- Home
- Skills
- Agents & Orchestration
- vir: Recall Your Past Coding Decisions
More screenshots
Works with the AI Agents you already use
vir: Recall Your Past Coding Decisions
Your agent checks your own past decisions, gotchas and patterns before it starts work, and cites the notes it uses.
Free
vir: Recall Your Past Coding Decisions
Example session with this skill installed
Stripe webhooks are failing signature verification again on the billing service. Have I hit this before? Check my notes before you change anything.
- Read your context and instructions
- Compiled the vir
I checked your vir notes before touching the code.
Search: vir query "stripe webhook signature verification failing" --json --limit 5, 3 matches above the relevance cutoff. I read the top two.
| Note | Type | Project | Date |
|---|---|---|---|
| [[stripe-webhook-raw-body-7c1e9a2f]] | gotcha | billing | 2026-06-14 |
| [[express-json-middleware-order-3b8d0e41]] | decision | billing | 2026-06-14 |
| [[stripe-cli-local-forwarding-9f2a6c13]] | tool | billing | 2026-05-02 |
What you learned last time ([[stripe-webhook-raw-body-7c1e9a2f]]): signature checks failed because express.json() parsed the body before stripe.webhooks.constructEvent saw it. Stripe signs the raw bytes, so any re-serialized body fails. You fixed it by mounting express.raw({ type: "application/json" }) on the webhook route only.
What you decided ([[express-json-middleware-order-3b8d0e41]]): register the webhook route before the global express.json(), not by excluding paths inside the JSON middleware. That way a new route can't silently break it.
Checked against the current code: src/server.ts now calls app.use(express.json()) on line 18, above the webhook route on line 31. Someone reordered it, so the bug you fixed in June is back.
Next steps
- Move the webhook route above
express.json(), per your June decision. Waiting for your go-ahead. - Re-test with
stripe listen --forward-to localhost:4000/webhooks/stripe([[stripe-cli-local-forwarding-9f2a6c13]]). - Optional: add a test that posts a signed payload, so the middleware order can't regress again.
I found no notes about webhook secret rotation, so I can't rule that out from your history.
Connects securely to your tools. The creator never sees your data.
About this skill
Your coding agent forgets everything between sessions. You end up re-solving bugs you already fixed and re-opening decisions you already made. This skill lets the agent check your own history first: notes that vir distills from your past Claude Code and Codex sessions.
What it does
- Checks before non-trivial work. It searches your notes for decisions, gotchas, patterns and tools related to the task.
- Cites its sources. Each note it relies on is linked by name, e.g.
[[stripe-webhook-raw-body-7c1e9a2f]], so you can open it. - Prefers your past choices over a fresh default, and says when it's doing so.
- Verifies against current code. Notes can be out of date, so when the code disagrees it trusts the code and points out the conflict.
- Is honest about misses. "No vir notes on this" instead of invented citations.
Safe by design
- Read-only: it never writes, moves or deletes notes.
- It never runs anything that spends money or changes state unless you ask.
- Searching is free:
vir query --jsonis retrieval only, with no LLM call. - If vir isn't installed, it says so once and gets out of the way.
Requirements
This skill reads notes made by the vir CLI (free, open source, MIT):
npm install -g @djolex999/vir-cli
vir init
vir run
Needs the vir CLI and a vault it has already built. Building the vault means vir distilling your past sessions with an LLM (API key, Claude Code subscription, or ChatGPT plan via Codex); the skill itself never spends anything. It's only as good as your history: a new install with few sessions has little to find. Notes can go stale, so the skill checks them against your current code and flags conflicts. Search is keyword-based unless you set up local embeddings (vir embed --setup). vir runs on macOS and Linux only; Windows isn't supported yet.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
This skill has been used 2 times.
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Permissions
File Scopes
Shell: runs `vir query --json`, a local search with no LLM call and no network access. Read files: if the vir CLI isn't available, it reads ~/.vir/config.json to find your notes folder and searches the markdown notes there directly. It never writes, moves or deletes files, and never runs commands that spend money (vir run, vir connect, etc.) unless you ask. The vir CLI itself logs each search to ~/.vir/queries.jsonl; turn that off with "logQueries": false.