AGENT SKILL

widen-the-solution-space

Deliberately widen the set of candidate approaches before committing to one. Use before irreversible decisions (architecture, DB schema, data format, public API), when the first solution feels complicated or awkward, when the user is deciding rather than instructing ("what's the best way to do this?"), or when a previous attempt at the same problem failed. Skip for routine work with one obvious answer.

License 许可
MIT
最近更新
2026-11-07
Stars
★ 16
Forks
2

01 前言 Front Matter

name
widen-the-solution-space
description
Deliberately widen the set of candidate approaches before committing to one. Use before irreversible decisions (architecture, DB schema, data format, public API), when the first solution feels complicated or awkward, when the user is deciding rather than instructing ("what's the best way to do this?"), or when a previous attempt at the same problem failed. Skip for routine work with one obvious answer.

03 SKILL.md 正文

Diverge Then Converge

The most common failure: the first plausible idea that comes to mind immediately becomes the answer. Because it isn't bad, there's no felt need to look for anything better. The more fluently you can generate code, the deeper this trap.

Diverge: 12 Stimulus Axes

Run through these quickly, in your head. One line each is enough — don't force every one to be filled in.

  1. The obvious solution — whatever comes to mind first. Keep it as a baseline.
  2. Industry standard — how do people who already solved this problem usually do it?
  3. The simplest possible thing — embarrassingly simple. Surprisingly often the right answer.
  4. Build nothing — can an existing library, feature, or manual process replace this?
  5. Eliminate the problem — can something upstream change so this situation never arises in the first place?
  6. The opposite direction — what if the premise is flipped? (push↔pull, sync↔async, store↔recompute, eager↔lazy, shared↔copied). "Extract shared code" isn't always the right pole either way: Hncs's brand-agnostic statistics/orchestration (summarize(), run_per_brand_calibration()) belongs shared — copy-pasting it across brand files caused a real bug, a hardcoded datasets/leica/... path left in 5 of 6 files' error messages — while the per-brand evaluate_*.py loaders are deliberately kept copied, not shared, so experiments don't couple as they accumulate. Same axis, opposite answer depending on which piece of code it is.
  7. Solve it at a different layer — in the DB instead of the app, in infra instead of code, at build time instead of runtime
  8. A scoped-down version — a solution covering 80%. Is the remaining 20% actually needed?
  9. Borrow from an adjacent domain — does a pattern familiar from a different domain fit here?
  10. A different data structure or model — does representing the data differently make the problem easier?
  11. Buy or borrow — an external service or tool instead of building it yourself
  12. Work backward from the future — from a version of this working well a year from now, what comes first?

Converge

  • Keep only the 2-3 candidates actually worth evaluating. Drop the clearly-nonviable ones silently, without justifying each rejection.
  • Hand the remaining candidates to weigh-tradeoffs.
  • Mention in one line if a rejected candidate was a close call — the user might actually want that direction, or it becomes a lead if the project pivots later.
  • Principal-level angle: for an irreversible or externally-visible choice (schema, API, data format), weigh which candidate is easiest for other teams to adopt or extend — the option that wins on your metrics alone but is hardest for others to build on sets bad precedent.
  • Distinguished/Fellow angle: if one candidate is a multi-year, company-wide bet that could get open-sourced or presented at a conference, that candidate has to survive scrutiny from people outside this org entirely — a different bar than "wins on this quarter's metrics."
  • Executive angle (CTO/VP-Eng): if the winning candidate becomes the default, weigh how many engineers you can realistically hire or retain who already know it — a technically superior but rare skill set is a hiring bottleneck, not a win.

Output

Don't list all 12. Divergence mostly happens in your head — show the user only the narrowed 2-3 and why they were chosen. The idea list itself has no value to the user.

04 同仓库更多 Skills

songjiun10-collab
songjiun10-collab/Hncs

weigh-tradeoffs

Compare real alternatives explicitly and record why one was chosen. Use when picking between libraries, data structures, architectures, or storage options; when a decision is hard to reverse (schema, public API, data format, dependency); or when the user asks "which is better, A or B". Also use to decide how much deliberation a decision even deserves.

★ 16
songjiun10-collab
songjiun10-collab/Hncs

verify-before-claiming

Run the verification and read its output before saying anything is done, fixed, working, or passing. Use before any completion claim, before committing or opening a PR, before moving to the next task, and after a subagent or tool reports success. Triggers on the words "should work", "fixed", "done", "passing", "looks good" — and on the urge to celebrate.

★ 16
songjiun10-collab
songjiun10-collab/Hncs

verifiability-first

Turn the task into something verifiable and declare the success criterion before seeing results. Use when the request is vague about "done" ("make it work", "improve this"), when writing tests, when fixing a bug (write the failing reproduction first), or when comparing two approaches. Critical before any measurement or benchmark, where deciding the bar after seeing the numbers guarantees a "success".

★ 16
songjiun10-collab
songjiun10-collab/Hncs

threat-and-scale-check

Check trust boundaries, scale behavior, and layered safety. Use when handling user input, external API responses, files, or auth; when designing permissions, validation, or any safety mechanism; when data volume could grow; or when writing loops that touch I/O. Also use when reviewing whether a single safeguard is the only thing standing between a mistake and a disaster.

★ 16
songjiun10-collab
songjiun10-collab/Hncs

surgical-change

Keep every changed line traceable to the request. Use when editing existing code, especially code you didn't write; when tempted to clean up something you noticed while reading; when a diff is growing beyond the ask; or when deciding whether to delete code that looks dead.

★ 16
songjiun10-collab
songjiun10-collab/Hncs

simplicity-budget

Keep the solution as small as the problem — YAGNI and a complexity budget. Use when a design starts growing abstractions, config options, or extension points; when adding a dependency; when a file or function keeps getting longer; or when the user asks whether something is over-engineered. Also use before building anything justified by "might need it later".

★ 16