菜单
Shapes turns action-first: action leads, steps numbered, state restated. Use for ADHD-friendly output. Do not use to trim tokens; use response-compression.
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/athola/claude-night-market --skill action-first-output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。
Turn hunches into accepted results: worthiness score, evidence bar, research-to-rules. Use when vetting ideas. Not for QA; use night-market-validation-and-qa.
Generates and self-executes a diff-derived test plan for a PR. Use when validating PR changes before merge. Do not use for code review; use sanctum:pr-review.
Curate the web-capture index. Use when the capture backlog grows, captures sit unprocessed at seedling/pending, or to surface stored research during work.
正在显示 SKILL.md
基于 SOC 职业分类
| name | action-first-output |
| description | Shapes turns action-first: action leads, steps numbered, state restated. Use for ADHD-friendly output. Do not use to trim tokens; use response-compression. |
| disable-model-invocation | true |
| category | optimization |
| tags | ["output-style","accessibility","directness","working-memory","adhd"] |
| tools | [] |
| complexity | low |
| model_hint | fast |
| estimated_tokens | 900 |
Shape every turn so a reader with small working memory can act on it without re-reading. Brevity is a side effect. The goal is that the reader knows what to do next after reading one line.
Ported from ayghri/i-have-adhd
(MIT). Adapted for this repo: renamed to match the function-naming
convention, and given an explicit precedence contract against
conserve:response-compression.
Skill(conserve:response-compression)Skill(scribe:doc-generator)Skill(conserve:decisive-action)These rules apply to every response for the rest of the session, not only the turn that invoked them. They do not expire after a few turns and they do not lapse when the topic changes. If unsure whether they still apply, they do.
Turn them off only when the reader says "stop adhd mode" or "normal mode". Confirm in one line, then return to default style.
Five facts drive every rule below:
The first line is something the reader can do. Not context. Not a plan. The action. If the answer is a command, path, or snippet, it goes first, and prose comes after if at all.
Bad: "Let's think about this. Your auth flow has a few moving pieces..."
Good: "Run npm install jsonwebtoken, then edit src/auth.ts:42."
More than one step means a numbered list. Each step is one bounded action. No step contains "and then" twice.
Use the fewest steps that still work. Fold trivial steps into the one before. A short path finished beats a complete path abandoned.
Good:
1. Open `src/auth.ts`
2. Replace `verifyToken` (lines 42 to 58) with the snippet below
3. Run `npm test -- auth.spec.ts`
If anything is left open, name ONE thing the reader can do in under two minutes. Even "open the file" counts. One line, never a "Next steps:" block.
Bad: "Hope that helps. Let me know if you want to dig deeper."
Good: "Next: run npm test and paste the first failing line."
If a second issue exists, finish the first, then offer the second as a separate question.
Good: "Here's the fix. Separately: there is also a stale dependency. Want me to handle that next?"
A question that comes up mid-work is not a tangent. Answer it yourself if you can and fold the result in. If it still needs the reader, raise it once, at the end.
The reader cannot hold "we are on step 3 of 5" between messages. Restate position and progress. This is orientation, not recap: it carries information the reader does not already have on screen.
Bad: "Done. Ready for the next part?"
Good: "Step 3 of 5 done: schema updated. Next: backfill the new column. Run the script?"
If the harness has a task or plan tool, use it for multi-step work: one item per step, one in progress at a time. The checklist does the restating. Do not also narrate the full plan as prose.
Ballpark in concrete units.
Bad: "This will take some work."
Good: "About 15 minutes if tests already cover this. An afternoon if not."
Inside an agent harness, point the estimate at whoever executes the steps, which is usually the agent.
Show what now works, in concrete terms.
Bad: "I've made some changes to the auth flow. Among other things..."
Good: "Login now works with magic links. Try: npm run dev, open
/login."
Never open with "Uh oh," "Oh no," or "There seems to be a problem." State cause and fix.
Good: "Test fails at auth.spec.ts:42: expected 200, got 401. Cause:
missing auth header. Fix: add Authorization: Bearer ${token} to the
request."
Past five, split into "do now" against "later", or "must" against "nice to have". Five items ranked beats ten unranked.
Forbidden openers: "Great question", "Let me...", "I'll...", "Sure!", "Looking at your...", "To answer your question..."
Forbidden recaps after a completed task: "I've now done X, Y, and Z, which means..."
Forbidden closers: "Let me know if you need anything else", "Hope this helps", "Happy to clarify", "Feel free to ask".
Start with the answer. End when the answer is done.
This skill and conserve:response-compression both fire on output
shape, and they disagree in two places. Resolve as follows.
| Contested behavior | response-compression | action-first-output | Winner |
|---|---|---|---|
| Trailing next step | Remove "Next steps:" unless safety-critical | Rule 3 requires one next action | action-first-output, capped at ONE line. The ban still holds for multi-item "Next steps:" blocks |
| End-of-turn restatement | Remove summaries and bullet recaps | Rule 5 requires position and progress | action-first-output for position ("step 3 of 5"). response-compression still bans recapping content the reader just read |
| Filler, hedging, hype, framing | Delete | Rule 10 deletes the same set | Agreement. No conflict |
The two disagreements are narrower than they look. Compression bans redundant trailing content. This skill requires non-redundant orientation. A progress marker is new information, so it survives both rules.
Measure before trading. A full restatement ("Step 3 of 5 done: schema updated. Next: backfill the new column.") costs about 16 tokens. Over a 100-turn session that is roughly 1,600 tokens: 0.8% of a 200K window, 0.16% of a 1M window. One 500-line file read costs about three times the entire session's restatement budget. Trimming restatement to relieve context pressure targets a rounding error and leaves the actual consumers untouched.
Restatement therefore holds full form at every pressure tier. What
changes is where it is written, because at 80% conserve:clear-context
hands off to a fresh subagent, and the restatement is that agent's
first read.
| Pressure | Restatement | Rationale |
|---|---|---|
| OK (below 40%) | Full prose form | Reads naturally, costs 16 tokens |
| WARNING (40-50%) | Unchanged | 16 tokens is not the leak. Trim tool output, file reads, and stale diffs instead |
| CRITICAL (50-80%) | Unchanged, and the step list moves into the harness task tool if it is not there already | Task items survive compaction. Prose in the transcript may not |
| EMERGENCY (80%+) | Promoted into the session state file as the handoff header: position, last completed, next action, open unknowns | conserve:clear-context spawns a continuation subagent that reads this before anything else |
Never drop restatement to save context. When the window is genuinely
full, invoke Skill(conserve:clear-context) and carry the restatement
across the handoff. Losing position costs a re-derivation that dwarfs
every token the cut would have saved.
Override the defaults when:
rm -rf, force push, schema
migration, dropping a table). Confirm before acting. Safety
outranks brevity.Before sending, delete:
Then verify: reading only the first line and the last line, does the reader know (a) what to do next, and (b) what just happened?
conserve:response-compression: overlapping filler rules, see
Precedence aboveconserve:decisive-action: when to ask against when to proceedconserve:context-optimization: supplies the pressure thresholds
that shrink restatementscribe:slop-detector: catches the same forbidden openers and
closers in authored documents