메뉴
Use when a behavior-locked changed diff needs post-implementation, pre-review cleanup for reuse, simplicity, efficiency, and appropriate abstraction; not for initial implementation or correctness/root-cause work.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when the user explicitly delegates autonomous end-to-end delivery across discovery, planning, implementation, and verification; not for a small execution-ready task.
Use when the user explicitly delegates autonomous end-to-end delivery across discovery, planning, implementation, and verification; not for a small execution-ready task.
Use when requirements are sufficiently known but implementation is broad, risky, architecture-sensitive, cross-file, multi-step, or strategy-unclear and needs a plan; not for still-vague discovery or an execution-ready task.
Use when the user explicitly requests TDD, test-first, or RED-GREEN-REFACTOR, or an already-selected workflow enters its internal TDD gate; ordinary implementation remains Ralph.
Use when requirements are sufficiently known but implementation is broad, risky, architecture-sensitive, cross-file, multi-step, or strategy-unclear and needs a plan; not for still-vague discovery or an execution-ready task.
Use when the user explicitly requests TDD, test-first, or RED-GREEN-REFACTOR, or an already-selected workflow enters its internal TDD gate; ordinary implementation remains Ralph.
| name | simplify |
| description | Use when a behavior-locked changed diff needs post-implementation, pre-review cleanup for reuse, simplicity, efficiency, and appropriate abstraction; not for initial implementation or correctness/root-cause work. |
| argument-hint | [<target>] |
This generated file is the Claude Code-facing runtime skill document. Claude Code slash commands should read this file directly; maintainers edit the source documents listed below instead.
Source order:
../../docs/skill-core/simplify.md../../docs/platforms/claude-code-runtime.md../../docs/platforms/claude-code-simplify.mdThe sections below are already composed for this platform. Do not ask the runtime model to load another platform's runtime document or invocation syntax.
Simplify reviews changed code for reuse, simplification, efficiency, and
altitude issues, then applies behavior-preserving fixes. Its main skill pass
is read-only discovery; accepted repository work-product cleanup is applied
through executor, with inline mutation only as a recorded LIGHT-tiny or
dispatch-unavailable fallback.
It is a skill, not an agent. In Oh No Harness, ralph uses it after the
behavior lock and BEFORE the single review round, so the sole review sees the
cleaned diff; the active platform runtime document supplies the matching
simplify skill. ralph owns a preflight quick diff scan and invokes this skill
only when that scan finds actual candidates or candidate uncertainty remains;
a clean quick scan records cleanup as not needed without invoking Simplify.
This is quality cleanup, not bug hunting. Use code-reviewer or the host's
code-review workflow for correctness bugs.
Simplify is the post-implementation, pre-review cleanup stage.
Use it after behavior is locked and before the caller's single review round, so that review sees the cleaned diff. It should improve reuse, clarity, maintainability, and efficiency without changing behavior, adding scope, or replacing implementation review.
Reuse, Simplification, Efficiency, and Altitude are review viewpoints, not four mandatory jobs.
no candidates when the scan finds nothing; do not create cleanup work
to satisfy a pass count.Record Cleanup depth: combined | four-viewpoint, its trigger, and the changed
files inspected. Viewpoint work is read-only discovery. Use discovery
subagents only when separate contexts can change the cleanup decision enough
to justify lifecycle cost. Every direct role dispatch reuses the target role's
required identity/result envelope and adds only Simplify's workflow delta: the
behavior lock and, for apply assignments, accepted cleanup finding IDs. If
cleanup creates a need for additional review or evidence, return it to the
caller; Simplify does not expand its own mandate.
Use for changed code that works but needs a quality pass before delivery:
Do not use when the requested change is functional behavior, a broad refactor, or a rewrite without a behavior lock.
Before editing, establish one of:
For a cleanup that alters control flow, structure, or shared/altitude mechanisms,
an executable lock — passing tests or a focused verification command — is
required; a prose lock (checklist or before/after description) suffices only for
purely local edits. If only a prose lock exists for such a change, record it as
reviewer follow-up under the Maintainability Debt Boundary instead of
applying it.
If no behavior lock exists, create a verification plan first and state the risk.
Simplify may fix behavior-preserving cleanup inside the reviewed diff, but it must not hide correctness, scope, security, or architecture concerns inside a cleanup pass. When a finding points to maintainability debt that cannot be fixed without changing behavior, widening scope, changing public contracts, or touching sensitive behavior, record it as deferred reviewer work instead of patching around it.
Classify every nontrivial finding as:
Maintainability finding:
- Type: behavior-preserving cleanup | reviewer follow-up | out-of-scope
- Evidence:
- Cost if ignored:
- Safe cleanup action:
- Reviewer needed: none | code-reviewer | verifier
Use reviewer follow-up for brittle coupling, unclear ownership, hidden state,
cross-boundary special cases, fragile tests, risky generated changes, or cleanup
that would require a behavior decision.
When called by ralph, limit cleanup to files changed in the current Ralph session.
When called directly, limit cleanup to the target files or paths named by the user.
Do not expand scope because nearby code looks messy.
If a PR number, branch name, or file path was passed as an argument, review that target instead of the default diff.
Run git diff @{upstream}...HEAD to get the unified diff under review. If
there is no upstream, use the smallest credible fallback such as
git diff main...HEAD, git diff master...HEAD, or git diff HEAD~1.
If there are uncommitted changes, or the range diff is empty, also run
git diff HEAD and include working-tree changes in scope. Treat this diff as
the review scope.
This phase is read-only discovery: do not edit repository work product or
.oh-no state while finding and classifying candidates.
Apply the Cleanup Depth Decision. A combined scan checks Reuse, Simplification, Efficiency, and Altitude in one bounded pass. When a named THOROUGH trigger selects four-viewpoint depth, launch the four independent cleanup subagents in one batch before waiting when the host concurrency limit permits it. Otherwise use the smallest bounded waves that preserve four independent contexts; use four labeled inline blocks only with a recorded dispatch-unavailable reason. The caller captures and cleans up every dispatched result.
Each pass returns findings with file, line, a one-line summary, and the
concrete cost: what is duplicated, wasted, fragile, or harder to maintain.
Each pass must also classify whether the finding is behavior-preserving cleanup,
reviewer follow-up, or out-of-scope under the Maintainability Debt Boundary.
Flag new code that re-implements something the codebase already has. Search shared utilities and files adjacent to the change, and name the existing helper or local pattern to use instead.
Flag unnecessary complexity the diff adds: redundant or derivable state, copy-paste with slight variation, deep nesting, dead code left behind, or Speculative abstraction. Name the simpler behavior-preserving form.
Flag wasted work the diff introduces: redundant computation or repeated I/O, independent operations run sequentially, or blocking work added to startup, tests, or hot paths. Name the cheaper behavior-preserving alternative.
Check that each change is implemented at the right depth, not as a fragile bandaid. Special cases layered on shared infrastructure are a sign the fix is not deep enough. Prefer generalizing the underlying mechanism over adding special cases, but skip anything that would change intended behavior or exceed the reviewed scope.
Capture the combined result or all four expanded results, then deduplicate
findings that point at the same line or mechanism. Select only accepted
behavior-preserving cleanup and dispatch one scoped executor assignment (or
up to 5 disjoint assignments when the caller's isolation policy permits). Reuse the
executor's required envelope and add only the behavior lock plus accepted
cleanup finding IDs. Simplify interprets the executor envelope; Mutation status: complete is not cleanup acceptance.
Inline application is allowed only with Mutation fallback: LIGHT-tiny for a
tiny LIGHT cleanup or Mutation fallback: dispatch-unavailable after a failed
dispatch attempt. Record the reason. .oh-no state and finding dispositions
remain caller-owned.
Skip any finding whose fix would change intended behavior, require changes well outside the reviewed diff, or that is a false positive. Note the skip rather than debating it.
Run the behavior lock again after cleanup. If the post-cleanup behavior lock
regresses from the pre-cleanup result, assign a focused executor revert/fix or
report the blocker; do not let a discovery role mutate the work product. If
cleanup changed structure, tests, or control flow, return that evidence need
to the caller. When the caller is ralph, its upcoming sole review inspects
the cleaned result; cleanup does not add a later review round.
Return:
None - this is a post-implementation mid-loop skill. Return control to the caller (ralph or direct invocation). If the cleanup pass changed structure, tests, or control flow, identify the evidence need for the caller; under ralph, the upcoming sole review covers the cleaned diff. Do not invoke another role or skill yourself.
This compact platform section is embedded in generated Claude Code-facing skill documents.
Claude Code-facing public skills live under skills-claude/. Generated
skills-claude/<skill>/SKILL.md files compose the matching skill core, this
compact runtime section, and any Claude Code skill-specific overlay such as
docs/platforms/claude-code-<skill>.md. Slash commands must delegate to the
matching generated skill document.
Use the host's structured question tool when available for approval, preference, scope, or next-step selection; otherwise ask one focused plain-text question and wait. Present options as actions the host agent will take.
When a core skill has a multi-phase approval handoff and the host exposes task tracking, create one task per phase and complete them sequentially.
Keep Claude prompts explicit and sectioned: state scope, non-goals, constraints, approval gates, expected evidence, and output format. Preserve long-running context in artifacts before compaction, task handoff, or subagent dispatch.
Dispatch only after the active skill's trigger fires, then read
docs/platforms/claude-code.md ## Role Dispatch for the full host contract.
Prefer oh-no-harness:<role>, request the whole independent batch before
waiting, capture every final result, and clean up only after integration. An
approved-plan handoff is dispatch authorization for eligible isolated roles;
plugin-agent unavailability uses the documented embedded-role fallback.
This mechanism is trigger-loaded, not embedded in every workflow decision. For
any dispatched plan-reviewer or code-reviewer pair (every dispatched review),
or a named THOROUGH debugger pair, both legs MUST be requested in a single
batch: issue both subagent tool calls in the same assistant turn (or with
Background: yes for both) BEFORE waiting on either result; a serial
dispatch-wait-dispatch sequence is not a valid pair. The two legs' packet bodies
MUST be identical except the single Assigned perspective: line (Lens A on the
primary leg, Lens B on the diversity leg); leg identity (primary vs
diversity) is carried ONLY by the host dispatch metadata (the description field
and the model override), never inside the packet text. Read the role's concrete
stored primary and validated secondary top-tier model from the session
<OH_NO_MODEL_DIVERSITY> block. The primary leg is
unoverridden and uses the declared-frontmatter primary; the
secondary leg carries an explicit NATIVE model override. Claim
model-diversity-pair only when the primary is not host-default and the
secondary differs from it. Otherwise default to two independent same-model
instances as same-model-parallel-fallback with the reason recorded; an
explicit require-model-diversity demand transitions to PAUSED when the
diversity leg is unavailable. Fusion Rescue uses its Claude Code overlay's
three-panel assignment instead of this two-leg shape.
This platform overlay is source content for the generated Claude Code-facing
simplify runtime document, after the shared core and
docs/platforms/claude-code-runtime.md.
When the core selects combined depth, run one combined pass. When it selects
four-viewpoint depth, prefer Workflow Promise.all when available; otherwise
issue all four background Task or Agent requests before waiting or inspecting
results. Each request gets exactly one angle: Reuse, Simplification, Efficiency, or Altitude. These are read-only discovery workers: dispatch each as oh-no-harness:explore — Task and Agent set subagent_type: oh-no-harness:explore, a Workflow agent() call sets agentType: "oh-no-harness:explore". If dispatch is unavailable, use the shared core's inline labeled-block fallback; the core owns selection and fallback semantics.