Skip to content

CLAUDE.md

Latest commit

 

History

History
72 lines (52 loc) · 2.37 KB

CLAUDE.md

File metadata and controls

72 lines (52 loc) · 2.37 KB
Raw

scafld Claude Contract

Read AGENTS.md first. It owns the full scafld contract.

Default Flow

scafld plan <task-id> --title "Title"
scafld harden <task-id>
scafld approve <task-id>
scafld build <task-id>
scafld review <task-id>
scafld complete <task-id>
scafld status <task-id>
scafld handoff <task-id>

Boundaries

  • Use scafld harden to strengthen the draft before approval.
  • Use scafld build to open one phase, then run it again after implementation to record evidence.
  • Use scafld review as the adversarial gate.
  • Use scafld status --json for automation.
  • Use scafld handoff for compact model context without moving state.

For real review, use --provider claude or --provider codex. --provider local is smoke-test only and cannot satisfy complete.

Inside the scafld repo, use ./bin/scafld or go run ./cmd/scafld; do not use a copied compiled binary.

Claude Code Integration Notes

aster is a small governance repo, not a product app. Most work here is in docs, GitHub workflows, helper scripts, and scafld-managed specs.

MUST READ: AGENTS.md - the canonical agent guide covering invariants, modes, validation, and conventions. Read it before doing any work.

Tool Usage

  • Always use Read before Edit to understand existing code and ensure correct string matching.
  • Use Grep and Glob for codebase exploration instead of bash find/grep.
  • Prefer Edit (targeted replacement) over Write (full file overwrite) for existing files.

Spec Management

Always use the scafld CLI for spec lifecycle management. Never manually move, copy, or rename spec files. Never manually change the status field.

Entering scafld Modes

  • Plan mode: Read .scafld/prompts/plan.md, then explore and generate a spec.
  • Exec mode: Read .scafld/prompts/exec.md, then load the approved spec and execute.
  • Review mode: Run scafld review <task-id> with an external adversarial provider. Read findings from scafld review, scafld status, or scafld handoff; do not write a separate review file.

Prompting Patterns

"Let's plan [feature]. Create a task spec."
"Execute the [task-id] spec."
"Review the [task-id] spec."
"Show me the current phase status."

Useful Commands

npm run check
npm run docs:build
RUNX_ROOT=/home/kam/dev/runx bash scripts/proving-ground.sh
scafld list
scafld status <task-id> --json