spar-exocortex

Validates algorithm and code-design solutions through adversarial propose-attack-measure loops with execution-backed testing.

Updated Mar 30, 2026
One-click install
npx skills add https://github.com/rubrical-works/idpf-praxis-skills --skill spar-exocortex-rubrical-works
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: spar-exocortex
Source: https://github.com/rubrical-works/idpf-praxis-skills/tree/main/Skills/spar-exocortex
Command: npx skills add https://github.com/rubrical-works/idpf-praxis-skills --skill spar-exocortex-rubrical-works

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires ajv, and includes scripts (resource) components.

What problem does it solve? Plausible-looking algorithmic and code-design solutions often hide failure modes—complexity blow-ups, correctness bugs, or operational fragility—that pure reasoning misses. This Skill stress-tests a baseline solution by dispatching an attacker subagent to produce a concrete breaking input, a challenger subagent to propose a surviving alternative, and an execution harness to run both candidates against shared tests, so a judge decides based on measured evidence rather than argument. ## Core Features & Use Cases - Adversarial propose-attack-measure loop: Baseline proposal, attacker producing concrete failing inputs, challenger proposing a different approach, execution harness running both candidates, and a judge deciding or demanding round two. - Execution-backed validation: Runs candidate implementations in isolated subprocesses via JavaScript/TypeScript (Node) and Python adapters, capturing pass/fail/timeout outcomes and wall-clock timings. - Schema-validated subagent outputs: JSON Schema validation (via ajv or an inline no-Node fallback) enforces concrete attacker inputs, anti-overlap between baseline and challenger, and structured judge decisions. - Use Case: Ask whether your LRU cache with TTL is correct under adversarial load—the attacker constructs a 10,000 short-TTL insert wave, the challenger proposes an eager-expiry min-heap design, and the harness shows the baseline taking 17x longer on the attack input, so the judge endorses the challenger. ## Quick Start Invoke /spar-exocortex with your algorithmic or code-design problem, for example: stress-test my LRU cache with TTL implementation in JavaScript.

Frequently Asked Questions about spar-exocortex

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I stress-test an algorithm implementation for hidden failure modes?▼

Invoke /spar-exocortex with your problem statement. It drafts a baseline implementation, dispatches an attacker subagent to produce a concrete failing input, has a challenger propose a surviving alternative, then executes both against shared tests so a judge decides on measured evidence.

What languages does the spar-exocortex execution harness support?▼

The harness ships adapters for JavaScript and TypeScript (requiring Node.js 18+) and Python (requiring python3). One adapter suffices per invocation; problems targeting other languages halt with a scope-statement diagnostic since there is no reasoning-only fallback.

When should I use spar-exocortex instead of engage-exocortex?▼

Use spar-exocortex when you have a single cleanest-looking answer you want pressure-tested with executable validation. Use engage-exocortex for open-ended cooperative exploration of multiple approaches, or when no supported runtime is available for execution.

Does spar-exocortex work without Node.js installed?▼

Partially. Without Node, schema validation falls back to inline prose-based checking at higher token cost, but the execution harness still requires either Node.js 18+ or Python 3. With neither runtime, the skill halts because execution-backed validation is intrinsic.

Why does the attacker output get rejected during validation?▼

The attacker schema requires a concrete failing input or a scenario-plus-trigger object for operational failures. Verbal-only criticism without executable data fails validation and triggers one re-dispatch; a second failure marks the path as attackerValidationFailed.

What is the operational-graft step in the judge decision?▼

For ops-critical problems, the judge must ask whether the losing candidate wins on extensibility, simplicity, transparency, or cost. If so, it names concrete grafts from the loser into the winner, catching cases where the architecturally optimal solution is operationally fragile.