Watson Toolkit

24 skills that fix the ways Claude actually breaks: agreeing with everything, skipping the thinking, guessing at fixes instead of finding root causes, writing like a robot, declaring victory without proof, and not being able to work while you're away.

Skills follow the open Agent Skills format, so they work in Claude Code, Cowork, Codex, and any harness that reads SKILL.md directories.

Install

Cowork: Settings > Add marketplace > https://github.com/jeremyknows/watson-toolkit

Claude Code (marketplace):

/plugin marketplace add jeremyknows/watson-toolkit
/plugin install watson-toolkit@watson-toolkit

Claude Code / Codex / anything else (npx):

npx watson-toolkit

Copies the skill directories into ~/.claude/skills/. For other harnesses, point the installer at your skills directory or copy plugins/watson-toolkit/skills/* wherever your harness loads skills from.

What's in here

Think before building

Claude's default is to start writing code the second you describe a problem. These skills force the design conversation to happen first.

| Skill | What it does | |-------|--------------| | brainstorming | Interviews you about what you actually want before producing a spec. Without it, Claude skips straight to code without understanding the problem. | | grill-me | Walks every branch of your decision tree. Forces you to resolve the ambiguities you were hoping to ignore. Doesn't stop until there's shared understanding. | | grill-with-docs | Same relentless interview, but grounded in your project docs — it reads them first, then grills you on the gaps between what's written and what you're asking for. | | plan-review | Two phases: first, is this the right thing to build? Then, is this the right way to build it? Catches bad plans before they turn into bad code. | | writing-plans | Turns specs into step-by-step implementation plans that coordinate across files. The thing that goes between deciding what to build and building it. |

Push back

Claude agrees with you too much. These skills add real resistance.

| Skill | What it does | |-------|--------------| | intellectual-honesty | Honest assessment over comfortable agreement. Has a framework for separating what's actually being said from what's being presented, and where reasoning breaks down. | | receiving-feedback | Anti-sycophancy protocol for processing feedback on anything: code reviews, doc edits, strategy. Rules for when to push back and when to accept. | | complete-code-review | Spawns 5 parallel reviewers with confidence scoring. The disagreements between reviewers are where the real design decisions live. | | prism | Parallel adversarial review by independent specialists. Disagreements are more valuable than consensus. | | boil-the-ocean | The under-build counter. Catches "MVP shipped" rationalizations at the moment they happen — TODO comments, workarounds instead of fixes, missing tests — and asks whether the missing piece is bounded (finish it now) or unbounded (flag it and stop). | | verification-before-completion | No "done" without evidence. Requires running the verification command and reading its output before any success claim. The gate between "should work" and "works." |

Build and debug

| Skill | What it does | |-------|--------------| | systematic-debugging | Four phases: observe, hypothesize, test, fix. You have to finish observing before you can propose a fix. Stops the spiral of random patches. | | autoresearch | Autonomous iteration. Sets a goal, loops (modify, verify, keep or discard, repeat), runs bounded iterations until the metric converges. | | sprint-cowork | Walk-away work. Set goals with acceptance criteria, start a sprint, leave for hours. A director fires every 25 minutes, dispatches workers, tracks progress, handles stalls, writes a report when it's done. You can also give it a vague topic like "improve security" and it'll scope the work itself in the first cycle, then execute. | | baton | Session continuity. When work will continue in a future session, write a "baton" — a cold-start-ready continuation prompt with state inventory, self-audit, and a dry-run as a cold reader. The next session picks up without re-discovery. |

Write like a human

Claude has tells. These find and fix them.

| Skill | What it does | |-------|--------------| | humanizer | Catches 15+ specific AI writing patterns: inflated symbolism, em dash overuse, "delve"/"tapestry"/"landscape", rule of three, promotional hedging. Rewrites to remove them. Based on Wikipedia's "Signs of AI writing" guide. | | doc-coauthoring | Three stages: context interview, iterative refinement, then reader testing with a fresh Claude that catches assumptions the author and AI share but a new reader wouldn't. |

Explore and create

| Skill | What it does | |-------|--------------| | playground | Builds single-file interactive HTML tools. Controls on one side, live preview on the other, copyable output. Templates for design, data, concept maps, code exploration. | | random-thought | Picks a random insight from your own workspace docs. Good for rediscovering things you wrote and forgot about. | | x-fetch | Pulls tweet text, threads, articles, and engagement stats from X URLs via fxtwitter. Works where direct scraping doesn't. | | last30days | Deep recency research across 10+ sources — Reddit, X, YouTube, Hacker News, Bluesky, the web — synthesized into a grounded, cited report of the last 30 days on any topic. |

Skill and plugin tooling

Build your own, audit what you have, publish when it's ready.

| Skill | What it does | |-------|--------------| | skill-doctor | Health audits with scoring (target: 11/14+), reference extraction for progressive disclosure, and vetting new skill installs before they pollute your context. | | publish-skills | Pre-publish checklist: frontmatter, directory structure, LICENSE, README, hardcoded reference scan — plus a sanitizer pipeline for skills authored inside a private workspace. Catches the stuff that makes skills fail on install. | | publish-cowork-plugin | Validates, packages, and publishes Cowork/CC plugins. Structure validation through marketplace setup. |

Provenance: which skills are original, which are third-party

This toolkit bundles original work alongside excellent open-source skills from other authors. Original skills are maintained here (and in standalone repos); third-party skills are included unmodified from their upstream and retain their original licenses — credit and maintenance belong to their authors.

| Skill | Origin | Upstream | |-------|--------|----------| | prism | Original | jeremyknows/PRISM | | complete-code-review | Original | jeremyknows/complete-code-review | | skill-doctor | Original | jeremyknows/skill-doctor | | publish-skills | Original | jeremyknows/publish-skills | | x-fetch | Original | jeremyknows/x-fetch | | random-thought | Original | jeremyknows/random-thought | | intellectual-honesty | Original | this repo | | receiving-feedback | Original | this repo | | plan-review | Original | this repo | | sprint-cowork | Original | this repo | | publish-cowork-plugin | Original | this repo | | boil-the-ocean | Original | this repo | | baton | Original (portable edition) | this repo | | playground | Original, derived from Anthropic's official plugin collection | anthropics/claude-plugins-official | | brainstorming | Fork of obra/superpowers (substantially modified) | obra/superpowers | | systematic-debugging | Fork of obra/superpowers (substantially modified) | obra/superpowers | | writing-plans | Fork of obra/superpowers (substantially modified) | obra/superpowers | | grill-me | Third-party, unmodified | mattpocock/skills by Matt Pocock | | grill-with-docs | Third-party, unmodified | mattpocock/skills by Matt Pocock | | verification-before-completion | Third-party, unmodified | obra/superpowers by Jesse Vincent | | humanizer | Third-party, unmodified | blader/humanizer by @blader | | autoresearch | Third-party, unmodified | uditgoenka/autoresearch by @uditgoenka | | last30days | Third-party, unmodified | mvanhorn/last30days-skill by @mvanhorn | | doc-coauthoring | Third-party, unmodified | Anthropic |

If you only want one skill, the standalone repos linked above install individually — you don't need the whole toolkit.

Maintenance

Skills are authored and battle-tested in a private agent workspace, then published here through an automated pipeline that refreshes every skill from its canonical source, strips workspace-private content, and runs a leak scan before anything is staged. The pipeline (built on the bundled publish-skills sanitizer) means this repo is a generated mirror — skill content is never edited here directly.

License

MIT. See LICENSE. Bundled third-party skills retain their original licenses — see individual skill directories.

Changelog

2.0.0 — 2026-06-02

• Every skill refreshed to its current canonical version (the toolkit had been frozen since April; prism alone gained simplicity-pass mode, wiki mode, and custom review panels since then)
• New skills: grill-with-docs, boil-the-ocean, last30days, baton (portable edition), verification-before-completion (obra's upstream)
• Removed: decide — its default data sources assume private infrastructure; it returns to the toolkit after a standalone refactor
• prism sanitized — legacy workspace paths and internal role names in the published skill replaced with generic equivalents
• New provenance table — original vs. third-party, with upstream links and attribution
• New publish pipeline — replaces the old bidirectional sync scripts with a one-way generated-mirror model (fleet canonical → sanitize → review → publish)
• Harness-agnostic framing — Agent Skills format; CC, Cowork, Codex, npm install paths
• Removed: dead sync scripts (sync-watson-toolkit.sh, sync-github-skills.sh, SYNC.md, sync-manifest.txt)

1.5.0 — 2026-03-30

• Curated 20-skill set, npx installer, publish-spec compliance, attribution corrections