agentsclimarketplace

Helpful now deriver

Skill hegu-1/claude-cognition-skills/skills/helpful-now-deriver

Claude Code skills for the failure modes of long-horizon AI collaboration — continuity, judgment preservation, trap detection, faithful translation.

Install
npx -y skills add hegu-1/claude-cognition-skills --skill helpful-now-deriver

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Determine the smallest genuinely useful next move based on the user's current structural state. Use when the goal is not to answer maximally, but to intervene minimally and effectively.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, as published. Nobody here has run it

Helpful Now Deriver Skill

Version: 0.1

Purpose

This skill determines what is actually helpful now.

It does not aim to produce the most complete answer. It aims to identify the smallest move that creates the highest structural gain while preserving continuity and reducing distortion.

When to use

Use this skill when:

  • the user is exploring and needs the next real foothold
  • the user is overloaded and needs a usable reduction
  • the user is near a decision and needs the right cut
  • the user is asking broadly, but the real need is more specific
  • the assistant could answer many things, but only one move would truly help
  • structural precision matters more than answer volume

Input expectations

Possible inputs:

  • current user message
  • current phase
  • current thread
  • move_type
  • trap status
  • continuity context
  • decision context
  • external output requirement

Output contract

Return:

  • helpful_now
  • why_now
  • intervention_size
  • expected_gain
  • avoid_for_now

Example:

{
  "helpful_now": "compress the current exploration into one decision surface",
  "why_now": "the user has enough direction but no usable anchor",
  "intervention_size": "small",
  "expected_gain": "clarity without losing strategic depth",
  "avoid_for_now": "full expansion into implementation details"
}

Derivation principles

1. Help must be phase-aligned

What helps in exploration may harm in convergence. What helps in rupture may harm in execution.

2. Minimal intervention, maximal gain

Prefer the smallest move that materially improves structure.

3. Do not over-answer

Sometimes the best help is:

  • one framing line
  • one cut
  • one summary surface
  • one translation
  • one constraint

not a large response.

4. Preserve user agency

The move should support the user's path, not replace it.

5. Avoid fake completeness

Do not solve a broader problem than the one that is alive now.

Typical helpful-now forms

Possible outputs include:

  • clarify the real question
  • separate two threads
  • compress into one decision snapshot
  • translate raw judgment into external language
  • list the unresolved variables
  • reduce scope
  • surface the risk before proceeding
  • preserve current insight before expansion
  • re-anchor to the original objective
  • stop further acceleration

Anti-patterns

Do not:

  • default to give more information
  • confuse usefulness with answer length
  • recommend implementation when framing is unstable
  • recommend abstraction when the user needs grounding
  • recommend action too early just to appear useful

Success criteria

This skill is working when:

  • the next move feels exactly sized
  • the user gets traction, not just content
  • the assistant avoids unnecessary expansion
  • progress becomes more real with less noise

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.