菜单
Externalize uncertain time with calibrated best, realistic, and worst ranges, hidden costs, checkpoints, and stopping rules. Use for estimates, open-ended research or refinement, work likely to expand, and tasks where elapsed-time feedback should revise the forecast.
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。
基于 SOC 职业分类
Manually route the current request through the installed neurodivergent-support skill pack. Use only when the user explicitly invokes `/nd` in Claude Code, `$nd` in Codex, selects ND Router from the skills UI, asks to activate ND support for the current task, or requests ND status/off controls. Select the smallest relevant subset of sibling `nd-*` skills and continue useful work instead of merely describing them.
When presenting options or a backlog, show only the top 3. Hidden items are noted but not shown until one choice is completed.
Before starting a new task thread, surface any tasks that are more than halfway complete. Ask whether to close one first. One-tap dismiss. Never blocks.
Detect complex or multi-step requests and suggest structured planning before acting. Prevents diving into implementation before the problem is understood.
Stabilize task execution when the user explicitly reports frustration, shame, panic, anger, rejection, or overwhelm that is affecting decisions. Preserve work, separate observable facts from interpretations, lower immediate decision load, and choose one reversible next action without turning the exchange into therapy.
Redesign the task environment when a known intention repeatedly fails at the point of action. Use when reminders arrive too early or late, tools or state are hard to find, choices or steps create friction, a system depends on memory, or repeated failure shows that effort alone is not enough.
| name | nd-time-blindness |
| description | Externalize uncertain time with calibrated best, realistic, and worst ranges, hidden costs, checkpoints, and stopping rules. Use for estimates, open-ended research or refinement, work likely to expand, and tasks where elapsed-time feedback should revise the forecast. |
Time blindness: the reduced ability to intuitively sense how long things take. This is one of the most functionally impactful ADHD symptoms. In development work, it manifests as chronic underestimation ("this should take an hour" → 8 hours), missed deadlines, and the constant surprise that things took longer than expected.
An agent that gives honest, calibrated estimates with ND-aware adjustments significantly reduces the gap between planned and actual time.
Apply when:
Never give only the "ideal" estimate. Always give a range with explicit hidden costs:
Time estimate:
- Best case (no surprises, you stay focused): 1 hour
- Realistic (typical session with distractions + one unexpected thing): 2-3 hours
- Discovery overhead (if [unknown] turns out to be more complex): 4-6 hours
Hidden costs I'm counting:
- [Reading/understanding existing code]: ~30 min
- [Testing and edge cases]: ~45 min
- [Debugging the one thing that won't work as expected]: ~30 min
Apply the 2.5× rule for ND-adjusted estimates. For any task involving:
Multiply the "best case" by 2.5 to get the realistic estimate.
Surface the planning fallacy by name when estimates seem optimistic:
"Quick note: 'this looks like 30 minutes' estimates almost always run longer when they involve [unfamiliar code / debugging / multiple files]. The realistic range is probably [N-M hours]."
Never round down. If you're uncertain, round up.
Mark effort on the task, not just time:
Effort level predicts the chance of time overrun more reliably than initial time estimates.
Name the next checkpoint and its evidence. For open-ended research, retries, or refinement, include a stop condition.
Revise from elapsed evidence. Compare forecast with observed time at the checkpoint and update the remaining range. Treat a confident estimate without task data as a hypothesis, not a promise.
Time blindness in ADHD is not a misconception about time — it's a failure of the prospective working memory system that normally allows people to sense the passage of time and anticipate how long tasks will take. The brain runs without a reliable internal clock.
The planning fallacy (Kahneman & Tversky, 1979) affects everyone — ADHD amplifies it. Reference class forecasting (comparing to similar past tasks) is the most reliable correction, which is why explicit range estimates with labeled hidden costs outperform single-point estimates.
The 2.5× rule is a rough empirical correction based on software project estimation research (Hofstadter's Law and variations). It's imprecise but systematically closer to reality than naive estimates for work with any unknowns.