Design & Motion · leonxlnx/taste-skill

full-output-enforcement

Name: full-output-enforcement
Author: leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Pure behavioral constraint prompt (49-line single file): Establishes 'completeness' as the top priority baseline, directly overriding the model's default bias toward brevity.

skills/output-skill/SKILL.md

Treat every task as production-critical. A partial output is a broken output. Do not optimize for brevity — optimize for completeness.

'A partial output is a broken output' serves as the core value anchor across the entire file; rooted in the same research conclusions as research/laziness in the repository (truncation is a behavioral choice rather than a decoding failure), it uses explicit phrasing to counteract the 'stop pressure built into alignment layers'.

Bans placeholder patterns via 'hard failure' blacklists across three categories: in-code, in-text, and structural shortcuts, providing directly verifiable forbidden literal strings (e.g., // rest of code, 'for brevity', skeleton placeholders instead of full text).

skills/output-skill/SKILL.md

**Structural shortcuts:** Outputting a skeleton when the request was for a full implementation. Showing the first and last section while skipping the middle. Replacing repeated logic with one example and a description. Describing what code should do instead of writing it.

The checklist provides concrete literal strings for verification (code blocks forbid // ..., // TODO, bare ...; text forbids 'Let me know if you want me to continue', 'I'll leave that as an exercise', etc.), enabling the model to self-audit item by item.

Three-phase execution pipeline (Scope -> Build -> Cross-check): Counts and locks deliverable quantities first, generates content, and then cross-checks quantities against the original request, replenishing any missing parts.

skills/output-skill/SKILL.md

3. **Cross-check** — Before output, re-read the original request. Compare your deliverable count against the scope count. If anything is missing, add it before responding.

Uses 'deliverable count verification' to transform 'deliver 5 components when the user asks for 5' from a verbal promise into an actionable checking procedure.

Chunking protocol designed for token limits in long outputs: Writes at full quality until reaching a clean breakpoint (end of function/file/section) before pausing, outputs a fixed PAUSED marker, and resumes precisely from the breakpoint upon 'continue' without repeating previous content.

skills/output-skill/SKILL.md

[PAUSED — X of Y complete. Send "continue" to resume from: next section name]

Chunking does not mean compressing or jumping to conclusions (explicitly forbidden in text); rather, it means 'clean breakpoint + resumable state', relying on the host chat context to retain that state.

Quick Check wrap-up: Pre-delivery review for forbidden strings, presence of all requirements, verifiable executable code blocks, and absence of omissions made to save space.

skills/output-skill/SKILL.md

Code blocks contain actual runnable code, not descriptions of what code would do

All self-check items focus directly on 'output format and integrity', aligning one-to-one with the skill's objective (countering truncation/placeholders).

2Core Capabilities

01Enforced complete output: Deliverable counting + item-by-item generation + requirement cross-checking

02Hard bans on three placeholder categories (in-code / in-text / structural shortcuts) with verifiable forbidden strings

03Token-limit chunking and resumption protocol: Clean breakpoints + [PAUSED — X of Y] + precise resumption on continue

04Pre-delivery quick self-check (forbidden strings / completeness / runnability / unabridged content)

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Can be auto-loaded

For Administrators

Evidence & Triggers (File Paths & Excerpts)

SKILL.md: Pure prompt instructions and rule text, with no script execution or external impact

Audit Notes: (Evidence: Directory contains only a single SKILL.md file, confirmed via git ls-files; token scan shows zero hits for URL/curl/wget/fetch/bash/exec/API key/.env/fs/writeFile). Skill-level zero scripts, zero external network calls, zero credential access, zero local read/write; purely an output behavioral constraint. The only consideration is a lack of arbitration when its mandatory completeness phrasing conflicts with an explicit user request for brevity, which is a conversational-level rather than a code-level risk.

5Second-Pass Verification

[ok] 目录资产面（含隐藏/未跟踪文件） — git ls-files + ls -A 双查：该 skill 目录仅 1 个文件 SKILL.md，无 scripts/references/模板/数据/隐藏文件，无可执行代码。
[ok] 执行类 token 扫描 — 逐文件扫描 bash/exec/child_process/curl/wget/fetch/axios/process.env/API_KEY/.env/fs./writeFile/.github 等零命中（brutalist 一处 'sans-serifs.' 单词结尾 fs. 为误报，人工复核排除）。
[ok] 引文逐条 grep -F 字节复核 — 本 JSON 所用全部 quote 均以 grep -F 在 pin commit 工作树中逐条验证存在，无 paraphrase 冒充。
[ok] 元数据与 pin 核对 — 本地 git HEAD == 任务 pin ccbc15639c97057cbfcf32ecebc38ef716e4bb37（README 显示该提交为 docs/sponsor 修订 #102，2026-08-24）；GitHub API 实查 stars=85698、MIT、pushed_at=2026-08-24T15:23:56Z。frontmatter name 与任务给定名一致。
[ok] 功能声明 vs 实际能力夸大检查 — description 声称均由 SKILL.md 正文纯 prompt 机制兑现，无脚本可调用、无越界承诺（详见本文件 how_it_works 引文）。
[ok] 『handles token-limit splits cleanly』是否有真实机制 — 正文给出完整协议：全质量写到干净断点→输出 [PAUSED — X of Y complete. Send "continue"…]→continue 后『pick up exactly where you stopped. No recap, no repetition.』，非空头承诺。
[ok] 禁用清单与『用户要求简短』的冲突仲裁 — 正文无冲突仲裁条款（复查无 unless/user explicitly 等豁免表述）；已在 risks 中明示此缺口，不夸大其覆盖范围。
[ok] 占位串检查是否覆盖常见变体 — bare `...`、/* ... */、// TODO、'Let me know if you want me to continue' 等均以字面列入；属提示强度问题而非缺陷，正文以『hard failures. Never produce them.』设最高语气。

6Verdict & Scenarios

Elevates 'do not truncate' into an executable protocol: Deliverable counts, forbidden literal lists, and continuation markers are all self-auditable and verifiable.

The token chunking protocol is a rare pragmatic design among similar constraints: Provides a clean breakpoint and recovery path when truncation is inevitable, without compromising quality.

Zero dependencies, 49 lines, universally applicable across domains, making it the skill with the lowest migration cost in the family.

Backed by research: Addresses the root cause that 'laziness is a behavioral choice', rather than relying on empty admonitions.

Best for: Suitable for all generative tasks where 'delivery must be complete': long file/multi-component/full-site code generation, migration refactoring, and comprehensive documentation/list generation; especially suitable as a global or task-level behavioral constraint in workflows where the host model is known to truncate or use placeholders.

Not for: Not suitable for scenarios requiring intentionally concise output (summaries, rapid prototyping, conversational Q&A) — it directly conflicts with brevity goals; not suitable for users expecting the skill to enforce validation via scripts (pure prompt without an execution layer); does not resolve design quality issues, only guaranteeing 'completeness'.

Install Agent-Ready

① Verified Mirror 2026-09-09

Option A · Direct DownloadDownload full-output-enforcement.tar.gz

sha256: 499ed544fe930be6…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 3cca18b368.