AgentSkillsCN

maceff-experimental-self-motivation

当马尔可夫推荐器建议 BUILD 模式，或当实验已验证假设且有必要实施时，在关口调用。设置 BUILD 工作模式。在 AUTO_MODE 下，构建通常紧随实验之后——原型生活在实验 CA 中。

SKILL.md
--- frontmatter
name: maceff-experimental-self-motivation
description: "Invoke at gate points when the Markov recommender suggests EXPERIMENT mode, or when a discovery needs structured investigation before building. Sets EXPERIMENT work mode and guides hypothesis-driven task creation."
allowed-tools: Bash, Read, Grep, Glob, Write, Edit

Transition into EXPERIMENT mode — structured investigation with hypotheses and protocols.

When to Invoke

  • •Markov recommender suggests transition to EXPERIMENT at a gate point
  • •A discovery needs validation before committing to implementation
  • •An idea should be investigated rather than immediately built
  • •You need empirical data before making architectural decisions

Policy Engagement

bash
macf_tools policy navigate mode_system
macf_tools policy navigate experiments

Read the sections that answer: "What are work modes?" and "What makes a good experiment?"

Set Work Mode

bash
macf_tools events append work_mode_change '{"mode": "EXPERIMENT"}'

ULTRATHINK Reflection

Before creating tasks, pause and think deeply:

  1. •What hypothesis needs testing? What assumption am I making that data could confirm or refute?
  2. •What's the simplest experiment that would answer this? Prefer 30-minute smoke tests over multi-phase protocols.
  3. •What would I build IF the hypothesis is confirmed? Knowing the downstream action focuses the experiment.
  4. •Can I reuse an existing prototype? Experiment CAs can contain prototypes as applied learnings.
  5. •What ideas from the bank are ready for investigation? Check macf_tools idea list --status captured for experiment candidates.

Execution

  1. •Record the transition in task notes: "EXPERIMENT mode activated via maceff-experimental-self-motivation. Investigating: {hypothesis}."
  2. •Create an experiment task (or use existing): macf_tools task create experiment "Title"
  3. •Write a quick protocol or hypothesis statement
  4. •Execute: gather data, run tests, analyze results
  5. •Prototypes go INSIDE the experiment CA directory (not standalone repos)
  6. •Capture findings as learnings when complete

Accountability

If you were recommended EXPERIMENT but chose a different mode, you MUST justify in task notes: "Recommender suggested EXPERIMENT (p=X%). Override: chose {MODE} because {reason}."

Anti-Pattern: The Premature Build

EXPERIMENT exists to GATE building behind validation. If you find yourself writing production code instead of gathering data — you skipped experimentation. Prototypes within experiments are fine; committing to project repos before validation is not.