Versioned like code
Every skill is a file: reviewed in a pull request, released under a tag, pinned by the consumer. Upgrading is a diff, not a rumour.
AI-DRIVEN SDLC
AI Marketplace
Your team’s expertise, packaged as agent skills your whole org runs the same way — versioned, reviewed, pinned like any dependency.
claude code
/plugin install aws-finops@binbash
- aws-finops-investigateskill
- aws-finops-optimizeskill
- billing-cost-managementmcp
enabled inclaude codecoworkci✓ pinned by tag
THE PROBLEM
Everyone is using AI. Nobody is using the same AI.
Your best engineer has a prompt that works. It lives in their terminal history, and it leaves when they do.
- The same question gets a different answer depending on who asks, and what they pasted in.
- Nothing is versioned. No diff, no review, no way to trace what produced last quarter’s report.
- Every repo reinvents its conventions, so what an agent learns in one carries to none of the rest.
- CI cannot run any of it. The knowledge is in people’s heads, not in something a pipeline installs.
WHAT IT IS
A registry, not a prompt library
One private catalog of plugins — skills, commands, subagents, hooks and MCP servers — installed from a pinned release and identical everywhere it lands.
Portable by standard
Open Agent Skills layout, tools reached over MCP — so the content travels to whatever agent your team runs next.
Packaged, not documented
Install once and the whole capability arrives configured — no runbook, no per-developer setup.
Everywhere at once
Checked into the repo’s settings, so every contributor and every CI run gets the same catalog. Nothing depends on who installed what.
THE CATALOG
What we have already built for ourselves
binbash runs every one of these in production — across its own engineering, delivery and go-to-market work.
11 plugins · 23 skills · one private registry
Cloud cost & FinOps
aws-finops
Diagnoses real AWS spend — deltas, anomalies, tag hygiene, forecast — then ranks the savings.
Cloud cost & FinOps
aws-cost-estimation
Prices a workload before it exists, from live AWS pricing. Describe it, get the monthly breakdown.
Security & architecture review
well-architected-iac-review
Reviews Terraform, CloudFormation, CDK or a diagram against the AWS Well-Architected Framework.
Security & architecture review
aws-security-eks-audit
Audits an EKS cluster against the CIS Benchmark plus AWS, NSA/CISA and Pod Security guidance.
Architecture diagrams
diagrams-as-code
House-style architecture diagrams generated from Python — reviewed in a PR, regenerated not redrawn.
Architecture diagrams
diagrams-drawio-mcp
Editable draw.io canvases driven over MCP, with AWS and Azure icon catalogs.
Architecture diagrams
diagrams-migration-mermaid
Current- and target-state diagrams read straight out of a migration assessment, in Mermaid.
Delivery & go-to-market
binbash-assessment-suite
Migration, FinOps and infrastructure assessments, each with the summary that goes to the client.
Delivery & go-to-market
validate-sow
Reads a statement of work and reports what the scope promises that no task actually covers.
Delivery & go-to-market
release-management
Bumps the right versions, writes the changelog, opens the release PR. A person still cuts the tag.
Delivery & go-to-market
sales-toolkit
Account research, pipeline review, weighted forecasting and competitive battlecards.
OPEN STANDARDS, CLAUDE-NATIVE
Portable content. Packaged for the tools your team already opens.
The knowledge is written once, in formats nobody owns. The packaging is where we lean into Claude — because that is where the ecosystem is.
Written once, in the open
Plain Markdown with YAML frontmatter, tools reached over the Model Context Protocol. Neither is vendor-specific, so the content outlives any one agent.
- Agent Skills — one SKILL.md per capability
- Model Context Protocol for every tool
- AGENTS.md so every agent reads one set of rules
Delivered as Claude Code plugins
A private marketplace your team adds once. Engineers get it in the terminal; everyone else gets the same skills in Claude Cowork, no command line.
- A private Claude Code plugin marketplace
- Claude Cowork for the people who never open a terminal
- Pinned by release tag, private by default
binbash is part of the Claude Partner Network, and builds on Anthropic’s Claude across engineering, delivery and go-to-market.
How we build yours
01 AUDIT_
Map what your teams already do with AI
Repo by repo: what agent config exists, which conventions each team invented for itself, and which workflows are repeatable enough to package first.
02 DESIGN_
Set the conventions once
Naming taxonomy, categories, authoring and release rules, one instructions file your whole toolchain reads — written before the first skill, so the tenth still fits.
03 BUILD_
Package the highest-leverage workflows
Each becomes a skill bundled with the commands, hooks and MCP servers it needs, validated strictly on every pull request.
04 ROLLOUT_
Pin it across every repository and pipeline
Checked into each consumer, pinned to a tag, working for CI as well as for people — with a production-readiness gate before done.
WE RUN IT OURSELVES
Not a slide. The way we work.
Strict validation on every pull request
Manifests, skill frontmatter and the symlinks that keep one copy of every skill body, all checked before merge.
An agent prepares the release. A person cuts the tag.
Bumps, changelog and release PR are automated. The tag stays human — it is the contract every consumer pins to.
Consumed by pinning, never by copying
Our own repos install at a pinned tag. Upgrading is a one-line diff in a reviewed PR; rolling back is the same line.
AI review on demand, not on autopilot
Invoked deliberately on the changes that warrant it, not billed against every push. The gate stays a human decision.
Ready to build your own AI marketplace?
Tell us what your team already does with AI and we will show you what it looks like packaged. Want a walkthrough of ours? Ask for access.