Intuitive Reduce Entropy

---
name: intuitive-reduce-entropy
description: Find and rank material simplification candidates when the cleanup target or plan risks are unclear.
---

# Intuitive Reduce Entropy

Use this skill to find a bounded set of changes that remove real maintainer or
execution friction. Prefer deletion, reuse, and one canonical owner over new
workflow, code, test, or documentation surfaces.

## Modes

| Mode | Use when | Output | Redirect when |
| --- | --- | --- | --- |
| Repo entropy | The repo feels messy, stale, hard to navigate, or has no known cleanup owner. | Ranked maintenance candidates and specialist owners. | A concrete seam is already selected; use its specialist or `$intuitive-refactor`. |
| Plan entropy | An idea, draft plan, or preflight needs missing decisions, weak assumptions, scope leaks, or proof gaps found. | Ranked plan risks and the next planning owner. | The contract is already approved; use `$intuitive-flow`. |
| Discovery loop | The user asks to continue until meaningful cleanup saturates. | Fresh bounded rounds with deduplicated clear, parked, and rejected items. | A selected queue is ready for `$intuitive-refactor`. |

Pick the mode and discovery intensity (quick scan, selection scan, or
saturation scan) and name them in the packet, with a redirect only when another
mode or skill fits better.

## Read First

Load only the reference needed for the active pass:

| Need | Read |
| --- | --- |
| Mode selection, discovery-loop behavior, plan-ledger boundaries | [Discovery modes](references/discovery-modes.md) |
| High-noise summaries, materiality gate, no-change threshold | [High noise and materiality](references/high-noise-and-materiality.md) |
| Ranking, architecture review, delegation, specialist routing | [Ranking and routing](references/ranking-and-routing.md) |
| Candidate packet, user decision gates, handoff and final shapes | [Handoff and reporting](references/handoff-and-reporting.md) |

For broad repo scans, read the repo's thin orientation surface and run the
bundled high-noise summary before searching planning archives, logs, generated
trees, or other high-noise roots. Deep-read only enough to prove a candidate.

## Candidate Gate

Every selected candidate must pass both gates:

1. Demand sanity: the change deserves to exist now. For additions, explain why
   reuse, narrowing, deletion, or documentation cannot achieve the outcome. For
   removals, show why preserving the surface is stale, misleading, or costly.
2. Materiality: evidence shows false confidence, live source drift, a stale
   reachable surface, real workflow friction, or recurring rediscovery.

Use `scripts/materiality-gate.mjs` when a deterministic gate helps compare a
batch. Scripts provide evidence; they do not override repo facts or a required
human decision.

Rank deletion and cleanup candidates in this order:

1. Stale surfaces with an active replacement.
2. Compatibility aliases, wrappers, or legacy paths without a live contract.
3. Duplicate owners for one concept, state envelope, fixture, or rule.
4. Pass-through modules and tests/docs that keep stale concepts alive.
5. Bounded source-of-truth or verification drift.

Formatting, line motion, speculative abstraction, and taste are not selected
candidates. Park work that needs product direction, public migration approval,
credentials, hardware, paid services, or unavailable proof.

## Ownership Boundary

- This skill owns discovery, ranking, evidence, and routing.
- `$intuitive-refactor` owns mutation of selected cleanup seams.
- `$intuitive-doc`, `$intuitive-init`, and `$intuitive-tests` own their
  specialist surfaces.
- `$grill-with-docs-batch` owns unresolved human/domain decisions.
- `$intuitive-preflight` owns the approved execution contract.
- `$intuitive-flow` starts only after that contract is approved.

Use the [shared delegation policy](../_shared/references/delegation.md) for
independent read-only probes. Keep the main session responsible for ranking,
architecture decisions, and final synthesis.

## Stop And Handoff

Stop when a decision-complete ranked packet is ready, or when a saturation pass
finds no new material candidate. The user selects the subset and approves
implementation. If nothing passes, say `Selected candidates: none` and stop
there, with no gate, commit, or follow-up proposal.

End with a packet the next owner can act on without repeating the audit.
Always include selected candidates (with evidence, specialist owner, and proof
command), `Recommended next action:`, and `Shortcut:` (a ready-to-paste
command, or `park`/`none` when there is no useful next step). Add parked items
or saturation status only when they carry information.