taches-cc-resources: A Claude Code Command, Skill and Subagent Pack
A collection of my favorite custom Claude Code resources to make life easier.
At a glance
- What is it?
- glittercowboy/taches-cc-resources bundles 27 slash commands, 9 skills and 3 auditor subagents for Claude Code, installed through a plugin marketplace or by copying files into ~/.claude. It is a personal toolkit, not a framework, and the last push to main was on 2026-04-01.
- Who is it for?
- Adopt it if you already run Claude Code daily and want staged meta-prompting, todo handoff, thinking-model commands and extension generators without writing them yourself. Skip it if you need a versioned, supported product with release notes: the README documents no rollback path, there are no retrieved releases, and the last push was 2026-04-01.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 167 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What taches-cc-resources solves for Claude Code users
Claude Code ships with a small set of built-in behaviours. Everything specific to how you plan, debug or capture tasks has to be written by hand, and the README frames the project as a response to that gap: a growing collection of custom resources built for real workflows, written by someone who states the guiding assumption as "it's your responsibility to assume everything is possible." That is a personal philosophy, not a product promise, and it shapes what you get.
The pack covers four kinds of work. Commands (27 of them) are slash commands that expand into structured workflows: meta-prompting, todo management, context handoff, thinking models, deep analysis. Skills (9) are described as autonomous workflows that research, generate and self-heal, including generators for MCP servers, agent skills, slash commands, subagents and hooks. Agents (3) are subagents named skill-auditor, slash-command-auditor and subagent-auditor, used by the audit commands. The intended user is a solo developer who already treats Claude Code as a working environment rather than a chat window, and who wants the planning and debugging scaffolding to live in the tool instead of in a notes file. If you use Claude Code occasionally, most of this will sit unused.
How the commands, skills and subagents fit together
The repository splits into commands/, skills/, agents/, docs/ and .claude-plugin/ at the top level. Commands are markdown files that Claude Code expands; the README links each one, for example ./commands/create-prompt.md, ./commands/whats-next.md and ./commands/debug.md. Skills are directories, for instance ./skills/create-plans/, with the README describing create-plans as a pipeline: Brief, Roadmap, Research (if needed), PLAN.md, Execute, SUMMARY.md. Agents are single markdown files under agents/.
The design idea worth naming is that commands and skills are not independent. Several commands are wrappers that invoke a skill: /create-agent-skill, /create-meta-prompt, /create-slash-command, /create-subagent and /create-hook all route into the corresponding skill, and the three /audit-* commands invoke the auditor subagents. So the pack has an extension layer and a review layer. The extension layer generates new resources; the review layer checks them against best practices. /heal-skill closes the loop by fixing skills based on execution issues.
Two mechanisms stand out. Meta-prompting separates analysis from execution: /create-prompt generates an optimized prompt with XML structure, and /run-prompt executes saved prompts in sub-agent contexts, so the generating context does not pollute the running one. Context handoff writes a document you reference with @whats-next.md to resume work in a fresh session. Both are attempts to manage context as a finite resource, which is the honest problem this collection is aimed at.
create-plans adds a domain-awareness hook: it can optionally load framework-specific expertise from ~/.claude/skills/expertise/ (the README gives macos-apps and iphone-apps as examples) so that plans are concrete rather than generic. Those expertise skills are themselves produced with create-agent-skill and are described as exhaustive knowledge bases of 5k to 10k or more lines. That is a large artefact generated by another part of the same pack, and the README does not describe how you keep such a knowledge base current.
Installing taches-cc-resources and running a first command
The README gives two install paths and recommends the first. The plugin route adds the repository as a marketplace and then installs the plugin:
# Add the marketplace
claude plugin marketplace add glittercowboy/taches-cc-resources
# Install the plugin
claude plugin install taches-cc-resourcesAfter that, the README says to start a new Claude Code session to use the commands and skills. If the plugin route is unavailable to you, the manual route clones the repository and copies the two directories into your Claude configuration:
# Clone the repo
git clone https://github.com/glittercowboy/taches-cc-resources.git
cd taches-cc-resources
# Install commands
cp -r commands/* ~/.claude/commands/
# Install skills
cp -r skills/* ~/.claude/skills/Commands install globally to ~/.claude/commands/ and skills to ~/.claude/skills/. The README states that project-specific data such as prompts and todos lives in each project's working directory, which is what makes the todo and handoff commands portable across projects.
A reasonable first use is the meta-prompting pair, because it does not require any generated artefact to exist yet. Describe what you want in natural language, let /create-prompt produce a structured prompt, then run it with /run-prompt so that execution happens in a fresh sub-agent context. You should end up with a saved prompt file and a result produced away from the conversation that generated it. If you want to see the review layer instead, run /audit-slash-command against one of the installed commands and read what the slash-command-auditor subagent reports.
Where the pack gets in the way
The manual install copies files into ~/.claude/commands/ and ~/.claude/skills/ with cp -r and no manifest. The README does not document rollback, so undoing an install means knowing which files came from this repository and removing them yourself. If you later install the plugin, you have two copies of the same commands and no documented precedence rule.
The deeper limitation is the personal-toolkit shape. There are no retrieved releases, so there is no version number to pin and no changelog to read before updating. The last push to main was on 2026-04-01, which is more than six months before today; treat the repository as stable-but-untouched rather than actively developed, and do not expect fixes to arrive on a schedule. The README does not document a test suite or a compatibility matrix against Claude Code versions, so a command that works today may depend on behaviour that changes.
The thinking-model commands are the weakest fit for anyone outside the author's habits. Twelve /consider:* commands (pareto, inversion, 5-whys, swot and so on) each encode a mental framework as a prompt. If you already have opinions about how to structure a decision, these add a layer of ceremony. And create-plans assumes a solo developer plus Claude workflow where PLAN.md is the prompt rather than documentation. On a team that expects plans to be reviewed as documents, that inversion will read as a missing artefact, not a feature.
How this differs from writing your own Claude Code commands
The obvious alternative is not another repository; it is authoring your own commands, skills and subagents directly in ~/.claude/. That approach gives you exactly the resources you need and nothing else, and it keeps you in control of naming and structure. The cost is that you write the scaffolding before you get any benefit, and you have to work out the conventions for command structure, skill layout and subagent configuration yourself.
This pack's answer is to include the scaffolding generators and an audit layer. /create-slash-command, /create-agent-skill, /create-subagent and /create-hook build new resources, and skill-auditor, slash-command-auditor and subagent-auditor review them for best-practices compliance. Writing your own commands by hand gives you no equivalent review step unless you build one. The trade-off runs the other way too: the auditors encode this author's view of best practices, and the README does not say what those criteria are or where they are documented, so you are adopting a standard you cannot read up front.
A second alternative is to use only the parts you want. Because commands are individual markdown files and skills are directories, the manual install can be narrowed: copy commands/create-prompt.md and commands/run-prompt.md, leave the rest. That is a legitimate way to use the repository, and it avoids the twelve /consider:* commands if they do not match how you think. The plugin install does not offer that granularity.
Maintenance, upgrades and the MIT licence
Maintenance is the weak point. The repository is not archived, but the last push was on 2026-04-01 and no releases were retrieved, so there is no published artefact to track. Upgrading by plugin means re-running the install and taking whatever main contains at that moment. Upgrading by git clone means pulling and re-copying, which overwrites any local edits you made to the copied commands in ~/.claude/commands/ and ~/.claude/skills/. If you plan to customise, keep your edits in a separate directory or track them, because the README offers no merge strategy.
The licence is MIT. In practical terms that permits use, modification and redistribution with the licence and copyright notice preserved, and it comes with no warranty. Because the manual install copies the files into your home directory, the licence text does not travel with them unless you keep it; if you redistribute a modified copy, keep the notice. That is the shape of the obligation, not legal advice, and if you are bundling this into something you ship, read the LICENSE file at the repository root yourself.
One cost the README does not price is the expertise skills. create-plans can load domain expertise from ~/.claude/skills/expertise/, and those knowledge bases are described as 5k to 10k or more lines generated with create-agent-skill. Generating one is cheap; verifying that a 10k-line knowledge base is still accurate is not, and nothing in the documentation describes a refresh path for it.
Editorial conclusion
Adopt it if you already run Claude Code daily and want staged meta-prompting, todo handoff, thinking-model commands and extension generators without writing them yourself. Skip it if you need a versioned, supported product with release notes: the README documents no rollback path, there are no retrieved releases, and the last push was 2026-04-01. Before installing, read commands/create-prompt.md and skills/create-plans/ and confirm that PLAN.md-as-prompt matches how you actually work.
Frequently asked questions
How do I install taches-cc-resources in Claude Code?
The README recommends the plugin route: run claude plugin marketplace add glittercowboy/taches-cc-resources, then claude plugin install taches-cc-resources, and start a new Claude Code session. Alternatively, clone the repository and copy commands/* into ~/.claude/commands/ and skills/* into ~/.claude/skills/.
What do I actually get in taches-cc-resources?
The README lists 27 slash commands, 9 skills and 3 subagents. The commands cover meta-prompting, todo management, context handoff, extension creation, auditing, thinking models and debugging; the skills generate plans, MCP servers, agent skills, meta-prompts, slash commands, subagents and hooks; the subagents are skill-auditor, slash-command-auditor and subagent-auditor.
Where does taches-cc-resources store prompts and todos?
The README states that project-specific data such as prompts and todos lives in each project's working directory, while the commands and skills themselves install globally to ~/.claude/commands/ and ~/.claude/skills/.
Is taches-cc-resources actively maintained?
The repository is not archived, but the last push was on 2026-04-01 and no releases were retrieved, so there is no published version to track. Treat it as a stable personal toolkit rather than something updated on a schedule.
What licence does taches-cc-resources use?
It is MIT licensed. That permits use, modification and redistribution provided the licence and copyright notice are preserved, and it comes with no warranty; the LICENSE file is at the repository root.
What is PLAN.md in the create-plans skill of taches-cc-resources?
The README describes create-plans as producing a plan where PLAN.md is the prompt rather than documentation that gets transformed later, following the sequence Brief, Roadmap, Research if needed, PLAN.md, Execute, SUMMARY.md. It can optionally load domain expertise from ~/.claude/skills/expertise/.
Community notes