Thinking Methods
Structured thinking skills that turn AI coding assistants into collaborators for decisions, brainstorming, and root-cause work.
- Source repo
- ysskrishna/ai-agent-skills
- Stars
- ★ 20
- Last updated
- 19d ago
- License
- MIT
- Primary language
- Python
- FA score
- 61/100 · Some gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- An engineering team choosing between Postgres, Redis and a managed queue uses tradeoff-analysis to get a structured comparison before committing.
- Main limitation
- Provides method scaffolding only — it does not read project code, logs or metrics, so output quality depends on context the user supplies.
- Source review
- 61/100 · Some gaps
What does this agent do, and when should you use it?
Thinking Methods is a collection of 17 SKILL.md skills built on the Agent Skills open standard, packaging methods such as five whys, tradeoff analysis, pre-mortem, SWOT, and six thinking hats as installable units for AI agents. Each skill is description-triggered: its frontmatter lists the situations and phrasings it fits and when to skip it, so users describe a problem rather than naming a method. Skills install as a bundle or individually into Claude Code, Codex, Gemini CLI, Cursor, OpenCode, GitHub Copilot CLI and dozens of other hosts, which then invoke them during conversation. Every skill is standalone, includes a light path for small questions, and ships a worked example in its references/ folder; thinking-method-selector acts as an entry point that picks and runs a suitable method. The repo validates skills through a GitHub Actions workflow, is MIT licensed, and is also published to ClawHub and skills.sh. Nothing in the repo calls a model itself — reasoning comes entirely from the host agent.
Each skill is a directory containing SKILL.md whose frontmatter declares applicable situations, trigger phrases, and exclusions, and whose body gives the step-by-step procedure for that method. When a user describes a situation inside a host agent, the host matches one of these descriptions and executes the steps, producing output such as a chained root-cause analysis (five-whys), a side-by-side comparison of Postgres, Redis and a managed queue (tradeoff-analysis), or a list of imagined failure causes before a launch (pre-mortem). thinking-method-selector inspects the problem first, chooses a method, and runs it. The repo also ships scripts/publish_clawhub.py, a maintainer script that plans and publishes the bundle and individual skills to ClawHub.
- An engineering team choosing between Postgres, Redis and a managed queue uses tradeoff-analysis to get a structured comparison before committing.
- A team about to ship a risky database cutover runs pre-mortem to surface failure modes and mitigations in advance.
- An on-call engineer facing a recurring deploy failure uses five-whys to reach a cause that can actually be fixed.
- A product manager with more work than capacity uses prioritization to produce an order and an explicit cut line.
- A designer or PM uses design-thinking to understand users before committing to a build.
- A user who does not know which method fits hands the problem to thinking-method-selector, which selects and runs one.
How do you install or deploy this agent?
Fastest path, works across 70+ hosts via skills.sh:
npx skills add ysskrishna/ai-agent-skillsInstall a single skill:
npx skills add ysskrishna/ai-agent-skills --skill pre-mortemClaude Code plugin marketplace:
/plugin marketplace add ysskrishna/ai-agent-skills
/plugin install ai-agent-skills@ai-agent-skillsCodex CLI and App (requires Codex CLI v0.122 or later):
codex plugin marketplace add ysskrishna/ai-agent-skills
codex plugin add ai-agent-skills@ai-agent-skillsGemini CLI:
gemini extensions install https://github.com/ysskrishna/ai-agent-skillsGitHub CLI users (v2.90.0+):
gh skill install ysskrishna/ai-agent-skillsOpenCode — add the plugin to opencode.json and restart:
{
"plugin": ["@ysskrishna/ai-agent-skills"]
}How do you use this agent?
After installing, describe the situation in plain language; the host matches a skill from its description:
We have to pick a job queue: Postgres, Redis or a managed service. Team of four.
Why did last night's deploy fail? Keep asking why until we reach something we can fix.You can also call a skill by name. After a skills.sh or gh skill install, use a slash command:
/pre-mortemClaude Code plugin installs namespace skills by plugin name:
/ai-agent-skills:pre-mortemWhen unsure which method fits, start with the selector:
/ai-agent-skills:thinking-method-selectorWhat are this agent's strengths and limitations?
- 17 independent skills let you install only what you need instead of loading everything into context.
- Description-triggered design states when each skill fits and when to skip it, so users describe problems instead of memorizing method names.
- Wide documented host coverage — Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Copilot CLI, ClawHub and more — with npx skills as a common entry point.
- Each skill has a light path for small questions plus a worked example in references/, avoiding over-processing trivial asks.
- thinking-method-selector lowers the cost of first use by choosing and running a method automatically.
- MIT licensed, so the skill text can be read and adapted directly.
- Provides method scaffolding only — it does not read project code, logs or metrics, so output quality depends on context the user supplies.
- Install commands differ per host (Claude Code, Codex, Gemini, Kimi, Muse, Grok and others), so multi-host teams maintain separate instructions; Muse also needs an experimental flag.
- Matching depends on the host model interpreting skill descriptions; the repo publishes no benchmark showing results beat plain conversation.
- Publishing and distributing to ClawHub and skills.sh relies on maintainer scripts such as scripts/publish_clawhub.py.
- Grok Build does not support the marketplace route and requires a direct install.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Thinking Methods This agent | 61 · Some gaps | — | ★ 20 | 19d ago | Python | Codex · Claude Code |
| Council Decision Workspace | 85 · Good | Desktop appFree + model costs | ★ 111 | 2mo ago | Python | OpenAI API |
| SandBase Skills Library | 50 · Major gaps | Agent plugin / skillFree + model costs | ★ 206 | 1mo ago | Python | Codex · Claude Code |
| Material Learning Studio | 92 · Excellent | Agent plugin / skillFree | ★ 1.3k | 16d ago | Python | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The repo is a pure SKILL.md prompt collection with no script execution, network calls, or credential access, so the least-privilege surface is naturally small (2). However install paths (npx skills add, multiple plugin marketplaces, grok --trust, Kimi 'Trust and install') pull in third-party toolchains and trust prompts, and the README never states what those installers write or read, so user_confirmation and data_flow_transparency are only 1. Sensitive data handling is unaddressed (skills process user decisions/business info with no non-exfiltration statement) — 1. Dependency security shows only an empty dependencies field and node>=18 engine, no lockfile or audit — 1. External effects are prompt-level only with no destructive defaults — 2. Rollback has per-platform uninstall/remove commands — 2. Attribution is clear (author, GitHub, site, MIT copyright) — 2.
The skills table, install commands, package.json version 1.4.1, and publish workflow are mutually consistent (2). Dependency availability rests on external tools with version gates (npx skills, gh skill v2.90, Codex v0.122, OpenCode 2.0.4) and no lockfile — 1. Failure messages cover only one publickey error fix, with no systematic error handling — 1.
Audience and scenarios are broad (17 skills, grouped by problem type, multiple trigger examples) — 3. Capability boundaries mention a light path and when to skip, but no systematic non-applicability statement — 2. Trigger precision depends on each SKILL.md frontmatter, which the README only summarizes — 2. Environment fit lists 18+ hosts with per-host install steps — 3.
Information architecture is clear (grouped tables, collapsible install blocks, anchors) — 3. Install notes are thorough — 3. Naming stability is a concern: ClawHub names diverge from directory names (first-principles vs first-principles-reasoning, six-thinking-hats vs six-hats-thinking) — 2. Examples exist but no FAQ — 2. Known limitations are nearly absent (only the Hermes plugin caveat) — 1. MIT license is complete — 3. A CHANGELOG link and release workflow exist but the file is not in evidence — 2. Maintenance responsibility names a single maintainer but offers no governance or response commitment — 2.
Output usability depends on SKILL.md content, of which only README summaries are in evidence — 2. Marginal value over built-in model reasoning is plausible but unsupported by comparison evidence — 2. Cost is low (pure prompts, per-skill install) with reasonable benefit — 2.
A test badge and validate-skills.sh reference exist, but the script and SKILL.md bodies are not in evidence, so claim_traceability is 1. Cross-source corroboration is limited to README/package.json/workflow agreement — 1. Fact/inference separation is weak: README leans on marketing phrasing ('upgrade your AI agents') without quantified support — 1.
- Installation relies on multiple third-party toolchains (npx skills, gh skill, plugin marketplaces); the README does not state what directories they write or data they read — verify before installing.
- ClawHub skill names diverge from repo directory names (first-principles-reasoning, six-hats-thinking), risking confusion when installing by name.
- SKILL.md bodies and validate-skills.sh are absent from evidence, so the test badge and method quality cannot be verified; this is a static source review with no execution.
- Publisher identity is unverified by the FollowAgents enterprise registry; a single maintainer with no governance or response commitment.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v1.4.1 Release Version v1.4.1
Release notes excerpt
Added - ClawHub bundle package — openclaw.plugin.json for OpenClaw bundle-plugin publish as @ysskrishna/ai-agent-skills (display name Thinking Methods). .clawhubignore excludes dev-only paths from package uploads. - publish_clawhub.py package — package plan and package publish sync the bundle via clawhub package publish (per-skill plan / publish unchanged).View on GitHub ↗ -
v1.4.0 Release Version v1.4.0
Release notes excerpt
Added - npm package @ysskrishna/ai-agent-skills with a files whitelist (skills, manifests, assets, OpenCode entry). Trusted publishing via GitHub Actions (publish-npm.yml, environment prod) on v* tags and manual dispatch. - OpenCode and Pi npm install documented in the README and [docs/opencode-setup.md](docs/opencode-setup.md) / [docs/pi-setup.md](docs/pi-setup.md). Changed …View on GitHub ↗ -
v1.3.1 Release Version v1.3.1
Release notes excerpt
Added - Listing metadata: display name Thinking Methods, a Codex subtitle of 30 characters or fewer, skills.sh groupings, the Matrix logo, and the horizontal lockup (mark, Thinking Methods on one line, AGENT SKILLS subtitle). Changed - Removed the root CLAUDE.md file so strict Claude plugin validation passes. - Plugin keyword lists now include the search terms people type.View on GitHub ↗