Agent Skill
2/7/2026

implementing-with-ralph-tui

Use this skill to orchestrate autonomous implementation loops with ralph-tui. It guides you through creating a bite-sized, spec-aligned task list (prd.json) and setting up the execution environment.

D
drevantonder
1GitHub Stars
1Views
npx skills add drevantonder/openfinance

SKILL.md

Nameimplementing-with-ralph-tui
DescriptionUse this skill to orchestrate autonomous implementation loops with ralph-tui. It guides you through creating a bite-sized, spec-aligned task list (prd.json) and setting up the execution environment.

name: implementing-with-ralph-tui description: Use this skill to orchestrate autonomous implementation loops with ralph-tui. It guides you through creating a bite-sized, spec-aligned task list (prd.json) and setting up the execution environment.

Implementing with Ralph TUI

Use when executing a change via ralph-tui.

Inputs

  • docs/changes/<slug>/proposal.md
  • docs/changes/<slug>/specs/*/spec.md
  • docs/changes/<slug>/design.md (optional)
  • Any other files in the change folder (specs take priority)

Behavior

  1. Study all files in docs/changes/<slug>/, with emphasis on delta specs under specs/.
  2. Use ralph-tui-create-json skill for prd.json structure, acceptance criteria formatting, and dependency conventions (we do not create prd.json from PRD.md).
  3. Story Sizing: Each story must be bite-sized and completable in ONE ralph-tui iteration (~one agent context window). If a requirement is large, split it into multiple stories.
  4. Coverage Map: Each story MUST note the requirement(s) it covers in the notes field (e.g., Covers: Req:Auth, Req:Shell). Ensure every requirement in the delta specs is covered by at least one story.
  5. Acceptance Criteria: Use spec-derived behavioral criteria. Do NOT include implementation-specific details (e.g., "Use ProviderRegistry") unless they are explicitly in the spec.
  6. Quality Gates: Append standard quality gates to every story:
    • pnpm lint passes
    • pnpm test passes (require "Tests added/updated for new behavior" and "Coverage >=90% not regressed" unless justified in notes)
    • pnpm build passes
  7. Infer story split and dependencies from proposal + deltas. Only ask if unclear.
  8. Set priorities using the ralph-tui convention (lower = higher):
    • 1 = highest priority
    • 2 = medium priority (default)
    • 3 = lower priority
    • 4+ = backlog Do not use incremental priorities; they should be semantic, not order-based.
  9. Infer a short description from proposal.md.
  10. Produce docs/changes/<slug>/prd.json using the ralph-tui JSON schema only when user chooses Ralph TUI path.
  11. Write prd.json at docs/changes/<slug>/prd.json (not a tasks/ folder).
  12. Stories must cover implementation only; do not add stories for creating specs that already exist in docs/changes/<slug>/specs/.
  13. Ask the user to review prd.json.
  14. Before running ralph-tui, offer to help create a worktree using gtr-git-worktree-runner:
git gtr new <branch>
cd "$(git gtr go <branch>)"
ralph-tui run --prd docs/changes/<slug>/prd.json
  1. Remind the user that ralph-tui creates per-story commits; offer to squash before merging.
  2. Implementation is done only via the selected run/implement mode (AFK/HITL/agentic), not eagerly.
  3. Remind the user to run ralph-tui outside opencode.

Modes

  • AFK loop: run ralph-tui run --prd docs/changes/<slug>/prd.json outside opencode.
  • HITL loop: run one iteration outside opencode, review, adjust spec, repeat.
Skills Info
Original Name:implementing-with-ralph-tuiAuthor:drevantonder