mons-rust

cargo add mons-rust

or

npm install mons-rust

automove

The active automove workflow lives here:

• canonical workflow: HOW_TO_ITERATE_ON_AUTOMOVE.md
• ideas backlog: AUTOMOVE_IDEAS.md
• compatibility entrypoint: docs/automove-experiments.md
• durable lessons: docs/automove-knowledge.md
• retired profile archive: docs/automove-archive.md

Useful scripts:

• ./scripts/run-automove-experiment.sh guardrails <candidate>
• ./scripts/run-automove-experiment.sh runtime-preflight <candidate>
• ./scripts/run-automove-experiment.sh triage-calibrate [reply_risk|opponent_mana|supermana|all]
• ./scripts/run-automove-experiment.sh preflight <candidate>
• SMART_TRIAGE_SURFACE=<surface> ./scripts/run-automove-experiment.sh triage <candidate>
• SMART_PROMOTION_TARGET_MODE=<fast|normal> ./scripts/run-automove-experiment.sh audit-screen <candidate>
• SMART_PROMOTION_TARGET_MODE=<fast|normal> ./scripts/run-automove-experiment.sh fast-screen <candidate>
• SMART_PROMOTION_TARGET_MODE=<fast|normal> ./scripts/run-automove-experiment.sh progressive <candidate>
• SMART_PROMOTION_TARGET_MODE=<fast|normal> ./scripts/run-automove-experiment.sh ladder <candidate>
• SMART_TRIAGE_SURFACE=<opening_reply|primary_pro> ./scripts/run-automove-experiment.sh pro-triage <candidate>
• ./scripts/run-automove-experiment.sh pro-audit-screen <candidate>
• ./scripts/run-automove-experiment.sh pro-fast-screen <candidate>
• ./scripts/run-automove-experiment.sh pro-progressive <candidate>
• ./scripts/run-automove-experiment.sh pro-ladder <candidate>
• ./scripts/run-automove-experiment.sh pre-screen <candidate>
• ./scripts/run-automove-experiment.sh pro-pre-screen <candidate>
• ./scripts/clean-experiment-artifacts.sh

Run triage-calibrate before candidate work on reply_risk, opponent_mana, or supermana. Then use guardrails -> triage -> runtime-preflight so weak ideas die before the expensive CPU gate. preflight still exists as the old all-in-one wrapper. Most candidates should die at triage, pro-triage, or the first earned duel stage. For mode-specific fast or normal ideas, run duel stages with SMART_PROMOTION_TARGET_MODE=<fast|normal> so the target mode is the improvement bar and the other client mode is only a non-regression check. Use audit-screen or pro-audit-screen only as occasional spot checks for clean triage rejects. pre-screen and pro-pre-screen remain available only as legacy noise diagnostics. Unless documented otherwise, new candidates should be deltas on runtime_current.

rules-tests runner

Fixtures are stored as chunk archives in rules-tests-chunks/ (default: 100000 fixtures per chunk).

Run all fixtures:

./scripts/run-rules-tests.sh

Useful options:

./scripts/run-rules-tests.sh --limit 100

./scripts/run-rules-tests.sh --log /tmp/rules-tests.log

./scripts/run-rules-tests.sh --chunks-dir ./rules-tests-chunks --verbose

rules-tests generator

Generate new random unique fixtures and repack chunks:

./scripts/generate-rules-tests.sh --target-new 100

Generate directly into a directory (continuous mode):

./scripts/generate-rules-tests.sh --dir /tmp/rules-tests-work

Pack a directory back into chunks:

./scripts/pack-rules-tests.sh --dir /tmp/rules-tests-work --chunks-dir ./rules-tests-chunks --chunk-size 100000

publishing to npm

./publish.sh