Skip to content
Agentmart
Submit a tool

mold

Iterative thinking amplifier for fuzzy ideas.

Skill by paulnsorensen

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/paulnsorensen-mold

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Iterative thinking amplifier for fuzzy ideas. Routes input to the right starting mode (Explore, Ground, Shape, Sketch, Grill, Diagnose), runs Validate Cycles to anchor claims, locks down interfaces in pseudocode, and only curdles a spec (and optional issues) after a two-key handshake plus a coherence self-check.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/mold

What's inside

Skill name
mold
Description
Iterative thinking amplifier for fuzzy ideas. Routes input to the right starting mode (Explore, Ground, Shape, Sketch, Grill, Diagnose), runs Validate Cycles to anchor claims, locks down interfaces in pseudocode, and only curdles a spec (and optional issues) after a two-key handshake plus a coherence self-check.
License
MIT
Compatibility
Works in markdown-based coding harnesses with sub-agent support. Heavy delegation to /briesearch and cheez-* skills. When tilth MCP is absent, substitute the host Read tool for local file references — the no-speculation invariant (Operating Principle 2) still holds.
Metadata
owner: cheese-flow · category: planning
Instructions
329 lines in SKILL.md · view on GitHub

Agent Skills spec issues

  • allowed-tools: must be a space-separated string
  • files: referenced but missing: references/sliced-bread.md

Files (9)

  • SKILL.md
  • diagnose-mode.md · references/
  • issue.md · references/
  • loop-detection.md · references/
  • routing.md · references/
  • sketch-mode.md · references/
  • spec.md · references/
  • state-schema.md · references/
  • validate-cycle.md · references/

SKILL.md (first lines)

# /mold

Use this skill when the user has a fuzzy idea and wants to converge — through
dialogue, evidence, and interface lockdown — on a coherent spec (and optional
issues) that downstream `/cheese` and `/cook` can consume without
re-asking design questions.

Do not use this skill for one-shot implementation, free-form rubber-ducking
without artifact intent (use `/culture`), or library-only research (use
`/briesearch` directly).

## What `/mold` is

A **thinking amplifier**, not a pre-prompt. The dialogue is the point;
artifacts are the by-product. The terminal step curdles whatever the
dialogue actually produced — never more, never less.

| Companion | Boundary |
| --- | --- |
| `/culture` | Same dialogue feel; **never writes**. Use it when there is no artifact intent. |
| `/briesearch` | External evidence dispatcher. `/mold` calls it through the Validate Cycle. |
| `/nih-audit` | Whole-repo build-vs-buy sweep. `/mold` calls it from Sketch when multiple library-shaped categories are in play, or offers it at Curdle for migration-style specs. |
| `/cook` | Implements a curdled spec. `/mold` ends with a hand-off offer, never an auto-invoke. |

## Operating principles

1. **No fixed entry point.** Inspect input shape and pick a starting mode.
   Announce the mode in one line. Low-confidence classifications default to
   `Explore`.
2. **Ground every load-bearing claim.** Never speculate about local code or
   external behavior without a `cheez-*` call or a Validate Cycle.
3. **State a hypothesis before researching.** A bare research dispatch is
   discouraged; the Validate Cycle frame forces commitment plus a judgment
   step.
4. **Lock down interfaces before curdling.** Every cross-slice seam
   gets a pseudocode signature with named unknowns, a recommended answer,
   and an explicit Sliced Bread slice (`domains/<name>`, `adapters/<name>`,
   `app`, or `domains/common`). Architecture rules in
   `references/sliced-bread.md` (repo root, not local to this skill).
5. **Two-key handshake.** Both the user (explicit verb) and the agent
   (coherence self-check) must agree before extraction.
6. **Heavy delegation.** `/briesearch` for external research, `cheez-search`
   / `cheez-read` for in-repo grounding. Do not reinvent.
7. **No production writes during the loop.** The only writes happen on
   Curdle, after explicit approval.

## Routing — input shape to starting mode

| Input shape | Start mode | Heuristic |
| --- | --- | --- |
| Stack trace, "X is broken/slow/flaky" | Diagnose | error markers, `file:line` refs, symptom verbs |
| File path, PR ref, existing spec under `.cheese/specs/` | Ground | concrete artifact exists; read it first |
| Half-baked design doc with signatures or schemas | Sketch | already has interfaces; refine them |
| "I want to add X" with concrete nouns | Shape | named the thing → jump to options |
| "Should we do X? thinking about Y" | Grill | tentative plan exists → stress-test it |
| Vague noun, half-sentence, "thinking about" | Explore | no grounded artifact, no chosen direction |

Detail in `references/routing.md`.

## The six modes

Curdle is a terminal state, not a mode.

### Explore — intent extraction
Job: collapse ambiguity with high-leverage questions. Borrow the `Beat 0`
framing (Job-To-Be-Done, Why Now, What This Unlocks, Who Has The Pain, Do
Nothing). Use lettered options to compress decisions.
Exit when: a problem statement plus one concrete pain point is articulated.

### Ground — anti-hallucination
Job: anchor every claim to evidence — code, docs, prior research. Probe
glossary conflicts against any harness convention files (`CONTEXT.md`,
`CLAUDE.md`, project root agent guides) found on entry.
**Sharpen fuzzy language**: when the user uses overloaded or ambiguous
terms ("account", "session", "user"), pause and resolve with a
canonical-term question (e.g. "you said 'account' — do you mean Customer
or User? Those are different things"). Resolved terms get logged in the
state file's `Decisions` block so later modes use the canonical name.
Invariant: never say "I think the code does X" without a `cheez-*` call.
Exit when: every load-bearing claim has a citation.

### Shape — option generation
Job: turn a grounded problem into 2+ candidate approaches with trade-offs.
Always include `Do Nothing`. Recommend with one-line rationale. Validate
Cycle any load-bearing assumption behind a recommendation.
Exit when: an option is picked (→ Sketch) or none survive (→ Explore).

### Sketch — interface lockdown
Job: lock modules, responsibilities, I/O contracts, and seams in pseudocode
signatures, anchored to Sliced Bread slices. Before drafting, parallel
`cheez-search` for sibling signatures in the same slice so new ones fit
conventions and respect the crust (the slice's public API). Run an **NIH
probe** (single Validate Cycle) before locking any signature whose
category is library-shaped — retry, validation, UUID, debounce, date,
argparse, clone, string-case, crypto, sanitizer, format, deep-equality.
If 2+ such categories are in play, call `/nih-audit` once instead.
Exit when: every public seam has a signature and a declared `slice` field;
   every cross-slice call imports from the target slice's crust, never its
   internals; library-shaped categories have an NIH probe verdict in
   `Decisions`. Detail in `references/sketch-mode.md`; architecture rules
   in `references/sliced-bread.md` (repo root, not local to this skill).

### Grill — adversarial clarification
Job: stress-test the chosen approach plus sketched interfaces. **One question
at a time**, paired with the agent's recommended answer. The recommendation
is non-optional. Traverse decision branches and contract corners. When grill
surfaces an unverified assumption, pause and run a Validate Cycle.
Exit when: every branch and contract corner is touched and agent confidence
is at least user confidence.

### Diagnose — symptom inputs
Job: entry mode for stack traces and "X is broken". Phases:
`Build a Loop → Reproduce → Hypothesize (3-5 ranked, falsifiable, parallel
Validate Cycles) → Confirm root cause`. Phase 0 (**Build a Loop**) is the
core discipline — agree on a fast, deterministic, falsifiable feedback
technique (failing test, curl/CLI script, headless browser, replay,
bisection harness, differential loop, ...) before chasing hypotheses.
The chosen loop becomes the Reproduction block in the bug-shaped spec, so
`/cook` can verify the fix against the same signal the diagnosis used.
Diagnose is **diagnostic-only** — hand-off to Shape ("what's the fix?")

From paulnsorensen/cheese-flow, MIT.

More from paulnsorensen/cheese-flow

Signals

Stars
1
Forks
1
Open issues
25
Last commit (this folder)
May 2, 2026 (6 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from paulnsorensen

Cite this page

mold skill. Agent Mart. https://agentmart.dev/skills/paulnsorensen-mold (last updated 2026-10-24)