Attemory System claims its released Spire agent can sustain long-horizon Slay the Spire play by delegating deterministic actions to domain tools while using an LLM for reasoning, with about 40% of runs reaching Act 3.
state: expiredheat: lowuncertainty: highconvergesscott: lowdomain-agents agent-harnesses long-running-agentsAttemory System
What is this?
Attemory System presents Spire as an autonomous Slay the Spire agent that delegates deterministic game actions to domain-specific tools while reserving the LLM for reasoning. The supplied results report support for Ironclad and Defect and, at the A20 difficulty, reaching Act 3 in roughly 40% of runs and Act 4 in roughly 10%, with multiple confirmed Heart wins. These architecture and performance claims appear to come from the creator; the snippets provide no independent validation or detailed methodology.
Why it matters to Scott
Spire independently embodies Scott’s Deterministic-AI Pendulum and Placement Judgment: deterministic domain tools handle reliable execution while the LLM supplies judgment. However, the supplied evidence is only a creator-reported, game-specific demonstration without independent validation or enough architectural detail to extend Scott’s position or change what he builds.
ip:concept.deterministic-ai-pendulumip:concept.placement-judgmentradar:concept.agent-harnessesradar:concept.game-agents
queries asked of Scott's wikis
- deterministic tools versus LLM reasoning in agent harnesses
- hybrid symbolic and LLM agent architectures
- long-horizon agent state and memory
- games as testbeds for agent reliability
- domain-specific tools for reducing agent errors
- evaluation metrics for long-running agents
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-09-04T07:35:35Z
No independent validation, implementation uptake, or substantive technical detail emerged within the observation window; the creator-reported game benchmark remains an isolated illustration of a familiar agent pattern.
2026-09-02T07:31:53Z
Re-evaluation adds no corroboration or technical detail; this remains a creator-reported, game-specific implementation of a familiar deterministic-tools-plus-LLM pattern.
2026-09-02T07:26:52Z
grounded: converges/low — Spire independently embodies Scott’s Deterministic-AI Pendulum and Placement Judgment: deterministic domain tools handle reliable execution while the LLM suppli
2026-09-02T07:24:13Z
case created — This is a usable first-party agent artifact with a specific long-horizon architecture and measurable performance claim.
Decision trace
- 09-04 17:35expireNo independent validation, implementation uptake, or substantive technical detail emerged within the observation window; the creator-reported game benchmark remains an isolated illustration of a famil
- 09-04 17:35alert_silentThe staleness trigger adds no consequential evidence, and the unvalidated release does not warrant attention outside a future substantive update.
- 09-04 17:35alert_routeThe staleness trigger adds no consequential evidence, and the unvalidated release does not warrant attention outside a future substantive update.
- 09-02 17:31repriceRe-evaluation adds no corroboration or technical detail; this remains a creator-reported, game-specific implementation of a familiar deterministic-tools-plus-LLM pattern.
- 09-02 17:31alert_silentThere is no new consequential delta beyond the already assessed repository release, and the performance claims remain independently unvalidated; it can wait for corroboration or a substantive technica
- 09-02 17:31alert_routeThere is no new consequential delta beyond the already assessed repository release, and the performance claims remain independently unvalidated; it can wait for corroboration or a substantive technica
- 09-02 17:29alert_silentThe repository release is an established creator-published event, but its reported Act 3/Act 4 performance is unvalidated and the available material lacks enough architectural or evaluation detail to
- 09-02 17:29alert_routeThe repository release is an established creator-published event, but its reported Act 3/Act 4 performance is unvalidated and the available material lacks enough architectural or evaluation detail to
- 09-02 17:26groundSpire independently embodies Scott’s Deterministic-AI Pendulum and Placement Judgment: deterministic domain tools handle reliable execution while the LLM supplies judgment. However, the supplied evide
- 09-02 17:24createThis is a usable first-party agent artifact with a specific long-horizon architecture and measurable performance claim.