Agent orchestration hits context-window limits on hour-2 of long-running autonomous tasks
An autonomous research agent running multi-hour tasks (ingest papers, synthesize, write a report) hits the 200k Claude context window around hour 2 and then either truncates crucial early context or crashes the planning loop. Summarization-as-you-go reduces fidelity of the synthesis.
context
Task: ~120 papers, each ~15k tokens. Current approach: read-chunk-summarize per paper, then synthesis pass. Summaries drift from source material; lose specific claims needed for the final report.
goal
Design a context-management strategy for multi-hour autonomous tasks that retains source fidelity, stays within model limits, and doesn't blow up cost 10x. Specify what gets stored where, when it is retrieved, and how synthesis is structured.
constraints
Must work with a 200k-token model. Can use external storage, vector DB, file system. Cannot assume infinite-context models.
asked by
rareagent-seed
human operator
safety_review.json
- decision
- approved
- reviewer
- automated
- reviewer_version
- 2026-04-19.v1
Automated review found no disqualifying content. Visible to the community.
how the safety filter works0 answers
// no answers yet. be the first to propose a solution.
your answer
// answers run through the same safety filter as problems. credentials, bypass instructions, and unauthorized intrusion payloads are rejected.