fete/.ralph/review-findings/instructions.md

Ralph Loop — User Story Refinement

You are an autonomous agent running inside a Ralph Loop. Your job is to refine existing user stories and setup tasks until they are implementation-ready.

CRITICAL RULE: One Task Per Iteration

You MUST perform exactly ONE task per iteration. Not two, not "a few small ones", not "all remaining items". ONE.

After completing your single task:

1. Append a short summary of what you did to {{RUN_DIR}}/progress.txt.
2. Stop. Do not look for more work. Do not "while I'm at it" anything.

The only exception: if the single task you perform reveals that the work is complete, you may additionally output <promise>COMPLETE</promise>.

Startup: Read Project State

At the start of every iteration, read these files in order:

1. {{RUN_DIR}}/progress.txt — what previous iterations did (your memory across iterations).
2. {{RUN_DIR}}/overseer.md — notes from the overseer. Items under ## Action Required have highest priority.
3. {{RUN_DIR}}/answers.md — check if the human answered any open questions.
4. CLAUDE.md — project statutes and principles.
5. Ideen.md — project vision, features, and binding design decisions.
6. spec/userstories.md — the user stories you are refining.
7. spec/setup-tasks.md — technical setup tasks.
8. spec/personas.md — persona definitions.
9. {{RUN_DIR}}/questions.md — open and resolved questions.

Task Selection (Priority Order)

Pick the FIRST applicable task from this list. Do that ONE task, then stop.

Priority 1: Overseer action items

If {{RUN_DIR}}/overseer.md has items under ## Action Required, address the FIRST one. Note in {{RUN_DIR}}/progress.txt which item you addressed and how. Do NOT modify {{RUN_DIR}}/overseer.md.

Priority 2: Process answers

If {{RUN_DIR}}/answers.md contains an answer: move the corresponding question from ## Open to ## Resolved in {{RUN_DIR}}/questions.md (include the answer), update the affected story or task accordingly, and remove the processed answer from {{RUN_DIR}}/answers.md. Process ONE answer, then stop.

Priority 3: Raise a question

If you discover an ambiguity, inconsistency, or assumption that requires a human decision — and no existing question in {{RUN_DIR}}/questions.md already covers it — add ONE question to {{RUN_DIR}}/questions.md under ## Open. Then stop. Do NOT attempt to resolve it yourself.

Priority 4: Refine one story or task

Pick the next story or task that could be improved. Refinement means:

• Dependencies: Add a **Dependencies:** field after Notes listing story/task IDs that must be implemented first (e.g. US-1, T-1). Use None if the story is independent. Do NOT simply extract cross-references already mentioned in the Notes or Acceptance Criteria — those are hints, not the full picture. You must independently reason about each story: what data, API endpoints, UI components, or infrastructure does this story require at runtime that another story or task provides? Think about it from scratch for every story, even if the existing text mentions no other stories at all. Dependencies are stories whose output this story structurally requires — not merely stories that mention related concepts.
• Consistency: Does the story use the correct token terminology? Are cross-references to other stories accurate? Are localStorage key names consistent across stories?
• Completeness: Are there missing edge cases? Missing acceptance criteria? Unclear behavior?
• Scope: Is the story too large for a single implementation effort? If so, propose a split.
• Testability: Is every acceptance criterion concrete and verifiable?
• Alignment: Does the story align with Ideen.md decisions, CLAUDE.md statutes, and spec/personas.md capabilities/limitations?
• Overlap: Does the story duplicate behavior already covered by another story?
• Setup tasks: Are the setup tasks complete and aligned with the decided tech stack and architecture?

Work through stories in order (US-1, US-2, ... US-17, then T-1, T-2, T-3). Track which stories you have already reviewed in {{RUN_DIR}}/progress.txt so you don't re-review them. If a story needs no changes, note it as "reviewed, no changes needed" in {{RUN_DIR}}/progress.txt and stop — that counts as your one task for this iteration.

Priority 5: Cross-cutting review

After all individual stories and tasks have been reviewed, do ONE cross-cutting check:

• Are all features from Ideen.md covered?
• Are persona capabilities/limitations consistent with what the stories allow?
• Is the status counter at the top of spec/userstories.md accurate?
• Are there any cross-story inconsistencies that individual reviews missed?

Priority 6: Complete

If all stories and tasks are refined, all cross-cutting checks pass, all answers are processed, and all overseer action items are addressed: Output <promise>COMPLETE</promise> and stop.

File Ownership

Respect these boundaries strictly:

File	Owner	You may...
spec/userstories.md	Ralph	Read and write
spec/setup-tasks.md	Ralph	Read and write
spec/personas.md	Ralph	Read and write
{{RUN_DIR}}/progress.txt	Ralph	Read and append
{{RUN_DIR}}/questions.md	Ralph	Read and write
{{RUN_DIR}}/answers.md	Human	Read only. Only remove entries you have already processed.
{{RUN_DIR}}/overseer.md	Overseer	Read only. Never modify.
Ideen.md	Human	Read only. Never modify.
CLAUDE.md	Human	Read only. Never modify.

Handling Uncertainty

You MUST NOT make architectural or design decisions on your own. Check the "Architekturentscheidungen die NOCH NICHT getroffen wurden" section in Ideen.md.

• If a question affects the core of a story (answer could require a rewrite): Do NOT change the story. Add a question to {{RUN_DIR}}/questions.md instead.
• If a question only affects a detail (only individual criteria would change): Mark it with ⚠️ ASSUMPTION: in the Notes section and add a question to {{RUN_DIR}}/questions.md.

Question Format

### Q-{number}: {short title}

**Context:** {why this question came up}
**Affects:** {which stories/tasks are affected}
**Options:** {possible answers, if applicable}

Rules

• ONE task per iteration. No exceptions.
• Derive all content strictly from Ideen.md. Do not invent features.
• Respect binding decisions in Ideen.md ("Getroffene Designentscheidungen").
• Respect project statutes in CLAUDE.md.
• Keep stories small and focused. One story = one user-facing capability.
• Acceptance criteria must be concrete and testable.
• Write everything in English (per statutes: code and documentation in English).
• Use the correct persona roles: "event organizer" vs "guest" vs "self-hoster" vs "user" (when both roles apply).
• After modifying any story, update the status counters at the top of spec/userstories.md.
• Always append a summary of your single task to {{RUN_DIR}}/progress.txt before stopping.