Hermes Agent context engine: the main model writes the compaction handoff on its cached prompt prefix
Evidence
These records cover research runs and issue checks. They have metadata only: counts, codes, booleans, times, and hashes. They contain no message text, request body, or key. Paths such as .work/... and short commit IDs in the research records refer to the research repository.
| File | Content |
|---|---|
| issue24-validation.json | Issue 24: assembled-capture refusal before the fix, effective-content matching after it, safe capture-ahead refusal, and synthetic checks on clean minimum and current Hermes source |
| issue24-live-dgx-20261009.json | Issue 24: one manual and one automatic live DGX test with synthetic context; exact request prefixes, saved-history reloads, reported cached tokens, and measurement limits |
| issue-12-whitespace.json | Issue 12: refusal before the fix, three warm compactions after it, and seven integration scenarios on clean Hermes e36a8180; synthetic conversations and a loopback fake server |
| plugin-integration.json | The integration check: install with the Hermes install command, manual and automatic compaction, fallback, tools, and rollback on unpatched Hermes 45871e10, against a loopback fake server |
| plugin-live.json | Live runs, plugin against the built-in compressor: DGX and LM Studio, manual and automatic, ten synthetic cases of 100,000 or more tokens each |
| lcm-bench.json | Live runs, plugin against hermes-lcm 0.21.0-rc2 and the built-in compressor: DGX, automatic compaction, ten synthetic cases of about 105,000 tokens |