菜单
Compresses prompts/skills into minimal goal-focused instructions. Trusts the model, drops what it already knows, maximizes action space. Use when asked to compress, condense, or minimize a prompt.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/doodledood/manifest-dev --skill compress-prompt命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Manifest builder. Turns shared understanding into a verifiable Manifest with Deliverables, Acceptance Criteria, Global Invariants, and an Initial Approach. Use when planning features, scoping refactors, debugging complex issues, or when the user asks to define, scope, plan, spec out, make a manifest, or break down a task.
Manifest executor. Works through Deliverables verifying every Acceptance Criterion and Global Invariant. Use when executing a manifest, running a plan, implementing a defined task, or when the user asks to run, execute, implement, or ship a manifest-backed plan.
Review a change along one specific quality dimension — bugs, design, simplicity, maintainability, testability, test quality, type safety, contracts, operational readiness, docs, prose value, change intent, or CLAUDE.md adherence. Loads exactly one dimension reference and audits the diff against it, returning a structured PASS/FAIL report with severities. Use when verifying a change before merge, auditing a diff for a named concern, or running a manifest acceptance gate.
正在显示 SKILL.md
基于 SOC 职业分类
| name | compress-prompt |
| description | Compresses prompts/skills into minimal goal-focused instructions. Trusts the model, drops what it already knows, maximizes action space. Use when asked to compress, condense, or minimize a prompt. |
| metadata | {"internal":true} |
Transform a prompt into the minimal instruction needed for the model to succeed. Not "preserve everything densely"—instead, "what's the least I need to say?"
Output: Display compressed result + stats. Optionally write to file with --output <path>.
$ARGUMENTS = prompt (file path or inline text) [--output path]
If file path: read content. If inline: use directly. If ambiguous: try as file first.
Trust capability, enforce discipline - Models know HOW to do tasks. But they cut corners, forget context, skip verification, declare victory early. Drop capability instructions, keep discipline guardrails.
Goal over process - State WHAT to achieve, not HOW. Let the model choose its approach.
Training filter - "Would a competent person need to be told this?" If no → drop it. Models are trained on millions of examples.
Maximize action space - Fewer constraints = more freedom = better results. Each constraint should earn its place.
Inline-typable brevity - Short enough you could type it verbally to a capable colleague.
Avoid arbitrary values - "Max 4 rounds" or "2-3 examples" become rigid rules. State the principle, not the number. Constrain productively while giving flexibility.
| KEEP | DROP |
|---|---|
| Core goal/purpose | Process/phases (capability) |
| Acceptance criteria (success conditions) | Examples the model knows |
| Novel constraints (counter-intuitive rules) | Obvious constraints (model defaults) |
| Execution discipline (write before proceeding, verify before finalizing) | Edge case handling (model trained on these) |
| Output format if non-standard | Explanations and rationale |
Execution discipline examples (KEEP these):
Training-redundant examples (DROP these):
Create todo list - Track: input validation, compression, verification iterations, output.
Verify with agent - Launch prompt-compression-verifier to check goal clarity, novel constraints preserved, no over-specification. Iterate until verification passes.
Single paragraph output - The compressed prompt must be one dense paragraph, not reformatted sections or bullets.
Non-destructive - Original file untouched. Display output + optional file save.
Compressed: {source}
Original: {tokens} tokens
Compressed: {tokens} tokens ({percentage}% reduction)
---
{compressed paragraph}
---
Verification: PASSED/INCOMPLETE ({iterations} iteration(s))
Before (1,247 tokens): Full code reviewer prompt with phases, edge cases, examples...
After (67 tokens):
Review code for bugs, security issues, performance problems; success = all critical issues identified with actionable fixes. Output JSON {file, line, issue, severity, fix}. Never approve code with critical issues.
Kept: Goal, acceptance criteria, output format, novel constraint (never approve with critical issues). Dropped: Process phases, edge case handling, examples, obvious constraints.