• A sprawling legacy student-information-system exploration keeps losing findings across long sessions. What helps?
• Finding every refund-caller path in a county billing codebase is extremely verbose. What should the main agent do?
• A city codebase agent crashes mid-exploration. How should recovery be designed?
• During a multi-hour municipal codebase tour, context fills with discovery noise. What Claude Code practice helps?
• Before spawning the next wave of exploration subagents, what should the coordinator do with the prior phase?
• A clerk extraction pipeline reports 97% overall accuracy but fails badly on handwritten hardship affidavits. What does this illustrate?
• How should a city monitor error rates even among high-confidence permit extractions?
• Before routing field-level confidence for municipal forms, what should the team do?
• Low-confidence or contradictory fields appear on a digital permit intake. What should happen first?
• When may a municipality safely reduce human review volume on automated extractions?
• Policy research subagents gather citations for a housing brief. What must synthesis preserve?
• Two credible transit reports disagree on ridership. What should synthesis do?
• Open-data findings for a parks briefing omit publication dates. What risk does that create, and what should travel with findings?
• How should a final civic research brief present certainty?
• When synthesizing mixed civic source types, how should financial tables and news articles be rendered?
• A transit-corridor investigation session has grown very long, and the team notices Claude Code responses becoming slower and less focused as irrelevant early exploration accumulates in context. What should the team consider doing?
• A water-quality research agent is given one very long combined document containing lab results, regulatory citations, and complaint logs, and it reliably cites content from the beginning and end but misses a key finding buried in the middle. What does this illustrate?
• A parks master-plan agent needs to work through messy intermediate reasoning about trail-usage data before producing a clean final recommendation, and the session is starting to run long. Which combination of practices helps?
• A public-health benefits research agent produces a summary of eligibility rules and lists several supporting claims, some pulled confidently from a verified state statute and others inferred loosely from an outdated blog post. What should the agent's output preserve so a human reviewer can calibrate trust appropriately?
• During a long housing-authority casework session, Claude Code's context window fills with dozens of prior tool outputs and file reads. The team notices that instructions given early in the session are now being followed less reliably than instructions given recently. What phenomenon does this describe, and what is a reasonable mitigation?
• In a multi-agent 911-dispatch data pipeline, a specialist subagent encounters an unrecoverable error partway through validating incident records and cannot complete its task. What should happen next in a well-designed multi-agent workflow?
• A court clerk's office feeds a 40-page hearing transcript into a single prompt asking Claude to extract every stipulated deadline into a structured JSON list. Deadlines mentioned on pages 15-25 are reliably missed, while ones near the start and end are captured correctly. What is the most likely cause, and what is a reasonable fix?
• A 311 request classifier returns structured JSON with a category field and a confidence field (0-1). The team adds a rule: any result with confidence < 0.6 is automatically routed to a human dispatcher instead of auto-filed. What is this design pattern primarily addressing?
• A coordinator agent spawns a subagent to extract structured fee data from scanned water-bill images, then immediately passes that subagent's raw output into a second subagent that calculates late-payment penalties, without checking the first output's shape. The second subagent silently computes penalties against a missing dueDate field as if it were zero. What does this scenario illustrate, and what should the coordinator do differently?
• A housing authority runs a long agentic session that reads forty case files, extracts each household's recertification date, and at the end writes a single consolidated waitlist report. Partway through, automatic compaction fires. The agent then produces a report covering only the last few cases and describes the earlier ones vaguely. What is the best explanation and remedy?
• During a long session, a parks-department agent read summerprogramroster.csv early on, then spent many turns on other work and went through a compaction. The user now asks it to update three campers' session assignments in that file. What should the agent do first?