Harness marketplace
Buy and sell the parts of a coding agent’s harness: skills, reviewers, memory files and playbooks that already work on somebody’s team. Each one is a packaged SKILL.md with reference files and a runnable test that drops straight into .claude/skills/ - installable via MCP, runnable as a CLI, and purchasable by humans or by agents paying in USDC over x402.
In the harness marketplace
24 of the currently published listings.
Arabic Dialect Engine — dialect-correct Arabic, not MT slop
Produces dialect-correct Arabic (Egyptian, Gulf, Levantine, Maghrebi, MSA) for any text — code-switching, RTL handling, register matching, and machin…
Agent Memory Architect — memory that resists poisoning
Design trust-tiered agent memory: short/long-term separation, write-time validation, provenance tracking, and defense against memory poisoning — the…
Incident Commander — structured response from detection to postmortem
Runs incident response end-to-end: SEV classification, war-room roles, rollback-first strategy, status-update cadence, timelines, blameless postmorte…
Secrets Leak Responder — rotate first, ask questions later
The runbook for 'I committed an API key'. Rotation order that minimises blast radius, git-history scrubbing, provider-specific playbooks for 13+ serv…
MCP Server Architect — build and harden MCP servers to spec
Design, review, and secure MCP servers against the current spec: stateless cores, OAuth 2.1 Resource Server, tool-poisoning defense, transport select…
AI Red-Team Auditor — continuous adversarial testing for agents
Continuous adversarial testing for LLM apps: prompt injection, jailbreaks, data leakage, tool abuse, memory poisoning, MCP tool poisoning — with CI g…
AI Code Security Reviewer — catch what the agent smuggled in
Security review built for agent-written code: hallucinated packages (slopsquatting), invented APIs, silently removed guardrails, insecure-by-default…
Prompt Injection Scanner — adversarial scan before you ship
Runs a 7-probe-family adversarial scan on any system prompt, SKILL.md, or agent config. Returns SAFE TO SHIP / SHIP WITH FIXES / DO NOT SHIP with the…
Git Timeline Film — play a repository's year
Any repo's history as a two-minute film: branches peel and rejoin, releases pinned, a daily heatmap on the same axis, and holds that zoom in on real…
Fine-Tuning GPT Models
Taking a pretrained GPT and adapting it to a new dataset without undoing what it already knows.
Preprocessing and Tokenizing Textual Datasets
Turning raw text into the tokenized, memory-mapped shards a GPT training loop can actually stream.
Training GPT Models with PyTorch
A training runbook for GPT models in PyTorch: torch.compile, DDP scaling, checkpoints that actually resume, and LR schedules tied to real steps.
Tools, Not Clicks: make any website agent-ready with WebMCP
Retrofit a site so AI agents call typed tools instead of screenshotting the page and guessing which rectangle is the button. The exact document.model…
Spinning up an AI employee: the Hermes agent stack playbook
Nine layers, one vendor each, and the 24-hour/7-day sequence that turns them into a $5k/month retainer. Hermes on an Orgo desktop, with its own email…
harness-shape: judge an agent framework by its imports
Three structural checks read straight from a repo's imports. It counts and names modules. It never scores.
concept-flow-film: any codebase or process as an animated flow
Point it at a GitHub repo or describe a process in words. Get a Remotion film. Nothing is drawn that cannot cite its source.
HyDE, Federated and Streaming RAG
The three retrieval patterns that fix what a better index cannot: the shape of the question, where the data is allowed to live, and what time it is.
Agent Driven Code Editing
A prompt, review and verify loop for changing code with an agent instead of editing by hand, with the checks that keep the diff honest.
Agent Context Files
Structure the files a long-running agent reads so its instructions stay current instead of going stale.
Skunky: Local-First Compartmentalized LLM Work
Hand bounded, sanitized work to untrusted LLM workers without exposing raw project files, identifiers, local paths, or secrets.
Branded ECharts Visualizations
Turn your brand's colours + fonts into polished, on-brand Apache ECharts charts — bar, line, pie, heatmap, radar, dashboards — rendered to SVG + PNG.…
Auditing a Page's Extractable Units
Inventories the liftable blocks on a page — direct answer, definition, list, table, Q&A — so you know how many ways AI can quote it.
Optimizing for People Also Ask
Maps the related follow-up questions a page should answer and flags the gaps, so you capture the whole question cluster.
Structuring Comparison Tables for Extraction
Validates that a comparison table has enough options, enough attributes, and no empty cells — the form AI lifts for 'X vs Y'.
How buying works
Free skills download after signup; paid skills are priced per listing. Sellers keep 80% of each sale, settled through Stripe Connect or in USDC via x402.