Anneal Temper
Run fixed-point, multi-agent deepen cycles on complex but bounded architecture or plans until variance-based convergence before you commit to full implementation.
Overview
anneal-temper is an agent skill most often used in Validate (also Build, Ship) that iteratively deepens architecture or plans via multi-agent fixed-point refinement until variance-based convergence.
Install
npx skills add https://github.com/krzemienski/anneal --skill temperWhat is this skill?
- Fixed-point deepen loop that refines architecture or plans across repeated agent passes
- Variance-based convergence criteria to stop iteration instead of endless rewrites
- Multi-agent plan-and-review workflow tuned for complex but scoped tasks
- Anneal-style tempering: progressive refinement rather than one-shot generation
- Best when a single chat pass leaves architecture still drifting between revisions
Adoption & trust: 1 GitHub stars.
What problem does it solve?
You have a complex but scoped design or plan that keeps changing across agent passes with no clear signal to stop refining and start building.
Who is it for?
Solo builders facing bounded, high-complexity architecture or planning work who want multi-agent deepen cycles with an explicit convergence stop rule.
Skip if: Simple one-shot tasks, already-approved specs, or work where iterative multi-agent refinement would add cost without narrowing design variance.
When should I use this skill?
Best for complex but scoped tasks that need fixed-point iterative refinement of architecture or plans with variance-based convergence via multi-agent deepen.
What do I get? / Deliverables
You get a stabilized architecture or plan artifact after variance-based fixed-point iterations, ready to hand off to implementation, review, or deployment prep without endless rewrite loops.
- Converged architecture or plan document after fixed-point iterations
- Documented convergence rationale (variance threshold met)
Recommended Skills
Journey fit
Spans multiple journey phases - primary shelf plus alternate fits below.
Canonical shelf is validate/scope because the skill temper-refines architecture and plans until they stabilize—work that belongs after discovery but before heavy build. Scope subphase fits iterative deepening of system boundaries, interfaces, and plan structure on tasks that are large yet explicitly bounded.
Where it fits
Deepen service boundaries and data flows across agent rounds until variance between iterations drops below your convergence threshold.
Temper a large implementation plan with fixed-point review passes before assigning files and milestones to your coding agent.
Run converge cycles on architecture review findings so security and design feedback integrate without reopening settled decisions endlessly.
How it compares
Use instead of ad-hoc multi-turn architecture chat when you need fixed-point convergence criteria rather than subjective “looks good enough.”
Common Questions / FAQ
Who is anneal-temper for?
It is for solo and indie builders using agentic coding tools who need to deepen complex but scoped architecture or plans with multi-agent iterations and a defined convergence rule.
When should I use anneal-temper?
Use it during Validate to converge scope and architecture, during Build when PM-style plans still drift across agent runs, and during Ship when review feedback needs fixed-point refinement before you lock the design—always for complex tasks with clear boundaries.
Is anneal-temper safe to install?
Treat it like any third-party agent skill: review the Security Audits panel on this Prism page and inspect SKILL.md in your repo before granting shell, network, or secrets access to your agent.
SKILL.md
READMESKILL.md - Anneal Temper
Fixed-point deepen architecture. Iterative refinement with variance-based convergence. Best for complex but scoped tasks. # anneal-temper Source: ./temper { "name": "anneal-temper", "author": { "name": "Nick Krzemienski" }, "source": "./temper", "version": "0.1.0", "keywords": [ "plan", "review", "multi-agent", "deepen", "fixed-point", "anneal" ], "description": "Fixed-point deepen architecture. Iterative refinement with variance-based convergence. Best for complex but scoped tasks." }