cosmtrek/mindwalk is a local visualization tool that replays coding-agent sessions as movement across a three-dimensional map of a codebase. It helps developers inspect which parts of a repository an agent explored, read, or changed, and catalogue add-ons connect the viewer with supported agent session workflows.
Borrowing it
Nothing to install: this file belongs to cosmtrek/mindwalk. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/cosmtrek/mindwalk/master/.claude/skills/verify/SKILL.mdgit clone --depth 1 https://github.com/cosmtrek/mindwalkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/cosmtrek/mindwalk/verify)<a href="https://agentmods.dev/skills/cosmtrek/mindwalk/verify"><img src="https://agentmods.dev/badge/skills/cosmtrek/mindwalk/verify/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/cosmtrek/mindwalk/verify"><img src="https://agentmods.dev/badge/skills/cosmtrek/mindwalk/verify.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Rogue Agent · line 19 Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.Fix: Remove any persistence mechanisms (cron jobs, startup scripts, state files). Skills should not maintain state across sessions without explicit user consent.
- medium Server-Side Request Forgery · line 41 Code issues a request to a loopback, link-local, or private-range host. This can reach internal services not meant to be exposed and is a common SSRF pivot.Fix: Avoid requests to loopback/link-local/private hosts from skill code. If internal access is intended, document it and validate the target against an allowlist.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00019 | $0.00969 |
| Opus 5.5 | $0.00008 | $0.00388 |
| Sonnet 5.5 | $0.00004 | $0.00194 |
| Haiku 4.5 | $0.00002 | $0.00097 |
Grade A, and why
verify scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 74 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verifying mindwalk
Build & launch
npm --prefix web run build # tsc -b && vite build → web/dist
go build -o bin/mindwalk ./cmd/mindwalk
bin/mindwalk serve --no-open --dev --port <PORT> # --dev serves web/dist from the working tree
Gotchas:
- Always pass
--no-openduring automated verification so repeated server starts do not create tabs or steal focus from the user's current work. - Port 8765 is the vite-proxy convention and is often already taken by a dev
server; pick another port and check the log for
bind: address already in use. - Sessions come from
~/.claude/projects,~/.codex/sessions, and~/.pi/agent/sessions— this machine has real data, no fixtures needed.testdata/claude-session.jsonlworks viamindwalk open. bin/mindwalk map <repo>(or the/?map=1&repo=<path>URL) serves the static citymap with no session.
Drive (headless Chrome + CDP, no npm installs)
System Chrome + raw CDP over Node's built-in WebSocket works. WebGL renders
under plain --headless=new on the real GPU (Metal, ~120fps); add
--use-angle=swiftshader --enable-unsafe-swiftshader only if GPU init fails
(software rendering is ~3x slower):
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" \
--headless=new --remote-debugging-port=9333 --user-data-dir=<tmp> \
--window-size=1440,900 --no-first-run about:blank &
New-tab endpoint needs PUT: fetch('http://127.0.0.1:9333/json/new?<url>', {method:'PUT'}).
Flows worth driving
- Load via
/?session=<key>deep link (legacy bare session id resolves only if unique); assert.session-row.active .session-title. - Readout
.deck-pos-countshowsN / Nafter load (playhead starts at end). [aria-label="Restart playback"]→1 / N;[aria-label="Play playback"]→ ~3 ticks/second at 1× (the.speed-btn/Skey cycles the multiplier); playback draws the ember trail + firefly.- Scene view lives in the dock's scene section: the top strip icon (tree/
mountain shape mirrors the current view) opens a compact pop with
Tree/Terrainradio rows plus the encoding legend; pressingVcycles views directly. Switching rebuilds the scene — watch forRuntime.exceptionThrown. The pop coexists with an open sheet (report stays open while switching). /?map=1&repo=<abs-path>renders the citymap with no trace and no transport (map-only mode); the rail-head folder icon ([aria-label="Open a repository map"]) opens a popover — primary card for the active session's repo (name + path), then an "or open any repository" path input — and opens the map viawindow.open, so in headless drive the URL directly instead of clicking.- The right-edge dock strip is a panel registry with two sections (scene /
session, hairline divider): View (pop),
Crosshair= Inspect (sheet; file details, teaching empty state when nothing is selected),Sparkles= Evaluate (sheet; judge report, status dot mirrors running/done/stale/failed). Clicking a report finding jumps the playhead to its evidence seq and selects the file; the rail rows mirror evaluation state with.rail-evalbadges. [aria-label="Export video"]records playback client-side (MediaRecorder → webm download): the label flips toRecording video, the transport, rail, and view toggle lock while recording, and the playhead restores afterwards. Capture the download in CDP withBrowser.setDownloadBehavior.- Rapid session switch (click uncached row, 150ms later a cached row) must end showing the last-clicked session's data.
- Bogus
?session=must fall back to the newest session with a console.warn.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2mo ago First seen · 74 lines · 19 tokens per session scan A 465299423795
verify is a skill published in the GitHub repository cosmtrek/mindwalk (1,387 stars, last pushed 2mo ago), licensed MIT. It adds 19 tokens to every session and 969 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
browse
Drive a real browser through Aside: open a page, read it, click through a flow, take screenshots, check console errors. (gstack).
cua-driver
Use Cua Driver for desktop or browser tasks that are awkward or unavailable through Bash/APIs, or when the user explicitly wants GUI interaction: app testing, visual bug reproduction, form filling, calendar entry, screenshots, and demo recording. Also covers Cua setup; not OpenAI Codex Computer Use or web research.
create-verification-skill
Create a repo-local verification skill and exhaustive feature map for driving a real app through its UI, CLI, or API. Use for "create a verification skill", "make a verify skill for this repo", or "document how agents can verify this app".
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
browser-qa
A browser-based quality check for deployed web pages and user flows. It uses browser automation to test rendering, navigation, forms, interactions, responsive behaviour, and accessibility-related issues.
test-electron-app
Drive the real running PostHog Electron app (live tRPC, workspace-server, real data) over CDP with agent-browser. Connect to the running app on port 9222, test desktop changes against a local Django stack, snapshot the accessibility tree, inspect network requests, and screenshot only when explicitly asked. Use when…