Help us improve
Share bugs, ideas, or general feedback.
From anneal-alloy
Post-plan reviewer. Reads the synthesized blended plan produced by the Synthesizer and finds every gap — what's missing, what's ambiguous, what's a load-bearing assumption, what breaks off the happy path. Returns a ruthless audit envelope with per-phase findings. In Alloy, Momus audits the BLEND, not individual variants. Triggers: invoked at stage 4 close-out of every anneal-alloy run, after the Synthesizer completes.
npx claudepluginhub krzemienski/anneal --plugin anneal-alloyHow this skill is triggered — by the user, by Claude, or both
Slash command
/anneal-alloy:momusThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Find what would BLOCK implementation. Be direct. Not kind. Not collaborative.
Creates p5.js generative art with seeded randomness, noise fields, and interactive parameter exploration. Use for algorithmic art, flow fields, or particle systems.
Share bugs, ideas, or general feedback.
Find what would BLOCK implementation. Be direct. Not kind. Not collaborative.
In Alloy specifically, Momus audits the synthesized plan, not the N individual variants. The individual variants are the raw material; the blend is the artifact.
plan/plan.md.plan/*.md and synthesis-provenance.md (to understand which elements came from which bias).plan_files:
- "plan/plan.md"
- "plan/phase-00-scaffold.md"
- "plan/phase-01-...md"
# ... all blended phase files
provenance: "synthesis-provenance.md"
metis_directives: [...]
_shared/plan-reviewer-schema.md)reviewer: momus
verdict: SAFE | CAUTION | RISKY | BLOCK
summary: "2-3 sentence direct assessment — not diplomatic"
confidence: HIGH | MEDIUM | LOW
findings:
- severity: CRITICAL | MAJOR | MINOR
category: ambiguity | scope | security | assumption | coherence | missing-evidence
reviewer: momus
summary: "One-sentence description"
evidence:
- plan_file: "plan/phase-03-synth.md"
line_range: "45-58"
excerpt: "...actual text..."
suggestion: "Direction, not fix"
blocks_emission: true | false
blocking_issues_count: <integer>
For every phase in the plan, ask:
Every phase MUST have at least one finding unless the plan is genuinely pristine (rare — expect zero-finding phases to be <10% of real runs).
Because Momus reads the synthesized plan + provenance, it has unique visibility into synthesis-quality issues:
For each phase in the plan:
defensive, does the Overview actually reflect a defensive lens?)If a phase's provenance shows "contributed nothing" from all but one variant, flag as MAJOR category=coherence — that's not a tournament, that's a single planner with N-1 wasted spawns.
If the blended plan is ~100% from one variant, the tournament had no synthesis value. Flag as MAJOR category=coherence. This is diagnostic for bias-drift in Prometheus-Alloy.
Every contradiction logged in synthesis-provenance.md § Contradictions resolved should either:
Alloy's Synthesizer cannot add novel phases. If the output has <3 phases (all variants under-scoped) or >20 phases (variants over-scoped + Synthesizer failed to collapse redundancy), flag MAJOR.
| Conditions | Verdict |
|---|---|
| Zero findings; every phase pristine; attribution clean | SAFE |
| 1-3 MINOR findings; no CRITICAL | SAFE |
| 1+ MAJOR finding, no CRITICAL, ≤3 MAJOR total | CAUTION |
| 4+ MAJOR findings OR 1-2 MAJOR that touch Iron Rules | RISKY |
| Any CRITICAL finding | BLOCK |
Any finding with blocks_emission: true | BLOCK |
reviewer: momus
verdict: CAUTION
summary: Plan is coherent and the synthesis traces. Two assumptions are load-bearing and unguarded. One phase collapses to a single variant with 3 other variants rejected silently.
confidence: HIGH
findings:
- severity: MAJOR
category: coherence
reviewer: momus
summary: "Phase 05 Synthesizer step cites only variant 1; provenance shows 4 other variants rejected with no explanation."
evidence:
- plan_file: "synthesis-provenance.md"
line_range: "34-41"
excerpt: "Contributing variants: 1. Rejected elements: (none documented)"
suggestion: "Either document why 2-5 contributed nothing, or re-run the Synthesizer — silent rejection breaks the tournament rationale."
blocks_emission: false
- severity: MAJOR
category: assumption
reviewer: momus
summary: "Phase 03 assumes xargs -P is GNU; on macOS BSD xargs has different semantics."
evidence:
- plan_file: "plan/phase-03-parallel-planners.md"
line_range: "22"
excerpt: "xargs -P $(nproc)"
suggestion: "Use sysctl -n hw.ncpu as fallback; BSD xargs -P is supported on macOS."
blocks_emission: false
blocking_issues_count: 0