moralespanitz/internal/embed/claude/skills/getting-started

SKILL: Getting Started

The bootstrap entry point for the Research Loop framework. Every research session starts here. This skill tells you what skills exist, when to use them, and what the mandatory workflow is.

---

You have skills. They give you superpowers.

<session-start-hook> <EXTREMELY_IMPORTANT>

You are operating inside Research Loop — an Agent OS for scientific research.

RIGHT NOW, read this file fully before doing anything else.

Core rules:

1. You have skills. They give you research superpowers.
2. Before any significant activity — selecting an idea, running experiments, writing, or preparing for submission — search for the relevant skill and follow it.
3. If a skill exists for what you are doing, you MUST follow it. Skills are not optional suggestions.
4. For new investigations: use idea-selection → execution → writing-papers → review-prep in order. Never jump ahead.
5. Never begin experiments before idea-selection has produced a written, checked hypothesis.
6. Never begin writing before execution has produced a complete, annotated experiment record.

To list all available skills:

ls skills/*/SKILL.md

To read a skill:

cat skills/<name>/SKILL.md

</EXTREMELY_IMPORTANT> </session-start-hook>

---

The mandatory research workflow

Every investigation follows this sequence. No shortcuts.

idea-selection
    │
    │  Output: hypothesis.md with core claim, uniqueness axis, best-case conclusion
    │
    ▼
execution
    │
    │  Output: autoresearch.jsonl with full experiment record, knowledge_graph.md with causal annotations
    │
    ▼
writing-papers
    │
    │  Output: complete draft with one-sentence core idea, self-contained figures, story-structured intro, conclusion that answers "so what?"
    │
    ▼
review-prep
    │
    │  Output: submitted paper + rejection protocol documented in lab_notebook.md
    ▼

---

Available skills

| Skill | When to use | Invoke with | |-------|-------------|-------------| | idea-selection | At the start of any new investigation; when evaluating whether to continue a stalled one | /idea-selection | | execution | Throughout the experiment loop; at every checkpoint to kill/pivot/continue | /execution | | writing-papers | When transitioning from experiments to drafting; when reviewing a draft | /writing-papers | | review-prep | Before submission; when a rejection arrives; when deciding whether to resubmit | /review-prep |

---

What to do right now

If you are starting a new investigation:

1. Read skills/idea-selection/SKILL.md
2. Complete the idea-selection checklist
3. Write hypothesis.md with the output
4. Only then: read skills/execution/SKILL.md and start the experiment loop

If you are resuming an investigation:

1. Read knowledge_graph.md — understand the full state of what has been tried and why
2. Read lab_notebook.md — understand where the investigation was left
3. Read autoresearch.jsonl — identify the last completed loop state
4. Apply the execution skill to decide: continue, pivot, or kill?

If you are starting to write:

1. Read skills/writing-papers/SKILL.md fully before writing a single sentence
2. Write the one-sentence core idea at the top of the draft document
3. Write the best-case conclusion before the introduction
4. Then write the introduction

If you are preparing for submission:

1. Read skills/review-prep/SKILL.md
2. Complete the self-review checklist
3. Run @reviewer for methodology audit
4. Document the rejection protocol in lab_notebook.md before submitting

---

State of the current workspace

To understand where you are, read these files in order:

hypothesis.md           # What you are trying to prove
knowledge_graph.md      # What has been tried and what was learned
lab_notebook.md         # Human-readable log of the investigation
autoresearch.jsonl      # Machine-readable checkpoint record

If none of these exist, you are starting fresh. Begin with idea-selection.

---

The underlying philosophy (Carlini, 2026)

The framework encoded in these skills is based on one principle:

Write papers with the goal of having an impact. That is what matters, is entirely under your control, and is lots of fun.

A best paper award is one sample from a distribution. You do not control the sampling process — that is determined by the committee, the timing, and who else submitted. What you control is the distribution.

Every skill in this framework is designed to shift that distribution: toward work that is important, well-executed, and clearly communicated. The award, if it comes, is someone noticing where your distribution ended up.

Focus on the distribution.