github oaustegard/claude-skills challenging-v0.8.1
challenging v0.8.1

challenging

Cross-context adversarial review for deliverables before shipping — blog posts, technical recommendations, analysis briefs, code, or any artifact where accuracy matters more than speed.

The value comes from fresh context: no shared blind spots with the conversation that produced the artifact, no accumulated goodwill. The adversary sees the draft cold and only the draft.

See SKILL.md for the full protocol, profile table, system prompts, and the drill loop. See CHANGELOG.md for version history.

Two paths

Environment Adversary Why
Claude Code (primary) Native sub-Claude via the Task tool No key, no network dependency, fresh context
claude.ai, Codex, headless Gemini 3.1 Pro (default) or Anthropic API No subagents available; Gemini also gives genuine cross-model diversity

Five profiles

Each lives in its own references/*.md file — read only the one you need.

Profile Use for Iteration
prose Blog posts, essays, articles parallel replay
analysis Research briefs, comparisons, synthesis parallel replay
code Scripts, implementations, PRs parallel replay
recommendation Technical decisions, architecture choices parallel replay
drill 5 Whys on one finding from a review sequential deepen

Review profiles replay in parallel — each pass independent, novelty tracked so confabulated findings get filtered. Drill deepens sequentially — one why-level per pass, conditioned on the chain so far, until bedrock or max depth, then a synthesis pass extracts root causes. Patches fix the instance; drills fix the class.

Verdicts

SHIP — clean, deliver. REVISE — real issues, sound core. RETHINK — structural problems, reconsider.

Provenance

  • VDD persona & anti-rationalization patternsdollspace.gay
  • Grainulation's confabulation-resistance heuristics
  • 5 Whys (drill profile) — adapted from Tim Kellogg's open-strix writeup

Complements

  • generative-thinking — generates distance before evaluation
  • convening-experts — synthesizes multiple role-based viewpoints
  • tiling-tree — exhaustive MECE partitioning of a solution space

This skill evaluates. It does not generate, synthesize, or partition.

Dependencies

No skill dependencies. requests is loaded lazily — only the API path needs it.

Credentials are only required for the API path (see SKILL.md).

Skill folder: challenging


Release of challenging version 0.8.1

📥 Download & Install

⬇️ Download challenging.zip

To install:

  1. Click the download link above (ignore the "Source code" archives below - they're auto-generated by GitHub)
  2. Go to Claude.ai Skills Settings
  3. Upload the downloaded ZIP file
  4. Requires paid Claude Pro or Team account

See official documentation for more details.

Recent Changes

32d781e challenging v0.8.1: add README
1308ae4 docs: Update CHANGELOG.md for released skills

Don't miss a new claude-skills release

NewReleases is sending notifications on new releases.