Skill Packs
Read this part first: these are not on sale. There is no checkout, no waitlist and no date. The three packs below are finished and tested, but they have not been listed anywhere you can buy them, and we are not going to pretend otherwise to collect your attention. There is nothing on this page to click. When that changes, this page will change with it.
What follows is a description, so that you know what exists.
A skill pack is a directory of instructions in the SKILL.md format — the thing a coding agent loads when it needs to know how to do a specific job properly. Each of these three encodes a procedure we run ourselves, together with the reference material and the scripts that make the procedure checkable rather than aspirational.
flaky-test-triage — $29
Find out which tests are actually flaky, prove it, then fix or quarantine them with a paper trail — instead of hitting rerun.
Loads when tests fail intermittently, when CI is red for no apparent reason, or when someone is about to retry a build and hope. It measures flake rate across historical runs before anyone touches a test, classifies the cause, reproduces the failure deterministically, and then either fixes it or quarantines it with an expiry date so the quarantine does not become permanent.
Includes a flake-rate script that reads JUnit XML from a batch of CI runs and ranks tests by what they are costing you, reference material on the common causes, a measurement guide, and a quarantine protocol.
agent-delegation-kit — $39
Turn a vague task into a delegation brief whose subagents return usable work — or tell you to do it yourself.
Loads when work is about to be handed to subagents, parallel agents, or a multi-agent workflow. The first thing it decides is whether delegating is cheaper than doing the work directly, which is the step most multi-agent setups skip. After that it covers writing the brief, choosing the fan-out shape, and verifying what comes back.
Includes a brief linter that scores a delegation brief against the failure modes we have actually hit, a brief template, documented fan-out patterns, and before/after examples of a weak brief being rewritten into a strong one.
mcp-server-production — $49
Take an MCP server from "works on my machine" to one a model uses correctly — tool surface, schemas, errors, auth.
Loads when building, reviewing or hardening an MCP server, and specifically when the symptom is that the agent picks the wrong tool, ignores your tools entirely, or drowns in oversized responses. Covers designing the tool surface, writing tool and parameter descriptions a model can act on, shaping responses to a token budget, designing errors a model can recover from, and handling auth.
Includes a tool-surface linter that checks a tools/list payload against the design rules, a conformance checklist, and reference material on tool design, error design and response shaping.
How this relates to the articles
Scoping an MCP Server's Permissions Before You Connect It is the permission half of the same problem mcp-server-production addresses from the design side. What an AI Agent Run Actually Costs builds the cost model that makes agent-delegation-kit's "should you delegate at all" question answerable with numbers.
The articles are free and complete on their own. They are not teasers with the useful part removed.