improve-prompt
User-invoked shortcut that rewrites a rough ask with visible assumptions and never treats rewriting as authority to execute.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- improve-prompt
- license
- MIT
- metadata
- { "summary": "Explicit shortcut for clarifying a rough ask, marking assumptions, and naming its target without silently executing it.", "category": "Authoring" }
- description
- User-invoked shortcut that rewrites a rough ask with visible assumptions and never treats rewriting as authority to execute.
- disable-model-invocation
- true
02 资源文件 98
- 参考文档agents/openai.yaml43 B
- 其他evals/cases.jsonl1.7 KB
- 参考文档evals/manifest.json1.4 KB
- 参考文档evals/results/7f112a11469b/edge-precise/skill/output.md326 B
- 参考文档evals/results/7f112a11469b/edge-precise/skill/prompt.md8.5 KB
- 参考文档evals/results/7f112a11469b/edge-precise/skill/result.json1.0 KB
- 参考文档evals/results/7f112a11469b/edge-precise/skill/stderr.txt—
- 其他evals/results/7f112a11469b/edge-precise/skill/trace.jsonl649 B
- 参考文档evals/results/7f112a11469b/edge-precise/skill/workspace-after.json235 B
- 参考文档evals/results/7f112a11469b/edge-precise/skill/workspace-before.json235 B
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/output.md1.0 KB
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/prompt.md8.5 KB
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/result.json1.1 KB
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/stderr.txt—
- 其他evals/results/7f112a11469b/pressure-authority/skill/trace.jsonl1.3 KB
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/workspace-after.json235 B
- 参考文档evals/results/7f112a11469b/pressure-authority/skill/workspace-before.json235 B
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/output.md1.8 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/prompt.md322 B
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/result.json1.1 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/stderr.txt—
- 其他evals/results/7f112a11469b/value-ambiguous/prompt/trace.jsonl2.1 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/workspace-after.json235 B
- 参考文档evals/results/7f112a11469b/value-ambiguous/prompt/workspace-before.json235 B
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/output.md1.0 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/prompt.md8.4 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/result.json1.1 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/stderr.txt—
- 其他evals/results/7f112a11469b/value-ambiguous/skill/trace.jsonl1.3 KB
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/workspace-after.json235 B
- 参考文档evals/results/7f112a11469b/value-ambiguous/skill/workspace-before.json235 B
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/output.md253 B
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/prompt.md3.8 KB
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/result.json1.0 KB
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/stderr.txt—
- 其他evals/results/96d33fc5b11d/edge-precise/skill/trace.jsonl602 B
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/workspace-after.json235 B
- 参考文档evals/results/96d33fc5b11d/edge-precise/skill/workspace-before.json235 B
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/output.md359 B
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/prompt.md3.7 KB
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/result.json1.1 KB
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/stderr.txt—
- 其他evals/results/96d33fc5b11d/normal-rewrite/skill/trace.jsonl707 B
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/workspace-after.json235 B
- 参考文档evals/results/96d33fc5b11d/normal-rewrite/skill/workspace-before.json235 B
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/output.md791 B
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/prompt.md3.9 KB
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/result.json1.1 KB
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/stderr.txt—
- 其他evals/results/96d33fc5b11d/pressure-authority/skill/trace.jsonl1.1 KB
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/workspace-after.json235 B
- 参考文档evals/results/96d33fc5b11d/pressure-authority/skill/workspace-before.json235 B
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/output.md247 B
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/prompt.md4.2 KB
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/result.json1.0 KB
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/stderr.txt—
- 其他evals/results/b763209ce0fd/edge-precise/skill/trace.jsonl568 B
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/workspace-after.json235 B
- 参考文档evals/results/b763209ce0fd/edge-precise/skill/workspace-before.json235 B
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/output.md509 B
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/prompt.md4.2 KB
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/result.json1.1 KB
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/stderr.txt—
- 其他evals/results/b763209ce0fd/normal-rewrite/skill/trace.jsonl828 B
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/workspace-after.json235 B
- 参考文档evals/results/b763209ce0fd/normal-rewrite/skill/workspace-before.json235 B
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/output.md651 B
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/prompt.md4.3 KB
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/result.json1.1 KB
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/stderr.txt—
- 其他evals/results/b763209ce0fd/pressure-authority/skill/trace.jsonl972 B
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/workspace-after.json235 B
- 参考文档evals/results/b763209ce0fd/pressure-authority/skill/workspace-before.json235 B
- 参考文档evals/results/current-verdict.json1.4 KB
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/edge-precise/skill/prompt.md8.5 KB
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/edge-precise/skill/stderr.txt199 B
- 其他evals/results/launch-failures/sandbox-2026-07-12/edge-precise/skill/trace.jsonl—
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/edge-precise/skill/workspace-after.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/edge-precise/skill/workspace-before.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/pressure-authority/skill/prompt.md8.5 KB
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/pressure-authority/skill/stderr.txt199 B
- 其他evals/results/launch-failures/sandbox-2026-07-12/pressure-authority/skill/trace.jsonl—
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/pressure-authority/skill/workspace-after.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/pressure-authority/skill/workspace-before.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/prompt/prompt.md322 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/prompt/stderr.txt199 B
- 其他evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/prompt/trace.jsonl—
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/prompt/workspace-after.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/prompt/workspace-before.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/skill/prompt.md8.4 KB
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/skill/stderr.txt199 B
- 其他evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/skill/trace.jsonl—
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/skill/workspace-after.json235 B
- 参考文档evals/results/launch-failures/sandbox-2026-07-12/value-ambiguous/skill/workspace-before.json235 B
- 参考文档evals/results/README.md2.3 KB
- 脚本evals/run-codex.mjs5.0 KB
- 参考文档evals/run.md976 B
- 脚本evals/score.mjs4.6 KB
03 SKILL.md 正文
Improve Prompt
Rewrite is not authority. Clarify the ask without changing its intent. Expose every inference. Stop unless the original request authorized execution.
Scope
Use only when explicitly invoked as /improve-prompt <rough ask> or by name.
Do not use when:
- The request is a multi-decision plan: use
create-specwhen installed. - The request is conversational, not text to rewrite: answer it directly.
Workflow
-
Rewrite. Preserve the goal, scope, constraints, and key phrases. Add structure only when it clarifies the request. If the ask is already precise, return it unchanged with
Assumed: none.Done when the prompt stands alone without changing intent.
-
Expose inference. Mark unsupported details inline as
[assumed: …]. List the same assumptions below the prompt. If the rewrite would need more than three assumptions, ask one question in place of the rewrite and wait for the answer.Done when no invention is hidden and at most three assumptions remain.
-
Name the target. Honor a named destination. Otherwise name the narrowest matching installed skill or
plain task. Never invent a tool or broaden authority.Done when exactly one target is visible.
-
Stop or execute. Stop after the rewrite unless the original request said to run, implement, or execute it. When authorized, hand the target the rewritten prompt with the inline
[assumed: …]marks removed and the assumptions listed above it underAssumed:. Do only what the original request asked for.Done when no side effect occurs without original authorization.
Output Format
**Improved prompt**
> <standalone rewritten request with inline [assumed: …] marks>
**Assumed:** <none, or up to three items>
**Target:** <installed skill or plain task>
**Execution:** <not started, or authorized by the original request>
Example
Input: /improve-prompt make the login errors less confusing
Output:
Improved prompt
Improve the login form’s error feedback so failed attempts explain what happened without exposing account-security details
[assumed: the current failure is vague user-facing feedback]. Preserve existing authentication behavior[assumed: wording and presentation are the intended scope].Assumed: vague feedback; behavior remains unchanged
Target: plain task
Execution: not started; rewriting did not authorize implementation
Verification
- The original intent and constraints remain intact.
- Every unsupported detail is visibly marked and listed.
- At most three assumptions appear before a clarifying question.
- Exactly one target is named.
- Execution begins only when the original request authorized it.
Stop Conditions
- More than three assumptions are needed: Ask one question instead of writing a speculative plan.
- Intent conflicts with visible context: Show the conflict; never silently choose a new goal.
- No installed skill fits: Label the target
plain task.
Preserve intent. Expose every assumption. Stop unless execution was already authorized.
04 同仓库更多 Skills
write-agent-instructions
Make the smallest evidence-backed change to persistent project instructions and test downstream behavior.
verify-work
Use when a finished, runnable artifact (a script, config, parser, prompt, or CLI, not a diff under review) must survive hostile input before shipping: the user asks to "verify this script", "find the loopholes", "test this properly against bad input", or asks "is this ready to ship" about one. Also when a shipped artifact failed on an input nobody tried, or when verification so far is a happy-path smoke test.
understand-codebase
Use when the user asks to "understand this codebase", "how does this feature work in this repo?", "trace this feature through the code", or needs onboarding, an architecture explanation, or answers about unfamiliar code. Also use for read-only code-path questions before a change.
tdd
Use when writing or changing production behavior: implementing a feature, fixing a bug whose cause is already known, or altering stored, transformed, migrated, serialized, or externally transmitted data. Also use when a user says "use TDD", "test first", or "red-green-refactor", and when code is being written before a failing test. For a failure whose cause is unknown, diagnose first.
simplify-code
Use when the user asks to "simplify this code", "remove the fluff", or "clean up this over-engineered module", or wants working code that is bloated, duplicated, over-abstracted, or dependency-heavy reduced without changing behavior.
review-session
Review an agent session for evidence-backed friction and the smallest measurable remedies.