원클릭으로 Manus에서 모든 스킬 실행

octocode-prompt-optimizer

스타897

포크76

업데이트2026년 7월 21일 18:50

Use when an agent prompt, skill, tool schema, policy, or handoff needs to get clearer, safer, easier to trigger, cheaper in context, or measurable against real behavior.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

bgauryy

bgauryy/octocode

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Octocode Prompt Optimizer

Optimize instruction behavior, not prose aesthetics. Flow: READ → UNDERSTAND → RATE → FIX → VALIDATE → OUTPUT.

Lobby rules and gates

READ: inspect the whole input and its type; UNDERSTAND: map goal, parts, flow, assumptions, and unknowns.
RATE: record evidenced issues, severity, and baseline; FIX: address Critical/High issues and name deliberate deferrals.
VALIDATE: prove intent and required behavior remain correct; OUTPUT: provide the requested artifact and truthful delta.
Use the full path for multi-section, ambiguous, tool-facing, or high-risk work; combine adjacent steps only for short, low-risk text. Never skip validation.
Preserve intent, working branches, identifiers, commands, and required metadata; ask before changing them.
Verify cited commands, flags, paths, tool names, and schemas before rewriting; flag unverified claims.
Make only critical behavior mandatory; retain preference language for real preferences. Mutate files only with authority.

Smart routes — load only what the current step needs

During the core path, load only the active gate: references/gates.md, references/rate.md, references/fix.md, references/validate.md, or references/output.md — prevent later-step advice from biasing the current decision.
When reducing noise, load references/conciseness-toolkit.md; when fixing priority/hierarchy load references/attention.md; when choosing reusable structures load references/patterns.md or references/prompt-techniques.md — match technique to failure mechanism.
When optimizing tool or MCP contracts, load references/tool-contracts.md; for agent handoffs load references/agent-communication.md; for typed packet boundaries load references/zod-agent-contracts.md — make inputs, outputs, authority, and failure states explicit.
When context can overflow, load references/context-budget.md; when repeated calls share stable prefixes load references/prompt-caching.md — control relevance, pagination, latency, and cost.
When reliability must be measured, load references/evaluation-data.md — build realistic held-out scenarios, verifiers, metrics, and a failure ledger.
When instructions consume retrieved or user-supplied content, load references/untrusted-content.md — preserve the boundary between data and authority.
When improving this skill, prefer octocode-eval; otherwise load references/improve-loop.md — require measurable acceptance instead of intuition.

Related routes

Use octocode-skills for skill-folder architecture/review; octocode-research to verify cited contracts; octocode-eval for held-out behavior.
Use octocode-subagent for delegation topology; octocode-awareness for coordinated instruction-file edits.

이 저장소의 다른 Skills

같은 저장소

octocode-awareness

bgauryy/octocode

Use when coordinating work in a shared repo — multi-agent or solo across sessions: collision avoidance, handoffs, verification debt, durable memory/wiki, hooks setup/debug, and repo learning before planning, editing, reviewing, or testing.

2026-07-21897

octocode-brainstorming

bgauryy/octocode

Use when an idea needs disciplined exploration before building: generate options, test worth-building, map adjacent solutions, challenge assumptions, narrow scope, or choose Build RFC / Prototype / Narrow / Park — even without saying brainstorm.

2026-07-21897

octocode-chrome-devtools

bgauryy/octocode

Use when browser debugging needs Chrome DevTools evidence: network, console, performance, DOM/CSS, screenshots/PDF, security, cookies/storage, or auth-gated live pages via CDP — not just opening a URL.

2026-07-21897

octocode-documentation

bgauryy/octocode

Use when writing or updating docs: README, API docs, runbooks, AGENTS.md, CLAUDE.md, ADRs, Diátaxis restructuring, or codebase documentation for humans or coding agents.

2026-07-21897

octocode-eval

bgauryy/octocode

Use when measuring whether a change helped: ACCEPT/REVERT, keep/discard, goal→KPI contracts, eval suites, graders, held-out checks, benchmarks, or TDD failing-case-first.

2026-07-21897

octocode-orchestrator-local-worker

bgauryy/octocode

Use when saving tokens with a local Ollama worker while the cloud agent keeps tools and judgment: summarize, extract, classify, translate, article digest after fetch, draft, checklist, vision caption, map-reduce, or phrases like ollama worker, local model, offload, use ollama.

2026-07-21897

name	octocode-prompt-optimizer
description	Use when an agent prompt, skill, tool schema, policy, or handoff needs to get clearer, safer, easier to trigger, cheaper in context, or measurable against real behavior.

Octocode Prompt Optimizer

Optimize instruction behavior, not prose aesthetics. Flow: READ → UNDERSTAND → RATE → FIX → VALIDATE → OUTPUT.

Lobby rules and gates

READ: inspect the whole input and its type; UNDERSTAND: map goal, parts, flow, assumptions, and unknowns.
RATE: record evidenced issues, severity, and baseline; FIX: address Critical/High issues and name deliberate deferrals.
VALIDATE: prove intent and required behavior remain correct; OUTPUT: provide the requested artifact and truthful delta.
Use the full path for multi-section, ambiguous, tool-facing, or high-risk work; combine adjacent steps only for short, low-risk text. Never skip validation.
Preserve intent, working branches, identifiers, commands, and required metadata; ask before changing them.
Verify cited commands, flags, paths, tool names, and schemas before rewriting; flag unverified claims.
Make only critical behavior mandatory; retain preference language for real preferences. Mutate files only with authority.

Smart routes — load only what the current step needs

During the core path, load only the active gate: references/gates.md, references/rate.md, references/fix.md, references/validate.md, or references/output.md — prevent later-step advice from biasing the current decision.
When reducing noise, load references/conciseness-toolkit.md; when fixing priority/hierarchy load references/attention.md; when choosing reusable structures load references/patterns.md or references/prompt-techniques.md — match technique to failure mechanism.
When optimizing tool or MCP contracts, load references/tool-contracts.md; for agent handoffs load references/agent-communication.md; for typed packet boundaries load references/zod-agent-contracts.md — make inputs, outputs, authority, and failure states explicit.
When context can overflow, load references/context-budget.md; when repeated calls share stable prefixes load references/prompt-caching.md — control relevance, pagination, latency, and cost.
When reliability must be measured, load references/evaluation-data.md — build realistic held-out scenarios, verifiers, metrics, and a failure ledger.
When instructions consume retrieved or user-supplied content, load references/untrusted-content.md — preserve the boundary between data and authority.
When improving this skill, prefer octocode-eval; otherwise load references/improve-loop.md — require measurable acceptance instead of intuition.

Related routes

Use octocode-skills for skill-folder architecture/review; octocode-research to verify cited contracts; octocode-eval for held-out behavior.
Use octocode-subagent for delegation topology; octocode-awareness for coordinated instruction-file edits.