Jev Compaction
Use when a transcript has to be cut to a fixed size and you must choose which turns go.
Install
npx skills add kerpopule/hermes-jev-skills --skill jev-compaction -g -a claude-code -y- Repository
- kerpopule/hermes-jev-skills1.3k
- Category
- Engineering
- Updated
- 6d ago
- License
- MIT
- Context cost
- Lightabout 1,116 tokens
- Runs scripts
- No, instructions only
What it does
Use when a transcript has to be cut to a fixed size and you must choose which turns go. Jev marks each turn keep, summarize or drop. Measured: it does not make a handoff better.
Choosing turns with Jev, and what actually makes a handoff work
Jev cannot write a summary. It can mark each turn of a transcript:
- keep: carries a decision, a constraint, a preference, unfinished work, or an exact value, path, id, command or error that later work depends on.
- summarize: background whose gist matters. In the digest this is the turn's first 400 characters, nothing more. Jev writes no gist.
- drop: chatter, superseded attempts, repeated output.
It judges a long turn on its first 350 and last 350 characters, redacted, 40 turns per request, and sees no other turn while it does.
Read this before you use it for a handoff
This skill used to say a handoff written from Jev's digest "stops losing the one line that mattered". We measured that on seven real sessions and 104 recall questions (scorecard) and it was wrong:
| the writer reads | capsule | recall alone | with one search of the old session |
|---|---|---|---|
| Jev's digest | 400 words | 37.5% | 68.3% |
| the plain last 24,000 characters | 400 words | 48.1% | 68.3% |
| the whole dialogue | 1,200 words | 58.7% | 75.0% |
| nothing: no handoff at all | 56.7% |
Jev's marks did beat the same marks handed out by recency (11 questions to 4), so the judgement is real. The digest built around it clips every other turn to 400 characters, and that cost more than the judgement earned. Nous Research found the same shape with a different Jev design (hermes-agent PR 116246).
So, for a handoff:
