fast-jev-compaction
Prune stale coding-agent tool history without rewriting the context that remains.
What Jev decides
Whether each old tool call and its result should remain available.
The library asks two Noul questions for each eligible tool call and result, then local policy keeps, truncates, or removes the pair while preserving ordinary conversation text verbatim.
- Jev role
- Judge whether older tool activity is still relevant to the active coding task.
- Input
- The current goal, fitted conversation history, tool inputs, and compact result previews.
- Output
- Two Noul probabilities per eligible tool call.
- Decision frequency
- When manual or automatic context compaction runs.
- Fallback
- Use the host agent's built-in compaction when Jev fails or reduction is too small.
How it works
Reported performance
Overview
fast-jev-compaction replaces summary-based compaction with selective pruning. Recent activity is protected, the conversation is fitted into Jev's request budget, and each older tool call is judged for whether the call and its full result still matter.
Keep, truncate, and drop are application actions derived from two Noul probabilities; they are not a Jev Choice response. Failures and insufficient reduction fall back to the host agent's normal compaction path.
- Language
- TypeScript
- License
- MIT
- Author
- tamaratran
Sources
Related projects
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.
Foreman
Supervise slower coding agents with parallel Jev assessments and deterministic intervention policy.
What Jev decides
The probability of completion, sufficient tests, satisfied requirements, useful progress, drift, a stuck worker, verification need, readiness to finish, and need for a human.