Reflection Engine: a single prompt that turns your AI assistant's memory into a candid portrait
A downloadable prompt that turns your AI assistant's memory of you into a candid, evidence-grounded portrait.
At a glance
- What is it?
- Reflection Engine is a downloadable markdown prompt, not a service. It asks an AI assistant that already holds years of your chat history to write back 22 evidence-cited answers about your patterns, and it only works if that memory is switched on.
- Who is it for?
- Reflection Engine suits people who have years of real conversation sitting in one assistant account, memory and chat history already enabled, and a willingness to read a blunt portrait without treating it as therapy. It is the wrong tool if your history is scattered across accounts, if your main assistant is a coding one, or if you want a repeatable measurement rather than a one-off read.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 46 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 18, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Reflection Engine actually is, and who it is for
Reflection Engine is a markdown file you attach to a conversation. The repository holds two top-level entries: README.md and Reflection-Engine-v1.3.md. There is no package, no binary, no server and no account. The README is explicit that it is "a prompt, not a product," with no install and no telemetry.
The problem it addresses is a mismatch between what an assistant knows and what it tells you. A model that has watched you discuss money, health, family and half-formed ideas over several years can summarise those conversations, but summarising is not the same as confronting. Reflection Engine is written to produce what the README calls "a candid, evidence-grounded portrait": blind spots, contradictions, expensive habits, and where the current trajectory leads. It returns 22 direct answers, each citing evidence, scoring its own confidence, and closing with one concrete thing to try that week.
The intended user is someone with a long, personal history inside a single assistant account. The README warns against using your coding assistant, on the grounds that "a corpus of pull requests produces a portrait of a developer, not of you." That is a narrower audience than the framing suggests. If your conversations are spread across three accounts, or you clear history regularly, the corpus is thin and the portrait will be too.
How the prompt gets its evidence, and why the model choice matters
The mechanism is retrieval by the host assistant, not by the prompt. Reflection Engine does not read your files or query an API. It instructs the model to look across what that assistant already stores: chat history, memory entries, and uploaded files. Everything depends on the provider's own memory features being active.
The README names the specific settings to check: memory and "reference chat history" in ChatGPT, memory and past-chat search in Claude, personal context in Gemini. With those off, the README states the model works from "a single blank conversation and the whole exercise falls apart." That is the single largest failure mode, and it is silent. You can attach the file, send the instruction, and receive a confident-sounding portrait built on nothing.
The second dependency is model capability. The README asks for the newest model with reasoning set as high as it goes, because the task requires holding evidence across years of material and weighing counterevidence against it. Its own warning is blunt: "A fast, cheap model will hand you a horoscope." There is no way to verify from the outside whether the model actually weighed counterevidence, which means the confidence scores in the output are self-reported and should be read as such.
One design detail is worth noting. The prompt tells the model that a third party wrote it, so the model does not mistake the questions for your own words or treat your curiosity as evidence about you. The README therefore instructs you not to add your name to the file.
Installing Reflection Engine and running a first session
There is nothing to install. The README's steps are a download and a conversation. Start by getting the file from the repository.
git clone https://github.com/kropdx/reflection-engine.git
cd reflection-engine
lsThe listing should show README.md and Reflection-Engine-v1.3.md. If you prefer not to clone, the README points at the file directly as Reflection-Engine-v1.3.md.
Next, open the assistant account where you actually talk, not a coding assistant. Confirm that memory and chat-history referencing are enabled in that provider's settings. Then start a fresh conversation, attach the markdown file, and send the instruction the README gives:
Please evaluate the attached markdown file and complete all tasks.Expect a long run. The README says good output takes a while, and that the result is 22 answers, each with cited evidence, a confidence score, and one concrete action for the week. Read the citations before the conclusions: they are the part you can check against your own recollection, and they tell you whether the model actually had your history or was improvising.
Where Reflection Engine breaks down
The most serious limitation is that the prompt cannot verify its own inputs. If memory is off, or the assistant's history search misses older conversations, the output still arrives in the same confident format. Nothing in the README describes a check that would catch this, and there is no error state to look for. The reader has to judge from the citations whether the corpus was real.
The second limitation is the corpus requirement itself. The README is direct that the tool "is only as good as the corpus it can see." A new account, a periodically cleared history, or a habit of using several assistants for different parts of life all reduce the input below what the task needs. The 22-answer format does not scale down; you get the same number of answers from thinner evidence.
The third is the nature of the output. The README states plainly that Reflection Engine "isn't therapy, and it isn't a diagnosis." It is a single read from a single model at a single moment. Run it twice and you may get different emphases, because there is no scoring rubric, no baseline and no versioning of results. Treating it as a measurement rather than a provocation misreads what it is.
Finally, the output is sensitive by design. The README cautions that it is "a blunt read on you, drawn from your most personal conversations," and advises keeping it private and thinking twice before pasting it into a shared workspace or team chat.
Reflection Engine compared with a plain summary request
The obvious alternative is asking your assistant directly: summarise what you have been working on, or tell me what patterns you notice. That costs nothing and requires no download.
The difference is structure and adversarial framing. A summary request invites the model to be agreeable; it has no reason to surface contradictions or to attach a confidence score to each claim. Reflection Engine's 22 answers each carry cited evidence and a self-assessed confidence level, and the set is chosen to include uncomfortable categories such as blind spots and expensive habits. The README's stated goal is to be "uncomfortable in a useful way, not cruel."
Whether that framing produces better output than a well-worded direct question is not something the repository demonstrates. There are no sample outputs in the repository, no comparison against a baseline prompt, and no evaluation. The case for Reflection Engine rests on the care taken in the prompt's construction and on the citation requirement, not on measured results. If you already get useful self-reflection from open-ended questions to your assistant, the marginal gain here is smaller than the README implies.
Maintenance, licensing and what the repository does not say
The last push to the repository was on 2026-08-03, roughly six weeks before this writing, and the repository is not archived. The project has no releases, so version 1.3 of the prompt exists only as a file in the tree rather than as a tagged artifact. Upgrading means replacing Reflection-Engine-v1.3.md with whatever file a later commit adds, and re-running the session; there is no changelog describing what changed between prompt versions.
Running cost is whatever your assistant provider charges for a long, high-reasoning conversation. The README does not estimate token usage, and the prompt's demand for maximum reasoning effort means the run is not cheap relative to an ordinary chat.
The license is not stated in the repository. The README attributes the work to Kevin Rose and links to X and Instagram profiles, but it does not specify terms for redistribution or modification. If you plan to fork the prompt, adapt it, or ship it inside a product, that is the first thing to resolve, and it is a question for the author rather than something the README answers. Nothing here constitutes legal advice.
On privacy, the README's claim is narrow and worth reading precisely: your data does not leave the AI provider you chose, and no server operated by the prompt's author is in the loop. That is a statement about the author's infrastructure, not about the provider's handling of your history.
Editorial conclusion
Reflection Engine suits people who have years of real conversation sitting in one assistant account, memory and chat history already enabled, and a willingness to read a blunt portrait without treating it as therapy. It is the wrong tool if your history is scattered across accounts, if your main assistant is a coding one, or if you want a repeatable measurement rather than a one-off read. Before you start, verify three things: that the assistant you plan to use has memory and chat-history referencing switched on, that you are attaching Reflection-Engine-v1.3.md to a fresh conversation, and that you have a private place to store the output, because the README states the result is sensitive and should not be pasted into shared workspaces.
Frequently asked questions
What is reflection in software engineering?
Reflection Engine is unrelated to the software-engineering sense of reflection, where a program inspects its own structure at runtime. This project is a markdown prompt that asks an AI assistant to inspect what it already knows about you and write back 22 evidence-cited answers.
Does Reflection Engine send my data anywhere?
The README states there is no service, no account, no install and no telemetry, and that nothing is sent back to the author of the prompt because there is no server in the loop. Your data stays with the AI provider you already use.
Which AI assistant should I use with Reflection Engine?
The README says to open the account where you actually talk, not your coding assistant, and to pick the newest model with reasoning set as high as it goes. It also names the memory settings to enable: memory and reference chat history in ChatGPT, memory and past-chat search in Claude, personal context in Gemini.
How many answers does Reflection Engine produce?
The README states you get 22 direct answers, each citing the evidence behind it, scoring its own confidence, and ending with one concrete thing you can try that week.
Community notes