prime
Bootstrap session context by reading READMEs, listing files, and building a mental model of the current project.
SKILL.md
Full skill instructions
Delegate the priming work to a Haiku agent. Do not run the reads/listings in the main thread.
Spawn one Agent with:
subagent_type: "general-purpose"model: "haiku"- A self-contained prompt instructing the agent to gather the context below and return a concise summary.
The agent's prompt should cover:
- List files in the current working directory (non-recursively first, then one level deep).
- Find and read any README, README.md, CONTRIBUTING.md, or similar documentation files in the current directory.
- Check for and read common project markers if present:
- package.json, Cargo.toml, pyproject.toml, go.mod, Makefile, justfile, flake.nix, shell.nix
- .claude/settings.json, .claude/CLAUDE.md, CLAUDE.md
- .envrc, .tool-versions, mise.toml
- If this is a git repo, run
git log --oneline -10to see recent activity.
Ask the agent to return a concise summary covering:
- What this project is and what it does
- Key directories and their purpose
- Language/framework/tooling in use
- Any conventions or patterns visible from the config files
- Recent git activity
Relay the agent's summary to the user. Keep it dense and useful. Skip anything that adds no value.
