Agent Skill
2/7/2026implementing-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
| 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. |
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.mddocs/changes/<slug>/specs/*/spec.mddocs/changes/<slug>/design.md(optional)- Any other files in the change folder (specs take priority)
Behavior
- Study all files in
docs/changes/<slug>/, with emphasis on delta specs underspecs/. - Use
ralph-tui-create-jsonskill for prd.json structure, acceptance criteria formatting, and dependency conventions (we do not create prd.json from PRD.md). - 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.
- Coverage Map: Each story MUST note the requirement(s) it covers in the
notesfield (e.g.,Covers: Req:Auth, Req:Shell). Ensure every requirement in the delta specs is covered by at least one story. - Acceptance Criteria: Use spec-derived behavioral criteria. Do NOT include implementation-specific details (e.g., "Use ProviderRegistry") unless they are explicitly in the spec.
- Quality Gates: Append standard quality gates to every story:
pnpm lintpassespnpm testpasses (require "Tests added/updated for new behavior" and "Coverage >=90% not regressed" unless justified innotes)pnpm buildpasses
- Infer story split and dependencies from proposal + deltas. Only ask if unclear.
- 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.
- Infer a short description from
proposal.md. - Produce
docs/changes/<slug>/prd.jsonusing the ralph-tui JSON schema only when user chooses Ralph TUI path. - Write prd.json at
docs/changes/<slug>/prd.json(not a tasks/ folder). - Stories must cover implementation only; do not add stories for creating specs that already exist in
docs/changes/<slug>/specs/. - Ask the user to review prd.json.
- 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
- Remind the user that ralph-tui creates per-story commits; offer to squash before merging.
- Implementation is done only via the selected run/implement mode (AFK/HITL/agentic), not eagerly.
- Remind the user to run ralph-tui outside opencode.
Modes
- AFK loop: run
ralph-tui run --prd docs/changes/<slug>/prd.jsonoutside opencode. - HITL loop: run one iteration outside opencode, review, adjust spec, repeat.
Skills Info
Original Name:implementing-with-ralph-tuiAuthor:drevantonder
Download