deep-work

Run an autonomous agent on a complex task with quality evaluation. Clarifies upfront, then works fully autonomously until the result meets the standard. Triggers: 'deep work', 'work on this seriously', 'high quality task', 'write a paper', 'complex task', 'do this properly', 'professional quality'.

Deep Work

Run an autonomous agent for complex, high-standard tasks. Clarifies any questions upfront, then works fully autonomously — plan, execute, evaluate, revise — until the result meets the quality standard.

agentic deep-work "<TASK>" --level <LEVEL>

Quality Levels

LevelStandard
high_schoolBasic correctness, simple structure
bachelorSolid understanding, proper methodology (default)
masterDepth of analysis, critical thinking, good writing
phdNovel contribution, rigorous methodology, publication-ready
professorExpert/authoritative, top-venue quality (NeurIPS/Nature/OSDI)

Examples

# PhD-level survey paper
agentic deep-work "Write a survey on context management in LLM agents, focus on compaction vs sub-agent trade-offs, target NeurIPS workshop" \
  --level phd

# Production code at master level
agentic deep-work "Build a REST API for user auth with OAuth2, include tests and docs" \
  --level master

# Expert-level analysis
agentic deep-work "Analyze system performance bottlenecks, provide data-backed recommendations" \
  --level professor

Options

FlagDescription
--level, -lQuality level: high_school / bachelor / master / phd / professor (default: bachelor)
--provider, -pLLM provider
--model, -mModel to use
--max-stepsMax total steps (default: 100)
--max-revisionsMax evaluation-revision cycles (default: 5)
--no-interactiveSkip clarification, start immediately