Claude Skill

eli5

Explain a technical spec or proposed change in plain language: what problem it solves, how the solution works, and every schema change.

LLM Mart · 0 points · 0 views 0 listing impressions 0 install-command copies
Virus-scanned Reviewed automatically before listing.

Full trust report

Download dzhng-skills-skills_engineering_eli5-f2fbf92.zip · 1 KB
Part of dzhng/skills — 32 skills

Install

skills CLI npx skills add https://github.com/dzhng/skills/tree/main/skills/engineering/eli5
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install dzhng-skills@llmmart
Git git clone https://github.com/dzhng/skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole dzhng/skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

ELI5

Turn the referenced spec or change into an accurate mental model for a technical reader who does not live in the code. Plain language, full precision: simplify the telling, never the claims.

The Register

Other skills point here when their output must be readable by someone who didn't live the work. The reader is technical — pseudocode and precise claims land fine — but they have read none of the code and none of the session's messages: no diff, no transcript, no labels the work invented along the way. Writing ELI5 means:

  • Walk one concrete scenario end to end — the triggering event, what happens today, what the change (or the unbuilt alternative) would do — instead of describing properties in the abstract.
  • Define every term of art at first use; never lean on labels the spec, code, or session invented.
  • Use small concrete examples without replacing precise claims with analogies.
  • Reach for pseudocode when explaining control flow, ordering, or timing. Prose describing when-something-fires reads as plausible and hides the gap; five lines of pseudocode make the gate visible and let the reader see the case you missed. Write it at the level of the decision, not the implementation — the conditions and their order, not real function signatures. The tell that you needed it: your prose contains "only when", "before", "unless", or "as soon as" and the reader still cannot say what happens on the second call.
  • The test: the text stands alone, without the diff, the spec, or the transcript. If the reader must ask "explain this part", it failed.

Workflow

  1. Read the complete referenced artifact and inspect current owners when the spec alone cannot establish behavior. When no artifact is named, explain the session's current change (working tree or branch diff). Separate what exists today from what is only proposed.
  2. Explain the problem through its user or operational consequence, including why the current design produces it.
  3. Explain the solution as one simple before/after data flow. Introduce each component by responsibility, not by filename or internal symbol.
  4. Inventory schema and durable-contract changes exhaustively: added, changed, removed, reset, and deliberately unchanged. Include cursor or wire-format cutovers when they affect stored data or readers. Say explicitly when there are no schema changes.

Output

Use these headings in order:

  1. Problem
  2. Solution
  3. Schema changes

End with one short sentence stating what users should notice after the change.

Rules

  • Lead with behavior and boundaries; mention implementation names only when they clarify ownership or a contract.
  • Distinguish canonical records from derived indexes, caches, summaries, and presentation grouping.
  • Call out destructive resets, migration requirements, compatibility behavior, eventual consistency, and intentional data loss directly.
  • Do not omit a schema change because it is operational rather than user-visible.
Files (skills)
  • SKILL.md 3.2 KB
    ---
    name: eli5
    description: "Explain a technical spec or proposed change in plain language: what problem it solves, how the solution works, and every schema change."
    disable-model-invocation: true
    ---
    
    # ELI5
    
    Turn the referenced spec or change into an accurate mental model for a technical
    reader who does not live in the code. Plain language, full precision: simplify
    the telling, never the claims.
    
    ## The Register
    
    Other skills point here when their output must be readable by someone who
    didn't live the work. The reader is technical — pseudocode and precise claims
    land fine — but they have read none of the code and none of the session's
    messages: no diff, no transcript, no labels the work invented along the way.
    Writing ELI5 means:
    
    - Walk one concrete scenario end to end — the triggering event, what happens
      today, what the change (or the unbuilt alternative) would do — instead of
      describing properties in the abstract.
    - Define every term of art at first use; never lean on labels the spec, code,
      or session invented.
    - Use small concrete examples without replacing precise claims with analogies.
    - Reach for pseudocode when explaining control flow, ordering, or timing.
      Prose describing when-something-fires reads as plausible and hides the gap;
      five lines of pseudocode make the gate visible and let the reader see the
      case you missed. Write it at the level of the decision, not the
      implementation — the conditions and their order, not real function
      signatures. The tell that you needed it: your prose contains "only when",
      "before", "unless", or "as soon as" and the reader still cannot say what
      happens on the second call.
    - The test: the text stands alone, without the diff, the spec, or the
      transcript. If the reader must ask "explain this part", it failed.
    
    ## Workflow
    
    1. Read the complete referenced artifact and inspect current owners when the
       spec alone cannot establish behavior. When no artifact is named, explain
       the session's current change (working tree or branch diff). Separate what
       exists today from what is only proposed.
    2. Explain the problem through its user or operational consequence, including
       why the current design produces it.
    3. Explain the solution as one simple before/after data flow. Introduce each
       component by responsibility, not by filename or internal symbol.
    4. Inventory schema and durable-contract changes exhaustively: added, changed,
       removed, reset, and deliberately unchanged. Include cursor or wire-format
       cutovers when they affect stored data or readers. Say explicitly when there
       are no schema changes.
    
    ## Output
    
    Use these headings in order:
    
    1. `Problem`
    2. `Solution`
    3. `Schema changes`
    
    End with one short sentence stating what users should notice after the change.
    
    ## Rules
    
    - Lead with behavior and boundaries; mention implementation names only when
      they clarify ownership or a contract.
    - Distinguish canonical records from derived indexes, caches, summaries, and
      presentation grouping.
    - Call out destructive resets, migration requirements, compatibility behavior,
      eventual consistency, and intentional data loss directly.
    - Do not omit a schema change because it is operational rather than
      user-visible.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.

Related