Skills¶

Skills are Claude Code slash commands bundled in the lightcone-cli plugin. They give the agent a structured, phase-by-phase workflow for the most common research operations.

If you're a researcher trying to use these, the Claude Code Workflow page in the user guide is the friendly version. This page is for maintainers.

Available skills¶

The /lc-from-* family is parallel by what you start from: a question, code, or a paper. /lc-from-paper is the entry point of a six-skill paper-reproduction bundle; the five bundle siblings stand alone and are user-invokable directly.

Project lifecycle¶

Paper-reproduction bundle (sibling skills)¶

Co-located with lc-from-paper so a single lc init brings the full toolkit. Each stands alone and is user-invokable; lc-from-paper dispatches them by role during the reproduction.

See the bundle README for the rationale behind co-location vs plugin install.

Reference skills (auto-primed via session-start)¶

Not direct entry points — invoked by other skills (or by Claude directly when relevant) to load reference content into the working session. The session-start hook names both in its primer, so Claude is aware they exist from the first turn.

These intentionally don't appear in the top-level README — researchers use the project-lifecycle skills directly; the reference skills are infrastructure.

How a skill is wired¶

Each skill is a claude/lightcone/skills/<name>/SKILL.md file with YAML frontmatter:

---
name: lc-new
description: >
  Scope a new ASTRA analysis from a research question...
allowed-tools: Read, Write(astra.yaml), Edit(astra.yaml), Glob, Grep, Bash(astra:*), ...
argument-hint: "[DESCRIPTION]"
---

The frontmatter configures Claude Code: which tools the skill may invoke, and what the slash command's argument hint looks like. The body is the prompt — phase definitions, rules, references to guide files, anti-patterns. The skill bundles its own helper scripts under scripts/ and its loop prompt template under assets/ when relevant.

Plugin layout¶

.claude-plugin/marketplace.json         # marketplace manifest at repo root

claude/lightcone/                        # plugin root
├── .claude-plugin/plugin.json           # plugin manifest (name, version, …)
├── skills/
│   ├── lc-new/{SKILL.md, references/*.md}
│   ├── lc-from-code/SKILL.md
│   ├── lc-from-paper/{SKILL.md, references/*.md, templates/{constitution.md, CLAUDE.md}}
│   ├── lc-feedback/SKILL.md
│   ├── ralph/{SKILL.md, references/*.md, scripts/ralph}
│   ├── paper-extraction/{SKILL.md, scripts/*.py}
│   ├── narrative/{SKILL.md, references/*.md}
│   ├── figure-comparison/{SKILL.md, scripts/*.py}
│   ├── check-sentence-by-sentence/SKILL.md
│   ├── astra/SKILL.md                  # reference: astra.yaml spec
│   └── lc-cli/SKILL.md                 # reference: lc workflow
├── agents/lc-extractor.md              # literature subagent for /lc-new
├── hooks/hooks.json                    # hook config (SessionStart, PostToolUse)
├── scripts/*.sh                        # hook handlers (venv, validate-on-save, session-start primer)
└── templates/CLAUDE.md                 # the project CLAUDE.md template

Both .claude-plugin/marketplace.json and claude/lightcone/ are force-included into the wheel via pyproject.toml::tool.hatch.build.targets.wheel.force-include, so lc init can run claude plugin marketplace add against the wheel's install directory whether you're working from source or PyPI.

Other plugin files¶

The two reference skills (/astra and /lc-cli) live under skills/ and are listed in the Reference skills section above. Remaining plugin files:

Authoring a new skill¶

See Authoring Skills.