A skill that makes agent-written reports reproducible by other agents
- Source
- x.com
- Date
I send around PDFs and markdown files to my coworkers so much and they use their agents to make sense of the world too, so I started using this skill methodology-appendix to make sure their agents could replicate what my agents found. --- name: methodology-appendix description: "Every data-backed report or memo ends with a Methodology / Replication Appendix: the signal choice and its sensitivity, verbatim queries, derivation pseudocode over named constants, the rebuild command, and drift/limits. Appended by default, without asking." --- # Methodology / Replication Appendix ## Rule Any report, PDF or page whose conclusions rest on data queries ends with a **Replication Appendix**. It is a working recipe, not a citation list. The test: > A reader with an agent and equivalent data access can regenerate the document > end to end from the appendix alone, without contacting the author. If a reader would have to ask "which column did you use?", "what date was this pulled?" or "how do I rebuild the PDF?", the appendix failed. ## Body voice The body states conclusions plainly. No process narration ("I was asked...", "on the first pass..."), no self-grading. The first sentence…
Context
Garry Tan shared a , a file of standing instructions an applies when a task matches its description, that makes his agents end every data-backed report with a replication appendix. Tan says he often sends PDFs and Markdown files to coworkers who use their own agents to make sense of them, and he wants those agents to be able to reproduce what his found. The skill sets the test: a reader with an agent and equivalent data access should be able to regenerate the whole document from the appendix alone, without contacting the author.
The appendix works as a recipe rather than a list of sources. It records what access is needed and the date the data was pulled, then every query that produced a number, copied exactly as it ran, and the arithmetic behind derived figures, written with each assumption as a named constant a reader can change. It ends with the commands that rebuild the PDF and a section on drift and limits, covering what was still changing at snapshot time, gaps in the data and retracted claims. The skill calls one section the most important: the choice of signal, meaning the column or threshold an analysis turns on. The appendix must name the alternatives and include an outcome sensitivity line saying how a different defensible choice would change the headline, because a replication that uses another column will not match the numbers.
Two rules keep the appendix honest. Queries are logged verbatim as they run, since the skill says reconstructed SQL is how appendices go wrong, and the appendix is updated in the same commit as any revised finding. The report body, meanwhile, states conclusions plainly and leaves the process to the appendix. The skill tells the agent to add the appendix automatically, without asking, to query-backed memos, PDFs for partners, executives or outside stakeholders, and pages of quantitative findings. That default depends on the agent loading the skill, which is not guaranteed. In Tan's own harness evaluations, Claude Code opened a skill nobody mentioned 0 times in 30 turns, while a rule shown on every turn was followed far more often. Tan reports no results from using this skill.
- agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
Checking sign-in…
Loading comments…





