0d64d8e559
Tracked: morrowind agent (py/cfg), skills/, training-data/, research/, notes/, specs/, test-results/, metrics/, heartbeat/, briefings/, memories/, skins/, hooks/, decisions.md, OPERATIONS.md, SOUL.md Excluded: screenshots, PNGs, binaries, sessions, databases, secrets, audio cache, timmy-config/ and timmy-telemetry/ (separate repos)
7 lines
1.8 KiB
JSON
7 lines
1.8 KiB
JSON
[
|
|
{
|
|
"prompt": "Create a Gitea issue on rockachopa/hermes-agent at http://143.198.27.163:3000 for the A-B-C-D eval test matrix. Token is in ~/.hermes/gitea_token_vps. Milestone ID is 16 (AutoLoRA P1). Use labels [86, 92, 90] (autolora, eval, training).\n\nTitle: [AutoLoRA] A-B-C-D Eval Test Matrix \u2014 Bare vs Sessions vs Curated vs Combined\n\nBody should describe:\n- A) Bare hermes3:8b \u2014 0.551 composite (DONE)\n- B) LoRA trained on ~364 compressed real sessions \u2014 noisy signal, real-world tool patterns\n- C) LoRA trained on 29 curated exemplars \u2014 pure values signal, hand-written gold standard\n- D) LoRA trained on sessions + curated combined \u2014 best of both\n\nEach variant runs through the same eval harness (run_eval.py + run_vibes.py) and gets compared via compare.py.\n\nThe key question: does a tiny pristine dataset (C) outperform a large noisy dataset (B)? And does combining them (D) beat both?\n\nSacred rule: any variant that degrades pastoral care or crisis response is REJECTED.\n\nAcceptance criteria: all 4 variants evaluated, comparison report generated, winner promoted to Ollama.\n\nAlso create a second issue on the SAME repo, same milestone, same labels, titled: [AutoLoRA] Audit Timmy-time-dashboard for Hermes harness duplication\n\nBody: Per rockachopa/Timmy-time-dashboard#1215 comment \u2014 the dashboard repo has legacy code that rebuilds things the Hermes harness already provides (tools.py, memory_system.py, thinking.py). Audit the dashboard codebase and identify all components that duplicate Hermes agent functionality. Recommend which to kill, which to keep, and which to redirect to Hermes. This prevents wasted cycles building harnesses we don't need. Reference: Timmy-time-dashboard#1215#issuecomment-9115.",
|
|
"chosen": "",
|
|
"session": "session_20260323_184349_3c75ce.json"
|
|
}
|
|
] |