[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-79669":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":9,"languages":9,"totalLinesOfCode":9,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":11,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":14,"starSnapshotCount":14,"syncStatus":25,"lastSyncTime":26,"discoverSource":27},79669,"directional-prompting","kingbootoshi\u002Fdirectional-prompting","kingbootoshi","Outcome-first plus directional language. A two-layer skill for writing prompts, agent directives, and skill descriptions. Works in Claude Code and Codex CLI.",null,111,9,101,8,0,3,1,"MIT License",false,"main",true,[],"2026-06-12 02:03:54","# Directional Prompting\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"plugins\u002Fdirectional-prompting\u002Fskills\u002Fdirectional-prompting\u002Fassets\u002Fhero.jpeg\" alt=\"Vibe Bot walking a glowing path that Bootoshi is pointing toward, with the words 'Directional Prompting — name the path. the trees disappear.'\" width=\"900\">\n\u003C\u002Fp>\n\nA two-layer skill for writing prompts, agent directives, skill descriptions, slash commands, and anywhere else an LLM reads instructions. Works the same in **Claude Code** and **OpenAI Codex CLI**.\n\nSame `SKILL.md`, same trigger surface, same outcome. Drop it into `~\u002F.claude\u002Fskills\u002Fdirectional-prompting\u002F` and `~\u002F.codex\u002Fskills\u002Fdirectional-prompting\u002F` and both agents pick it up natively.\n\n## The two layers\n\n**Layer 1 — Outcome.** Every non-trivial prompt opens with a block that names the destination: the goal, what \"done\" looks like, when to stop, the true invariants. This is the frame.\n\n**Layer 2 — Direction.** Inside that frame, every sentence names the path forward with positive verbs. \"Trace\", \"build\", \"use\", \"read\", \"return\", \"ask\", \"check\". The correct behavior is described so clearly and completely that the wrong behavior has no room to exist.\n\nOutcome without direction reads as wishful — the model knows where to go but not how to step. Direction without outcome wanders — the model walks crisp paths to nowhere. Both layers together: a model that knows the destination and walks toward it on every token.\n\n## Why both labs converge here\n\nModern frontier models follow instructions literally. The Claude 4.7 guide: *\"Positive examples showing how Claude can communicate with the appropriate level of concision tend to be more effective than negative examples or instructions that tell the model what not to do.\"* The GPT-5.5 guide: *\"GPT-5.5 is strongest when the prompt defines the target outcome, success criteria, constraints, and available context, then lets the model choose the path.\"*\n\nBoth labs converge on the same shape. Name the destination. Name the path. Skip the prohibitions.\n\n## What this skill does\n\nWhen the agent sees prompt-authoring language (`write a prompt`, `improve this prompt`, `audit this system prompt`, `outcome-first`, `make this prompt positive`, or any new skill, agent, or directive being authored), it:\n\n1. Opens the draft with an explicit outcome block: `Goal:`, `Success means:`, `Stop when:` (plus `Constraints:` when needed).\n2. Rewrites the body so every sentence leads with a positive verb naming the correct action.\n3. Audits the draft for `don't`, `do not`, `never`, `avoid`, `refrain`, `instead of`, `rather than`, `not allowed`, `prohibited`, `forbidden`, `won't`, `shouldn't` — and rewrites each as the positive replacement.\n4. Demotes decorative `ALWAYS` \u002F `NEVER` \u002F `MUST` down to plain prose so the absolute-rule signal stays loud where it matters.\n5. Keeps negation only in four narrow cases: hard safety boundaries, disambiguating near-identical paths, acceptable spaces too large to enumerate, and specific banned items where the positive form is genuinely ambiguous.\n\n## The before-and-after that proves it\n\n**Before** — no outcome block, mostly negatives, seven `don't`s:\n\n```text\nYou are a code reviewer. Don't be too harsh. Don't nitpick formatting.\nAvoid making assumptions about the author's intent. Never approve code\nwith obvious bugs. Don't suggest changes that aren't actionable. Try\nnot to be vague. Avoid emojis.\n```\n\n**After** — outcome on top, directional inside:\n\n```text\nGoal: Review the PR diff and decide whether to approve, request changes, or block.\n\nSuccess means:\n  - Verdict is one of: APPROVE, REQUEST_CHANGES, BLOCK\n  - Each comment names the file, line, and replacement code\n  - Comments cover correctness, security, clarity (skip formatting — the linter handles that)\n\nStop when: A verdict is issued and every comment is actionable.\n\nFocus on bugs you can reproduce, security boundaries, and unclear logic.\nAsk before interpreting intent — quote the line and request clarification.\nBlock merges on reproducible bugs. Write in plain text.\n```\n\nSame constraints, half the length. The model knows the destination (verdict plus actionable comments), how to stop (verdict issued), and every sentence in the body pulls forward.\n\n## Install\n\n### Claude Code\n\nAdd the marketplace, then install the plugin:\n\n```text\n\u002Fplugin marketplace add kingbootoshi\u002Fdirectional-prompting\n\u002Fplugin install directional-prompting\n```\n\nRestart Claude Code if the skill does not appear immediately. After install, the skill triggers automatically on prompt-authoring language, or you can invoke it explicitly.\n\n### OpenAI Codex CLI\n\nCodex looks for skills at `~\u002F.codex\u002Fskills\u002F\u003Cname>\u002F`. Drop the skill directory there:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fkingbootoshi\u002Fdirectional-prompting.git ~\u002F.directional-prompting\nmkdir -p ~\u002F.codex\u002Fskills\nln -sfn ~\u002F.directional-prompting\u002Fplugins\u002Fdirectional-prompting\u002Fskills\u002Fdirectional-prompting ~\u002F.codex\u002Fskills\u002Fdirectional-prompting\n```\n\nRestart Codex if needed. Invoke explicitly by asking Codex to \"use directional prompting on this\" or it will auto-select on prompt-authoring language.\n\n### One canonical location for both agents\n\nKeep the skill in one place and symlink it into both agents:\n\n```bash\n# clone once\ngit clone https:\u002F\u002Fgithub.com\u002Fkingbootoshi\u002Fdirectional-prompting.git ~\u002F.directional-prompting\n\n# Claude Code\nmkdir -p ~\u002F.claude\u002Fskills\nln -sfn ~\u002F.directional-prompting\u002Fplugins\u002Fdirectional-prompting\u002Fskills\u002Fdirectional-prompting ~\u002F.claude\u002Fskills\u002Fdirectional-prompting\n\n# Codex CLI\nmkdir -p ~\u002F.codex\u002Fskills\nln -sfn ~\u002F.directional-prompting\u002Fplugins\u002Fdirectional-prompting\u002Fskills\u002Fdirectional-prompting ~\u002F.codex\u002Fskills\u002Fdirectional-prompting\n```\n\nBoth agents follow symlinks for skill discovery, so updates to the cloned repo flow to both automatically.\n\n## When to fire it\n\nRun this skill whenever you write or audit any of the following:\n\n- System prompts for agents\n- `AGENTS.md`, `CLAUDE.md`, project instructions\n- Skill descriptions and `SKILL.md` bodies\n- Tool descriptions in JSONSchema\n- Slash command bodies\n- Cursor rules, Continue rules, any IDE-agent ruleset\n- Sub-agent prompts in orchestration code\n- Eval rubric instructions\n\nFor each draft, run the four-check pass inside the skill:\n\n1. **Outcome check.** Does the prompt open with goal plus success criteria plus stopping condition? If not, add the block.\n2. **Direction check.** Count negations in the body. Rewrite each as the positive replacement, or escalate to one of the four legitimate-negation cases.\n3. **Absolute-rule check.** Is every `ALWAYS` \u002F `NEVER` \u002F `MUST` a true invariant? Demote the decorative ones to plain prose.\n4. **Read-back.** Read the final prompt aloud. Every sentence should name a destination or a step toward it. Cut anything that does neither.\n\n## Why this matters more for agents\n\nA coding agent reads its system prompt on every turn. A negation that plants the wrong concept gets re-planted dozens of times per session. A vague outcome lets the agent's notion of \"done\" drift turn-by-turn.\n\nOutcome plus direction together re-load the correct frame on every turn — the agent's attention is structurally aimed at the destination, and every instruction in the body points toward it.\n\n## Layout\n\n```text\ndirectional-prompting\u002F\n  .claude-plugin\u002F\n    marketplace.json\n  plugins\u002F\n    directional-prompting\u002F\n      skills\u002F\n        directional-prompting\u002F\n          SKILL.md              # the skill itself\n          agents\u002F\n            openai.yaml         # Codex-specific UI metadata\n          assets\u002F\n            hero.jpeg\n```\n\n## Compatibility\n\nBuilt against the open [Agent Skills](https:\u002F\u002Fdevelopers.openai.com\u002Fcodex\u002Fskills) standard. Works in any agent that scans a skills directory for `SKILL.md`:\n\n| Agent | Personal path | Project path |\n| --- | --- | --- |\n| Claude Code | `~\u002F.claude\u002Fskills\u002Fdirectional-prompting\u002F` | `.claude\u002Fskills\u002Fdirectional-prompting\u002F` |\n| OpenAI Codex CLI | `~\u002F.codex\u002Fskills\u002Fdirectional-prompting\u002F` | `.agents\u002Fskills\u002Fdirectional-prompting\u002F` or `.codex\u002Fskills\u002Fdirectional-prompting\u002F` |\n| Cursor | n\u002Fa | `.cursor\u002Fskills\u002Fdirectional-prompting\u002F` |\n| Google Antigravity | `~\u002F.gemini\u002Fantigravity\u002Fskills\u002Fdirectional-prompting\u002F` | `.agent\u002Fskills\u002Fdirectional-prompting\u002F` |\n\nThe `agents\u002Fopenai.yaml` sidecar is Codex-specific UI metadata. Other agents ignore it.\n\n## License\n\nMIT.\n","Directional Prompting 是一个用于编写提示、代理指令和技能描述的两层技能工具，适用于 Claude Code 和 OpenAI Codex CLI。其核心功能包括明确目标和路径两个层面：首先定义任务的目标和成功标准，然后用积极动词详细描述实现步骤，确保模型能够清晰地理解和执行指令。该工具还能自动识别并优化负面表述，将其转换为正面指导，从而提高生成内容的质量和准确性。适用于需要精准控制大语言模型输出的各种场景，如编写复杂的系统提示、设计聊天机器人对话流程等。",2,"2026-06-11 03:58:12","CREATED_QUERY"]