PairFoundry is where that context gets built: repo configuration, chained prompts and review passes for Claude Code, Codex and the rest. Written from real projects, versioned as the tools change, and explicit about where each one stops working.
One-time purchase · Yours permanently · 14-day refund, no questions
# CLAUDE.md
## What this repo is
Payments service. Go 1.23, Postgres, deployed on Fly.
Money is involved: never guess at a rounding rule — ask.
## Before you change anything
- `make check` must pass. It runs vet, lint and the race detector.
- Migrations live in db/migrations and are append-only. Never edit one
that has shipped; write a new one.
- internal/ledger is double-entry. Every write needs a matching
counter-entry in the same transaction, or the invariant test fails.
## How to work here
1. Read the failing test before you read the source.
2. Change one thing, run `make check`, then continue.
3. If a change touches internal/ledger or db/migrations, stop and
summarise the plan before writing code.
## Do not
- Do not add dependencies without asking.
- Do not touch pkg/generated — it is regenerated from proto.
- Do not "fix" a flaky test by adding a retry.The gap
Every one of these is fixable, and none of them is fixed by a better model. They are fixed by what you put in the repo and what you ask for.
Nothing tells the agent that the ledger is double-entry, that migrations are append-only, or that pkg/generated is regenerated. So it writes code that compiles and violates the thing that actually matters.
Given a bug report, an agent will usually produce a confident theory and a patch for it. Ask it to reproduce first and the diagnosis changes about a third of the time.
The bottleneck moved. Most of the day is spent reading diffs you did not type, looking for the one wrong assumption buried in four hundred correct lines.
How the packs fit together
Each pack solves one stage well. They are designed to sit next to each other, but any one of them works on its own.
Memory files, skills, slash commands, hooks and a permission policy that fits how your repo actually works.
Open the packA staged pipeline from one-line request to reviewed PR, with an explicit gate between every stage.
Open the packSingle-purpose review passes, a debugging protocol that reproduces first, and refactors with a blast radius.
Open the packPrompt & skill packs
Every pack is a folder of working files — memory templates, skill definitions, slash commands, hooks, prompt chains. You read them in your browser and download the whole thing as a zip.
Everything that goes in the repo before you type a single prompt.
The chained prompts that take a one-line idea to a reviewed pull request.
The prompt arsenal for the unglamorous half: review, debug, refactor, backfill.
All three packs. Setup, pipeline, and review — the whole loop. All three, $89 instead of $117.
Learn
Nine lessons that get an agent set up properly on a codebase you did not write. No email required, nothing gated behind a signup form — read it and decide afterwards whether the paid material is worth anything to you.
The job does not become "writing prompts". It becomes specifying, bounding and reviewing — and those three are where all the difficulty moved.
CLAUDE.md, AGENTS.md and their cousins are the highest-leverage thirty minutes you will spend. Here is what belongs in one and what does not.
The difference between a useful agent and a frustrating one is usually the size of the change you asked for. Four techniques for keeping a diff reviewable.
The whole first track is free, permanently.
Tool directory
A maintained directory of AI coding tools with the parts that matter: what it runs on, what it costs, what it is genuinely good at, and where it falls down. We say when we haven't used something.
Anthropic's agentic coding tool. Lives in the terminal, reads and edits your repo directly, and takes instructions from a CLAUDE.md file you write.
From $20/mo (Pro) · usage-based API also availableOpenAI's open-source terminal agent. Runs locally in a sandbox, with cloud delegation for longer tasks.
Included with ChatGPT Plus/Pro/Business · or API keyA VS Code fork built around AI. Inline completion, chat over your codebase, and a background agent — all in one editor.
Free tier · Pro $20/mo · higher tiers for heavy useThe incumbent. Completion, chat and an agent mode across VS Code, JetBrains, Visual Studio and github.com.
Free tier · Pro from $10/mo · Business & Enterprise seatsOpen-source pair programming in your terminal. Git-native: every change lands as a commit.
Free and open source · you pay your own model costsGoogle's open-source terminal agent, with a notably generous free tier for individual developers.
Free tier with a personal Google account · paid for higher limitsWhere to start
Nothing here is gated behind a mailing list. The tutorials are open, the pack contents are described in full before you pay, and every purchase has a 14-day refund.
