agentleFS
TabbyMLTabbyML/pochi/.agents/skills/record-vscode-extension/SKILL.md

record-vscode-extension

Skill127 starsChanged 4 days agoApache-2.0View on GitHub

Record a run of the local Pochi VS Code extension as one shareable Jam link (video of the window, step marks and ✅/❌ checks as timestamped comments, a description with verdict, summary, and git state, and a console that holds the Pochi webview console and Pochi.log). Use together with run-vscode-extension when a run in packages/vscode or packages/vscode-webui should be captured and shared, e.g. a demo video for a pull request, evidence for a regression test document, or a reproduction for a bug report.

  • Pipes a download into a shell
  • Commits and pushes

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Pull request: When you create or update a PR for a UI or extension change, record the changed behavior and put the link in the PR body (a …

    line 29
  2. Distinctive:

    After pushing more changes that alter the behavior, record again and replace the link.

    line 39
  3. Very specific:

    Don't record for quick checks that only you need.

    line 57
  4. Very specific:

    Before cleanup, stop or cancel the recording.

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Record a Pochi VS Code extension run as a Jam

This skill turns a run into one Jam link: a video of the window, timestamped comments, a description, and logs. It does not start VS Code or drive the UI. That is the run-vscode-extension skill: start the instance with its env.sh up first, then drive the UI with its webview.mjs as usual while this skill records.

record.mjs lives in .agents/skills/record-vscode-extension/scripts/. Run it with node, and pass the same --port as the instance (default 9233). The examples use these functions:

r() { node .agents/skills/record-vscode-extension/scripts/record.mjs "$@"; }
w() { node .agents/skills/run-vscode-extension/scripts/webview.mjs "$@"; }

When to use

Record when someone other than you needs to see the run. Wherever the link goes, show the outcome next to it, for example [▶ Model menu: select a model](https://jam.dev/c/...) ✅ PASS, 3/3 checks.

  • Pull request. When you create or update a PR for a UI or extension change, record the changed behavior and put the link in the PR body (a "Demo" or "Test plan" section), so reviewers can watch it instead of checking out the branch.
    • New PR: include the link in the gh pr create --body. Then link the Jam back with jam update jam <id> --resource <pr-url> --label PR.
    • Existing PR: pass --resource <pr-url> --resource-label PR to stop. Then append the link to the body: gh pr view <n> --json body -q .body > /tmp/body.md, edit the file, and run gh pr edit <n> --body-file /tmp/body.md.
    • After pushing more changes that alter the behavior, record again and replace the link.
  • Regression testing. For a test document or checklist, record each case (or each flow) as its own Jam. Use the case name as the title, and add a check for each expected result. In the document, next to the case, write the link, the verdict, and the commit. Group a run with jam update jam <id> --folder "Regression <version or date>". A failed case's Jam is already the evidence for its bug report.
  • Bug report. Record the reproduction with a check that shows the wrong behavior, so the Jam ends as ❌ FAIL. Put the link in the issue (see the create-issue skill), and attach the issue to the Jam with --resource <issue-url>. After the fix, record the same steps again and link the two: jam update jam <fix-jam> --mark-related <bug-jam> --type fixes.
  • Review or demo. To show how a change behaves, including someone else's PR you are reviewing, record before and after (e.g. main and the PR branch). Link the two Jams with --mark-related ... --type relates_to.

Don't record for quick checks that only you need; run-vscode-extension is enough there.

Setup (once per machine)

Install the Jam CLI with curl -fsSL https://native.jam.dev/install | bash. It lands in ~/.local/bin/jam; JAM_BIN overrides the path. Then ask the user to run jam auth login, which opens a browser. Without the CLI, start fails with this hint.

Flow

r start "Model menu: select a model"           # Jam title; recorder runs in the background
r mark "1. Open the model menu"                # step marker, becomes a comment
w click label=model-select
r check "model menu opens" -- wait role=menu   # ✅/❌ comment; exits 1 on failure
w click 'role=menuitemradio[name="google/gemini-3.5-flash-lite"]'
r check "model is selected" -- wait '[aria-label="model-select"] >> text=gemini-3.5-flash-lite'
r check "menu closed" --expect 0 -- count role=menu
r check "icon renders crisply" --pass --note "screenshot /tmp/icon.png reviewed"
r stop --summary-file /tmp/summary.md \
  --resource https://github.com/TabbyML/pochi/pull/123 --resource-label PR
# ✓ Jam: https://jam.dev/c/...   <- give this link to the user
  • mark <label> puts a comment at this point of the video. Use it to label the steps.
  • check <name> -- <webview.mjs args> runs that webview.mjs command (wait, count, text, eval, ...; --timeout is in seconds and goes after --) and adds a ✅/❌ comment. It passes when the command exits 0. With --expect <output>, the command's trimmed stdout must also equal <output>. Scope locators so they cannot match your own prompt text.
  • check <name> --pass|--fail [--note <text>] records an outcome observed another way: reading a screenshot, a file, or a log.
  • check also works without a recording. It then only prints the result and sets the exit code.
  • stop uploads the recording and prints the link. --verdict pass|fail defaults to fail if any check failed, otherwise pass if there were checks. Use --summary <md> or --summary-file <file> to describe what the recording shows. Use bullet lists, because Jam drops tables. The description also gets the check list, the git branch, commit, and uncommitted diff stat, and whether the webview ran from Vite (dev) or the prod bundle.
  • cancel discards the recording; status shows it. One recording per port.

What the Jam contains

  • Video: start --video screencast (default) captures the workbench page through CDP, webviews included, even while the window is covered. It needs no permissions but misses native menus and dialogs. --video screen records the window itself and needs macOS Screen Recording permission for the app running the command.
  • Comments: each mark and check at its point in the video. A failed check includes the command and its error output.
  • Console: [mark] and [check ✓/✗] lines, [webview] Pochi webview console and uncaught errors (sidebar and task tabs), and [ext] Pochi.log lines, next to the workbench's own console. Read it back with jam get console <id> [--level error]. Read the video with jam get frames <id> --overview (right after upload it may report STILL_PROCESSING; retry later).

Notes

  • The recorder is a detached process with its own CDP connection, so it does not disturb the webview.mjs serve daemon and needs no webview reload. Its output goes to /tmp/pochi-vscode-debug-<port>/record.log (removed by env.sh down unless you pass --keep-state).
  • Reload Webviews, Reload Window, and restarting the serve daemon are fine mid-recording. If VS Code exits, the recording is cancelled and the recorder exits.
  • Before cleanup, stop or cancel the recording. env.sh down cancels a forgotten one.
  • Labels starting with 1. are rewritten to [1], because Jam renders them as a markdown list.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.