1---2name: ralph3description: Self-referential loop until task completion with architect verification4---56[RALPH + ULTRAWORK - ITERATION {{ITERATION}}/{{MAX}}]78Your previous attempt did not output the completion promise. Continue working on the task.910<Purpose>11Ralph is a persistence loop that keeps working on a task until it is fully complete and architect-verified. It wraps ultrawork's parallel execution with session persistence, automatic retry on failure, and mandatory verification before completion.12</Purpose>1314<Use_When>15- Task requires guaranteed completion with verification (not just "do your best")16- User says "ralph", "don't stop", "must complete", "finish this", or "keep going until done"17- Work may span multiple iterations and needs persistence across retries18- Task benefits from parallel execution with architect sign-off at the end19</Use_When>2021<Do_Not_Use_When>22- User wants a full autonomous pipeline from idea to code -- use `autopilot` instead23- User wants to explore or plan before committing -- use `plan` skill instead24- User wants a quick one-shot fix -- delegate directly to an executor agent25- User wants manual control over completion -- use `ultrawork` directly26</Do_Not_Use_When>2728<Why_This_Exists>29Complex tasks often fail silently: partial implementations get declared "done", tests get skipped, edge cases get forgotten. Ralph prevents this by looping until work is genuinely complete, requiring fresh verification evidence before allowing completion, and using explicit architect native-subagent verification to confirm quality.30</Why_This_Exists>3132<Execution_Policy>33- Fire independent agent calls simultaneously -- never wait sequentially for independent work34- Use `run_in_background: true` for long operations (installs, builds, test suites)35- The documented-leader preflight is not a general native-session gate. Run `omx ralplan preflight --json` only when native role routing reports `role_routing_unavailable` and Ralph attempts adapted Ralplan Planner, Architect, or Critic authority, adapted role-intent, or adapted consensus authority. On `unsupported_documented_leader_proof`, stop before that adapted authority and use a Codex surface with documented root proof or a reviewed alternative workflow. Do not infer root authority from `session_id`, undocumented `thread_id`, session/pointer/transcript/cwd state, absent child data, or prompt labels. Ordinary native planning, lifecycle, state, status, health, HUD, runtime, setup, install, sync, and unrelated delegation remain outside this preflight boundary and under their existing controls.36- When the native surface exposes `agent_type` role routing, set `agent_type` to an installed OMX role and never omit it for OMX work; use `reasoning_effort` for per-dispatch intensity when needed.37- **OMX adapted role-pass protocol:** when native routing is `role_routing_unavailable`, do not fabricate `agent_type`. On the exact reviewed Codex releases 0.144.5, 0.145.0, 0.146.1, and 0.148.0-alpha.5, only an attempted adapted Ralplan Planner, Architect, Critic, role-intent, or consensus authority path requires `omx ralplan preflight --json` and fails closed on `unsupported_documented_leader_proof`; every other version remains unknown and fails closed. Do not use prompt labels, task-name carriers, pending intents, markers, or `omx ralplan role-intent write` as substitutes.38- Preserve legacy Ralph tier intent through native reasoning effort: LOW -> `low`, STANDARD -> `medium`, THOROUGH -> `xhigh`39- Deliver the full implementation: no scope reduction, no partial completion, no deleting tests to make them pass40- Apply the shared workflow guidance pattern: outcome-first framing, concise visible updates for multi-step execution, local overrides for the active workflow branch, validation proportional to risk, explicit stop rules, and automatic continuation for safe reversible steps. Ask only for material, destructive, credentialed, external-production, or preference-dependent branches.41- Integrate with Codex goal mode when goal tools are available: inspect the active thread goal with `get_goal`, preserve it as the top-level stop condition, and only call `update_goal({status: "complete"})` after a Ralph completion audit proves the objective is actually achieved.42</Execution_Policy>4344<Steps>450. **Pre-context intake (required before planning/execution loop starts)**:46 - Assemble or load a context snapshot at `.omx/context/{task-slug}-{timestamp}.md` (UTC `YYYYMMDDTHHMMSSZ`).47 - Minimum snapshot fields:48 - task statement49 - desired outcome50 - known facts/evidence51 - constraints52 - unknowns/open questions53 - likely codebase touchpoints54 - If an existing relevant snapshot is available, reuse it and record the path in Ralph state.55 - If request ambiguity is high, gather brownfield facts first. `omx explore` is deprecated; use normal repository inspection tools/subagents for simple read-only repository lookups and `omx sparkshell` only for explicit shell-native read-only evidence. Then run `$deep-interview --quick <task>` to close critical gaps.56 - Do not begin Ralph execution work (delegation, implementation, or verification loops) until snapshot grounding exists. If forced to proceed quickly, note explicit risk tradeoffs.57 - A Ralplan-originated handoff alone does not require preflight. Before intake, run `omx ralplan preflight --json` only when native role routing is unavailable and the handoff attempts adapted Ralplan Planner, Architect, Critic, role-intent, or consensus authority. On `unsupported_documented_leader_proof`, record the reason in Execution Policy and stop before that authority; otherwise follow the existing intake controls.581. **Review progress**: Check TODO list and any prior iteration state592. **Continue from where you left off**: Pick up incomplete tasks603. **Delegate in parallel**: Route tasks to specialist native agents with explicit `agent_type` and appropriate `reasoning_effort`61 - Simple lookups: `reasoning_effort="low"` -- "What does this function return?"62 - Standard work: `reasoning_effort="medium"` -- "Add error handling to this module"63 - Complex analysis: `reasoning_effort="xhigh"` -- "Debug this race condition"64 - When Ralph is entered as a ralplan follow-up, start from the approved **available-agent-types roster** and make the delegation plan explicit: implementation lane, evidence/regression lane, and final sign-off lane using only known agent types654. **Run long operations in background**: Builds, installs, test suites use `run_in_background: true`665. **Visual task gate (when screenshot/reference images are present)**:67 - Run the Visual Ralph verdict step **before every next edit**.68 - Require structured JSON output: `score`, `verdict`, `category_match`, `differences[]`, `suggestions[]`, `reasoning`.69 - Persist verdict to `.omx/state/{scope}/ralph-progress.json` including numeric + qualitative feedback.70 - Default pass threshold: `score >= 90`.71 - **URL-based visual cloning tasks**: When the task description contains a target URL (e.g., "clone https://example.com"), route the work through `$visual-ralph`. `$web-clone` is hard-deprecated; Visual Ralph owns the migrated live-URL visual implementation use case and uses its built-in visual verdict step for measured visual scoring.726. **Verify completion with fresh evidence**:73 - If Codex goal mode is available, call `get_goal` before final verification to restate the active objective and include it in the evidence checklist.74 a. Identify what command proves the task is complete75 b. Run verification (test, build, lint)76 c. Read the output -- confirm it actually passed77 d. Check: zero pending/in_progress TODO items787. **Architect verification** (native role):79 - <5 files, <100 lines with full tests: `task(agent_type="architect", reasoning_effort="medium", prompt="...")` minimum80 - Standard changes: `task(agent_type="architect", reasoning_effort="medium", prompt="...")`81 - >20 files or security/architectural changes: `task(agent_type="architect", reasoning_effort="xhigh", prompt="...")`82 - Ralph floor: always run an explicit `architect` native subagent, even for small changes83 - On `role_routing_unavailable`, do not invoke `omx ralplan role-intent write` or manufacture an Architect identity. Run the documented-leader preflight only if Architect verification would attempt adapted Ralplan Architect authority; on the exact reviewed Codex releases 0.144.5, 0.145.0, 0.146.1, and 0.148.0-alpha.5 that adapted path is unavailable, while every other version remains unknown and fails closed. Surface the leader-proof diagnostic and the supported-surface recovery guidance, then stop before that authority. Ordinary Ralph delegation remains under its existing controls. Use an adapted route only after its documented positive root proof has been reviewed and implemented.847.5 **Mandatory Deslop Pass**:85 - After Step 7 passes, run `oh-my-codex:ai-slop-cleaner` on **all files changed during the Ralph session**.86 - Scope the cleaner to **changed files only**; do not widen the pass beyond Ralph-owned edits.87 - Run the cleaner in **standard mode** (not `--review`).88 - If the prompt contains `--no-deslop`, skip Step 7.5 entirely and proceed with the most recent successful verification evidence.897.6 **Regression Re-verification**:90 - After the deslop pass, re-run all tests/build/lint and read the output to confirm they still pass.91 - If post-deslop regression fails, roll back cleaner changes or fix and retry. Then rerun Step 7.5 and Step 7.6 until the regression is green.92 - Do not proceed to completion until post-deslop regression is green (unless `--no-deslop` explicitly skipped the deslop pass).938. **On approval**: If Codex goal mode is active, call `update_goal({status: "complete"})` before `/cancel`; report final elapsed time and token-budget usage when the tool returns it. Then run `/cancel` to cleanly exit and clean up all state files.949. **On rejection**: Fix the issues raised, then re-verify with the same `agent_type` and `reasoning_effort` profile95</Steps>9697<Tool_Usage>98- Use `ask_codex` with `agent_role: "architect"` for verification cross-checks when changes are security-sensitive, architectural, or involve complex multi-system integration99- Skip Codex consultation for simple feature additions, well-tested changes, or time-critical verification100- If MCP compatibility tools are unavailable, proceed with CLI/agent verification alone -- never block on external tools101- Use `omx state write/read --input '<json>' --json` for ralph mode state persistence between iterations102- Use Codex goal tools when present: `get_goal` to discover or re-check the active objective, `create_goal` only when the user/system explicitly requested a new goal and no active goal exists, and `update_goal` only after the audited objective is fully achieved.103- Persist context snapshot path in Ralph mode state so later phases and agents share the same grounding context104- Prefer CLI state commands. If an explicit MCP compatibility `omx_state` call reports that its stdio transport is unavailable/closed, do **not** retry the same MCP call. Retry once through the supported CLI parity surface with the same payload, preserving `workingDirectory` and `session_id`: `omx state write --input '<json>' --json`, `omx state read --input '<json>' --json`, or `omx state clear --input '<json>' --json`. If the CLI path also fails, continue with `.omx/context` / `.omx/plans` file-backed artifacts and report the state persistence blocker.105</Tool_Usage>106107## Goal Mode Integration108109Codex goal mode is the thread-level completion contract for long-running Ralph work. Ralph state tracks workflow mechanics; goal mode tracks whether the user objective is truly done. When the goal tools are available:1101111. Call `get_goal` during intake or before the first execution loop when the prompt/hook says an active thread goal exists.1122. If no goal exists, call `create_goal` only when the user or system explicitly asked for goal tracking; otherwise continue with Ralph state alone.1133. Treat `goal.objective` as binding acceptance scope. Newer user updates can refine the current branch, but do not silently narrow the goal.1144. Before completion, perform a prompt-to-artifact checklist and completion audit against real evidence:115 - restate the objective as deliverables/success criteria116 - map every prompt requirement, named workflow (`$ralplan`, `$ralph`), file, command, test, gate, and deliverable to evidence117 - inspect the actual files, command output, state, and tests behind each checklist item118 - identify missing, weakly verified, or uncovered requirements and continue if any remain1195. Call `update_goal({status: "complete"})` only when the audit shows no required work remains. Do not use passing tests, Ralph state, or architect approval as proxy proof unless they cover the whole goal.1206. If goal tools are unavailable, keep working through Ralph state and mention the missing goal-mode evidence in the final report.121122## State Management123124Use the CLI-first state surface for Ralph lifecycle state (`omx state write/read/clear --input '<json>' --json`). Explicit MCP compatibility tools (`state_write`, `state_read`, `state_clear`) remain acceptable only when already enabled.125126- **On start**:127 `omx state write --input '{"mode":"ralph","active":true,"iteration":1,"max_iterations":10,"current_phase":"executing","started_at":"<now>","state":{"context_snapshot_path":"<snapshot-path>"}}' --json`128- **On each iteration**:129 `omx state write --input '{"mode":"ralph","iteration":<current>,"current_phase":"executing"}' --json`130- **On verification/fix transition**:131 `omx state write --input '{"mode":"ralph","current_phase":"verifying"}' --json` or `omx state write --input '{"mode":"ralph","current_phase":"fixing"}' --json`132- **On completion** (only after the completion audit passes with real evidence):133 `omx state write --input '{"mode":"ralph","active":false,"current_phase":"complete","completed_at":"<now>","completion_audit":{"passed":true,"prompt_to_artifact_checklist":["<requirement mapped to artifact/evidence>"],"verification_evidence":["<fresh test/build/lint command and result>"]}}' --json`134- **Before the final answer**:135 1. Run fresh verification and read the output.136 2. Build `prompt_to_artifact_checklist` entries that map every user requirement, workflow gate, named file, command, PR/delivery requirement, and stop condition to a concrete artifact or evidence item.137 3. Build `verification_evidence` entries with concrete commands, exit status, files inspected, PR URLs, or other machine-checkable evidence.138 4. Write the Ralph completion state with a top-level `completion_audit` field on the Ralph state object. Do not write bare top-level `prompt_to_artifact_checklist` or `verification_evidence` fields by themselves; the Stop gate will reject them.139 5. Read the state back with `omx state read --input '{"mode":"ralph"}' --json` and verify `completion_audit.passed === true`, a non-empty checklist, and non-empty verification evidence before producing the final answer.140 6. If Codex goal mode is active, call `update_goal({status:"complete"})` only after this Ralph audit read-back succeeds.141- **On cancellation/cleanup**:142 run `$cancel` (which should call `omx state clear --input '{"mode":"ralph"}' --json`)143144145## Scenario Examples146147**Good:** The user says `continue` after the workflow already has a clear next step. Continue the current branch of work instead of restarting or re-asking the same question.148149**Good:** The user changes only the output shape or downstream delivery step (for example `make a PR`). Preserve earlier non-conflicting workflow constraints and apply the update locally.150151**Bad:** The user says `continue`, and the workflow restarts discovery or stops before the missing verification/evidence is gathered.152153<Examples>154<Good>155Correct parallel delegation:156```157task(agent_type="executor", reasoning_effort="low", prompt="Add type export for UserConfig")158task(agent_type="executor", reasoning_effort="medium", prompt="Implement the caching layer for API responses")159task(agent_type="executor", reasoning_effort="xhigh", prompt="Refactor auth module to support OAuth2 flow")160```161Why good: Three independent tasks fired simultaneously while explicitly selecting the installed `executor` native role, so the UI/tracker does not show default subagents; legacy tier intent is preserved through native reasoning effort (`LOW` -> `low`, `STANDARD` -> `medium`, `THOROUGH` -> `xhigh`).162</Good>163164<Good>165Correct verification before completion:166```1671. Run: npm test → Output: "42 passed, 0 failed"1682. Run: npm run build → Output: "Build succeeded"1693. Run: lsp_diagnostics → Output: 0 errors1704. task(agent_type="architect", reasoning_effort="medium", prompt="verify completion") → Verdict: "APPROVED"1715. Run /cancel172```173Why good: Fresh evidence at each step, architect verification, then clean exit.174</Good>175176<Bad>177Claiming completion without verification:178"All the changes look good, the implementation should work correctly. Task complete."179Why bad: Uses "should" and "look good" -- no fresh test/build output, no architect verification.180</Bad>181182<Bad>183Sequential execution of independent tasks:184```185task(agent_type="executor", reasoning_effort="low", prompt="Add type export") → wait →186task(agent_type="executor", reasoning_effort="medium", prompt="Implement caching") → wait →187task(agent_type="executor", reasoning_effort="xhigh", prompt="Refactor auth")188```189Why bad: These are independent tasks that should run in parallel, not sequentially.190</Bad>191</Examples>192193<Escalation_And_Stop_Conditions>194- Stop and report when a fundamental blocker requires user input (missing credentials, unclear requirements, external service down)195- Stop when the user says "stop", "cancel", or "abort" -- run `/cancel`196- Continue working when the hook system sends "The boulder never stops" -- this means the iteration continues197- If architect rejects verification, fix the issues and re-verify (do not stop)198- If the same issue recurs across 3+ iterations, report it as a potential fundamental problem199</Escalation_And_Stop_Conditions>200201<Final_Checklist>202- [ ] All requirements from the original task are met (no scope reduction)203- [ ] Zero pending or in_progress TODO items204- [ ] Fresh test run output shows all tests pass205- [ ] Fresh build output shows success206- [ ] lsp_diagnostics shows 0 errors on affected files207- [ ] Architect verification passed: on a routing-capable surface via explicit `task(agent_type="architect", reasoning_effort="medium"...)` minimum. On the exact reviewed Codex releases 0.144.5, 0.145.0, 0.146.1, and 0.148.0-alpha.5 when role routing is unavailable, no adapted Ralplan Architect pass is valid; every other version remains unknown and fails closed. When adapted authority is attempted, preflight must have stopped with the leader-proof diagnostic. Ordinary Ralph work remains subject to its existing controls.208- [ ] Codex goal-mode completion audit passed, and `update_goal({status: "complete"})` was called when an active goal exists209- [ ] ai-slop-cleaner pass completed on changed files (or --no-deslop specified)210- [ ] Post-deslop regression tests pass211- [ ] `/cancel` run for clean state cleanup212</Final_Checklist>213214<Advanced>215## PRD Mode (Optional)216217When the user provides the `--prd` flag, initialize a Product Requirements Document before starting the ralph loop.218219### Detecting PRD Mode220Check if `{{PROMPT}}` contains `--prd` or `--PRD`.221222Prompt-side `$ralph` workflow activation is lighter-weight than `omx ralph --prd ...`.223It seeds Ralph workflow state and guidance, but it does not implicitly launch the224CLI entrypoint or apply the PRD startup gate. Treat `omx ralph --prd ...` as the225explicit PRD-gated path.226227### Detecting `--no-deslop`228Check if `{{PROMPT}}` contains `--no-deslop`.229If `--no-deslop` is present, skip the deslop pass entirely after Step 7 and continue using the latest successful pre-deslop verification evidence.230231### Visual Reference Flags (Optional)232Ralph execution supports visual reference flags for screenshot tasks:233- Repeatable image inputs: `-i <image-path>` (can be used multiple times)234- Image directory input: `--images-dir <directory>`235236Example:237`ralph -i refs/hn.png -i refs/hn-item.png --images-dir ./screenshots "match HackerNews layout"`238239### PRD Workflow2401. Run deep-interview in quick mode before creating PRD artifacts:241 - Execute: `$deep-interview --quick <task>`242 - Complete a compact requirements pass (context, goals, scope, constraints, validation)243 - Persist interview output to `.omx/interviews/{slug}-{timestamp}.md`2442. Create canonical PRD/progress artifacts:245 - PRD: `.omx/plans/prd-{slug}.md`246 - Progress ledger: `.omx/state/{scope}/ralph-progress.json` (session scope when available, else root scope)2473. Parse the task (everything after `--prd` flag)2484. Break down into user stories:249250```json251{252 "project": "[Project Name]",253 "branchName": "ralph/[feature-name]",254 "description": "[Feature description]",255 "userStories": [256 {257 "id": "US-001",258 "title": "[Short title]",259 "description": "As a [user], I want to [action] so that [benefit].",260 "acceptanceCriteria": ["Criterion 1", "Typecheck passes"],261 "priority": 1,262 "passes": false263 }264 ]265}266```2672685. Initialize canonical progress ledger at `.omx/state/{scope}/ralph-progress.json`2696. Guidelines: right-sized stories (one session each), verifiable criteria, independent stories, priority order (foundational work first)2707. Proceed to normal ralph loop using user stories as the task list271272### Example273User input: `--prd build a todo app with React and TypeScript`274Workflow: Detect flag, extract task, create `.omx/plans/prd-{slug}.md`, create `.omx/state/{scope}/ralph-progress.json`, begin ralph loop.275276### Legacy compatibility277- During the compatibility window, Ralph `--prd` startup still validates machine-readable story state from `.omx/prd.json`.278- `.omx/plans/prd-{slug}.md` remains the canonical storage/documentation artifact, but it is not yet the startup validation source.279- If `.omx/prd.json` exists and canonical PRD is absent, migrate one-way into `.omx/plans/prd-{slug}.md`.280- If `.omx/progress.txt` exists and canonical progress ledger is absent, import one-way into `.omx/state/{scope}/ralph-progress.json`.281- Keep legacy files unchanged for one release cycle.282283## Background Execution Rules284285**Run in background** (`run_in_background: true`):286- Package installation (npm install, pip install, cargo build)287- Build processes (make, project build commands)288- Test suites289- Docker operations (docker build, docker pull)290291**Run blocking** (foreground):292- Quick status checks (git status, ls, pwd)293- File reads and edits294- Simple commands295</Advanced>296297Original task:298{{PROMPT}}
lovits