---
name: Reasoning Harness
slug: reasoning-harness
category: AI Engineering
description: Reasoning Harness calls an MCP reasoning tool before answering analytical, diagnostic, planning, or tradeoff-heavy questions. It is used to build a scaffold for multi-step judgments and reduce premature conclusions.
github: "https://github.com/jeremylongshore/claude-code-plugins-plus-skills/tree/main/plugins/community/ejentum-reasoning"
language: Python
stars: 2617
forks: 384
install: "npx degit https://github.com/jeremylongshore/claude-code-plugins-plus-skills/tree/main/plugins/community/ejentum-reasoning ~/.claude/skills/ejentum-reasoning"
installs_to: ~/.claude/skills/ejentum-reasoning
source_path: plugins/community/ejentum-reasoning/SKILL.md
collection_size: 53
category_size: 2451
collection_url: "https://dirskills.com/collections/jeremylongshore/claude-code-plugins-plus-skills"
added: 2026-08-12T04:42:56.723Z
last_synced: 2026-08-12T04:42:56.723Z
canonical_url: "https://dirskills.com/skills/reasoning-harness"
---

# Reasoning Harness

Reasoning Harness calls an MCP reasoning tool before answering analytical, diagnostic, planning, or tradeoff-heavy questions. It is used to build a scaffold for multi-step judgments and reduce premature conclusions.

**Install:**

```bash
npx degit https://github.com/jeremylongshore/claude-code-plugins-plus-skills/tree/main/plugins/community/ejentum-reasoning ~/.claude/skills/ejentum-reasoning
```

## README

# Reasoning Harness

When this skill triggers, call the `reasoning` tool from the `ejentum` MCP server. Pass a 1-2 sentence framing of WHAT you are reasoning about as the `query` argument. Be specific about the task, not what tool you want.

Good query: `diagnose why a microservice returns 503s under load`
Bad query: `help me think`

The tool returns a structured scaffold containing:

- `[NEGATIVE GATE]`: failure pattern to avoid
- `[PROCEDURE]`: steps to follow
- `[REASONING TOPOLOGY]`: decision flow with gates and traps
- `[TARGET PATTERN]`: correct shape your reasoning should take
- `[FALSIFICATION TEST]`: self-check criterion
- `Amplify:` signals to engage
- `Suppress:` failure modes to block

Absorb the scaffold internally and shape your response with it. The bracketed fields are instructions, not content to display. Do NOT echo the bracket labels, do NOT name the topology, do NOT meta-comment on calling the tool. The user-facing reply is naturally phrased and shaped by the injection.

If the API is unreachable or returns an error, proceed with native reasoning. The scaffold enhances; it is not a hard dependency.

Latency cost: ~1 second. Benefit: reasoning quality the model cannot reliably reproduce on its own for non-trivial tasks.