Damian's Life // model + effort picker

tiers.damianslife.com

Which model and reasoning effort for which domain and task. Toggle between normal mode (quality priority) and low-context mode (context pressure). Last updated 2026-04-19.

Mode toggle

Manual trigger: type the low-context variant alias. Auto trigger: low-context-suggest-hook fires at 70% context.

Shell aliases (already in .zshrc)

# Generic model + effort shortcuts
alias cc='claude --model opus --effort medium'      # default
alias ccol='claude --model opus --effort low'       # 4.7 low ~= 4.6 high at 1/3 tokens
alias ccom='claude --model opus --effort medium'    # explicit
alias ccoh='claude --model opus --effort high'      # deep work
alias ccsl='claude --model sonnet --effort low'     # sonnet low
alias ccsm='claude --model sonnet --effort medium'  # sonnet medium
alias ccsh='claude --model sonnet --effort high'    # sonnet high

# Haiku is NEVER main chat. Haiku only runs as Scout agents.

Normal mode - domain x alias

Domain	Task type	Alias	Model + effort
/darius	Architecture design, ADRs	ccoh	Opus 4.7 high
/cyrus	Psychological session	ccoh	Opus 4.7 high
/fix	Hard debugging	ccoh	Opus 4.7 high
/forge	Content drafting	ccoh	Opus 4.7 high
/jung	Analyst session	ccoh	Opus 4.7 high
/damian	Profile deepening	ccoh	Opus 4.7 high
/treasury	Tax work	ccoh	Opus 4.7 high
/treasury	Job applications	ccom	Opus 4.7 medium
/daedalus	Daily opener	ccom	Opus 4.7 medium
/pitch	Proposal writing	ccom	Opus 4.7 medium
/meridian	Bot fix	ccom	Opus 4.7 medium
/atrium	Website work	ccom	Opus 4.7 medium
/meister	Client deliverables	ccom	Opus 4.7 medium
/muse	Content creation	ccom	Opus 4.7 medium
/personalos	Brand work	ccom	Opus 4.7 medium
/solmere	Game design	ccom	Opus 4.7 medium
/vantabit	Studio operations	ccom	Opus 4.7 medium
/synqr	Business operations	ccom	Opus 4.7 medium
/audit	Domain audit	ccom	Opus 4.7 medium
/nexus	Meta-dashboard	ccom	Opus 4.7 medium
/titan	Bot tuning	ccol	Opus 4.7 low
/sora	Health bot tuning	ccol	Opus 4.7 low
/angie	PA bot tuning	ccol	Opus 4.7 low
/vega	Uber bot tuning	ccol	Opus 4.7 low
/kai	Finance bot	ccol	Opus 4.7 low
/cortex	AI memory work	ccol	Opus 4.7 low
/talon	Crypto bot	ccol	Opus 4.7 low
/lore	Book bot	ccol	Opus 4.7 low
/hermes	Router bot	ccol	Opus 4.7 low
/basket	Shopping bot	ccol	Opus 4.7 low
/oxide	Stack config	ccol	Opus 4.7 low
/utilities	Tool work	ccol	Opus 4.7 low
/wabisabi	Personal practice	ccol	Opus 4.7 low
/n8n	Workflow work	ccol	Opus 4.7 low
/newbot	Creating new domain	ccol	Opus 4.7 low
/obsidian	Vault work	ccol	Opus 4.7 low
/handoff	Status aggregation	ccsm	Sonnet medium
/weekly	Maintenance sweep	ccsm	Sonnet medium
/warframe	Game state / farming	ccsl	Sonnet low
/eve	Fitting / exploration	ccsm	Sonnet medium
/crypto	Market lookup	ccsm	Sonnet medium

Low-context mode - one tier down

Trigger when context is over 70%. Each command drops one effort tier. 20-30% longer conversations for 40-50% fewer tokens per turn.

Domain	Task type	Alias	Model + effort
/darius	Architecture design, ADRs	ccom	Opus 4.7 medium
/cyrus	Psychological session	ccom	Opus 4.7 medium
/fix	Hard debugging	ccom	Opus 4.7 medium
/forge	Content drafting	ccom	Opus 4.7 medium
/jung	Analyst session	ccom	Opus 4.7 medium
/damian	Profile deepening	ccom	Opus 4.7 medium
/treasury	Tax work	ccom	Opus 4.7 medium
/treasury	Job applications	ccol	Opus 4.7 low
/daedalus	Daily opener	ccol	Opus 4.7 low
/pitch	Proposal writing	ccol	Opus 4.7 low
/meridian	Bot fix	ccol	Opus 4.7 low
/atrium	Website work	ccol	Opus 4.7 low
/meister	Client deliverables	ccol	Opus 4.7 low
/muse	Content creation	ccol	Opus 4.7 low
/personalos	Brand work	ccol	Opus 4.7 low
/solmere	Game design	ccol	Opus 4.7 low
/vantabit	Studio operations	ccol	Opus 4.7 low
/synqr	Business operations	ccol	Opus 4.7 low
/audit	Domain audit	ccol	Opus 4.7 low
/titan	Bot tuning	ccsm	Sonnet medium
/sora, /angie, /vega, /kai, /cortex, /talon, /lore, /hermes, /basket, /oxide, /utilities, /wabisabi, /n8n, /newbot, /obsidian	Light work	ccsm	Sonnet medium
/handoff, /weekly, /eve, /crypto	Routine	ccsl	Sonnet low
/warframe	Game state / farming	ccsl	Sonnet low (floor)

Agent tier decision tree

Do I need an agent at all?

Scope under 5 files? Answer under 50 lines? Exact file known? Read directly. Otherwise fire an agent.

What kind of work?

Verbatim retrieval, quick factual check, state lookup - Scout (Haiku).

Research, pattern finding, multi-source comparison - Analyst Low or Medium (Sonnet).

Mechanical edits with verification, precise spec - Builder Low or Medium (Sonnet).

Execution with local judgment calls, trade-offs not obvious - Engineer (Opus 4.7 low).

Design, deep research, ADR writing, system-level choices - Architect Medium or High (Opus 4.7).

Escalation path

Scout fails - try Analyst Low - try Analyst Medium

Builder returns scope-exceeded - try Engineer - try Architect Medium - try Architect High

Full spec at ~/oxide/AGENT-PATTERNS.md. Six discipline rules apply to every agent call. Schema versioning + dual tool budget + drift re-injection built in.

Interchange guide

Stay in normal mode when

Context under 50% full
Session is discussion-heavy, quality matters
Architecture, psychology, content voice, tax accuracy, anything where wrong is expensive

Flip to low-context when

Context over 70% full
Session running long and you want it to finish
Routine work where 4.7 low is plenty
Follow-up question where context already has the answer

Signals to watch

Claude Code warning about context pressure
Responses getting slower or chunkier
low-context-suggest-hook fires