Four tasks: demote stale 'Right now' out of _CORE, rewrite 'How you talk'
in-voice with the 4 anti-tic rules + real exemplars, trim hedgy prose
for a leaner hot path, and a replay eval on the exact safe-trigger
prompts. Runs via `uv run` from the worktree (shared venv resolves to
the wrong checkout).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
Diagnosed 4 "safe" tics in real Talk-mode transcripts (menu-not-pick,
tag-question deferral, reassurance reflex, both-sides-ing judgment).
Approach C: rewrite the hot-path core in-voice, name the tics as hard
rules, embed exemplars from her own best moments; token diet + drop the
stale "Right now" from the always-on _CORE. Files: personas/lyra.md,
persona.py.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
The message-type-prompts plan (sub-project 2) predated the scouting desk + roster
build on this branch. Readjust it so the prompting work starts from reality:
- Add a "Readjustment 2026-07-04" section: the real failure shifted from mush to
MISSED TOOL CALLS; two new message types (READ, TABLE); HAND gains a
hero-vs-observed split; the scouting desk is a live per-turn injection layer to
compose with (not duplicate); BASE must cover the expanded toolset + identity
rules; source card grew to modes.py:67-169; Phase A still unbuilt.
- Taxonomy → READ | HAND | TABLE | MENTAL | STATUS | LOG | CHAT, with READ above
HAND (a villain's action lands on their file, not Brian's).
- New READ + TABLE fragments; BASE routes the full current tool set; STATUS
narrowed to pure logistics; classify tests add READ/TABLE + the READ↔HAND edge.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Observed live: an overheated MI50 returns a single char repeated ("?????") as a
successful 200, which neither the timeout nor the exception fallback catches — so
a degraded GPU would silently save capped garbage gists. Validate each summary
call's output: flag text (>=24 non-space chars) whose most-common non-whitespace
char exceeds 50%, raise DegenerateOutput, and let the existing retry->cloud
fallback handle it. Real prose (top char <20%) won't false-positive; short output
is exempt; cloud garbage raises rather than looping.
Tests: _looks_degenerate flags repeated-char / passes real prose / ignores short;
degenerate MI50 output falls back to cloud; cloud garbage raises. 177 pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015yrEb5qpPGv2FjyxrB7LLk
Phase A pipeline fixes (suppress the misfiring _route mood nudge + the
always-on mode-menu note in poker mode), Phase B classifier
(HAND/STATUS/MENTAL/LOG/CHAT) + per-type fragments replacing the
monolithic _CASH_CARD, Phase C MI50 tool-calling. NLH-only HAND
reasoning; PLO hands logged/replayed but not analyzed this pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
Seven tasks: contract module + conformance test, direct capture
endpoints (stack/buyin/start), hands API, reads/players API + route
conformance, chat-page stack quick-capture box, HUD quick inputs +
villain rename, and the iOS-PWA bottom safe-area fix. TDD with real
test/impl code grounded in the actual poker.py signatures and FastAPI
route patterns.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
Decompose into two sub-projects: (1) harden poker.py into a standalone
logging service with a complete REST API, a first-class documented
tool/API contract, and a human UI to log/edit/correct everything —
usable by Brian alone, zero LLM dependency; (2) wire Lyra in as a
client (classifier + message-type prompts), parked. Contract is
first-class because it's the shared seam for the cloud model, a
fine-tuned MI50 poker model, RTO, the human UI, and a future MCP wrap.
MCP deferred until a second host app exists.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
Two-track poker interaction: dumb data capture (stack/buyin/cashout)
that bypasses the LLM, plus a message-type classifier that injects
type-specific prompt fragments (HAND/STATUS/MENTAL/LOG/CHAT) in place
of the one broad poker card. Kills the false tilt-reads and the
coaching-essay-on-every-turn behavior. Includes the iOS-PWA bottom
safe-area fix and the 2nd input box from Brian's screenshot.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G796GsLCvJQKVN7hwV2cDx
serversdown