QUICK ANSWER

If someone already explained the change in a Loom recording, give your agent that link and ask it to use Clipy to work from the explanation. You do not need to write the whole request again. For the next handoff, record your screen with Clipy, say what you want done, and share the link.

Use Clipy to read this Loom recording and help me make the changes described: [link].

What does Claude Code need from a Loom recording?

For a useful feedback handoff, Claude needs the requested outcome, the relevant evidence, and access to the code it should change. The recording supplies evidence and narration. Your current instruction defines the work you want authorized.

Imagine a reviewer says: “After I change this to Weekly and press Save, it goes back to Daily. Keep the new choice.” A useful handoff preserves the words, identifies the control, and shows the transition. A transcript may be enough if the reviewer names the page and field. It may be insufficient if there are several unlabeled controls and every request relies on “this” or “that.”

Check four ingredients:

  • Words: Is there a usable transcript of what the reviewer said?
  • Time: Can each request be traced to the relevant point in the recording?
  • Visuals: Can the agent identify the target and inspect the observed state?
  • Scope: Does the agent know which repository, page, and requested changes to work on?

Do not assume that the original Loom link delivers all four. A browser player might be accessible to you while an automated import cannot access its transcript. A document might contain a title and duration without any narration. A successful download might still need media inspection before you can use it as evidence.

The goal is not to convert every recording into the largest possible bundle. It is to preserve enough information for the agent to understand and verify this request.

Which import path should you use?

Clipy has different paths for different jobs. They should not be presented as interchangeable.

Path Useful when What to expect
Downloader’s agent-context offer You already downloaded a public Loom and want its available words in your library Transcript and metadata; no frame extraction in this web path
CLI context import without sync You want a local context bundle Transcript-first output, with no synced visual classification
CLI context import with --sync You want an authenticated library document and possible visual enrichment Transcript upload, server classification, and selected local frame extraction when requested and available
Local-file context import You have an authorized exported recording and caption file Local bundle; requires a supplied transcript, with optional sync
Video-library migration You want to move recordings into another video library A separate migration decision, not a requirement for reading one recording’s feedback

Use a public share link only when the owner has authorized that sharing. An importer that cannot access a private link is not a reason to bypass the owner’s restrictions. An authorized export can provide a different route without changing the original recording’s visibility.

For one narrated feedback recording, start with its transcript and assess whether it identifies the targets. Add visual evidence where needed. For an interaction bug, expect that the sequence or key states may matter more than a summary.

How do you use the web offer after downloading a Loom?

Open the Loom downloader and use a recording you are allowed to download. Complete the download flow. The optional agent-context offer appears separately; the download does not depend on signing up for this extra step.

The offer is labeled “Want the agent-readable version too?” If you are signed out, it explains that saving the document requires a Clipy account. Use the displayed sign-in action, then open the saved document when it is ready. If a summary is still processing, the interface distinguishes that from the transcript already being available.

What matters is the content inside the document. This web path fetches Loom metadata and any readable Loom transcript. It does not download the media to extract images. Even if a classifier recognizes that the narration would benefit from pictures, that observation does not create pictures for this path.

When no readable transcript is available, the web flow can save a metadata-only document with the video’s details and a player. That is useful as a saved reference for a person, but it is not adequate visual evidence for Claude to understand a UI change. A title saying “Settings feedback” cannot tell the agent which setting should change.

For a transcript-rich explanation, read the saved words and check that they match the request. For feedback relying on visible targets, provide the missing images or use an appropriate visual import workflow. “Saved successfully” means the document was saved; it does not mean every implementation requirement has been captured.

How do you ask your agent to read a Loom recording?

Start with the recording link and the task. If the agent needs the Clipy skill or account access, ask it to handle setup and guide you through anything it needs. You can keep explaining what you want in ordinary language.

Use Clipy to read this Loom recording and list the requested changes before editing: [Loom link].

The agent handles the import. It may need an authorized download or transcript if it cannot access the source. A pasted private link does not grant it access to someone else’s account.

What if you want the context in your Clipy library?

Save this recording’s context to my Clipy library so I can refer to it later.

Follow the account connection prompt if needed. Ask the agent to confirm that the context was saved and whether visual evidence is available. Importing context does not automatically create a public watch recording or reproduce Loom’s original diagnostics.

How do you give the imported context to Claude Code?

For a local bundle, provide Claude Code the returned path and ask it to read the document in the repository session where you want the work done. The path must be accessible from that environment. A file on your laptop is not automatically available to a remote agent workspace.

For a synced document, use the authenticated Clipy connection described in the MCP documentation. Imported context documents are distinct from native recordings. The relevant tools include get_context_document for metadata and availability, and read_context_document for the content. Use the document identifier returned by the import rather than guessing a public recording URL.

A focused initial prompt can be:

Read the imported Loom context document with ID <document-id>.
This is feedback on the settings page in the current repository.

Before editing, list each requested change with its timestamp,
the visible target if established, and any missing evidence.
Separate the reviewer's requests from your implementation ideas.
Do not invent labels, dimensions, or root causes that the source
and repository do not establish.

For local context, replace the first line with the actual local bundle path. The remainder can stay the same.

Review the returned list. It should preserve the intended requests without promoting tentative suggestions into commitments. “Could we make this smaller?” and “Make the Save button smaller” carry different certainty. If the context does not identify what “this” means, answer that narrow question before the agent edits the wrong element.

Once the task is clear, ask the agent to implement the agreed changes and verify the specific behaviors. Anthropic’s common workflow documentation includes debugging, testing, and planning patterns; adapt those to your repository rather than treating a video handoff as a replacement for normal code inspection.

What would a reviewable Loom feedback brief look like?

The following is an illustrative example, not a claim about a tested import or customer outcome. Suppose a recording contains three requests on a notification-settings page:

Moment Narrated request Required evidence Open question
00:12 Keep Weekly selected after Save Before value, Save action, and reset state Does reopening the panel also reset it?
00:31 Put the helper text below the frequency control The named control and helper text Should mobile use the same placement?
00:49 Make the success message stay visible longer Current message and its disappearance What duration is intended?

A useful brief preserves those differences. The first request has a clear expected behavior. The second needs a responsive-layout decision if the recording does not answer it. The third is not permission to invent an exact duration and report it as the reviewer’s requirement.

You can turn the first row into a task:

Task: Preserve the selected notification frequency after saving.
Expected: Selecting Weekly and saving leaves Weekly visible.
Observed: The recording shows the control returning to Daily.
Evidence: Review the sequence around 00:12 in the source recording
and the corresponding context, if those moments were imported.
Unknown: Whether the saved backend value is also wrong.
Verification: Reproduce selection and Save; inspect the visible
value and confirm the persisted value using the app's normal flow.

Notice the separation between observed UI state and persisted data. A visible reset does not prove that the server stored the wrong value. The agent must inspect that part.

If the import contains only narration, do not fill the “Observed” field with a made-up visual result. Attribute it to the speaker and add the screenshot or sequence needed to confirm it. That distinction keeps a plausible summary from becoming false evidence.

What if the Loom is private or missing a transcript?

Ask the agent what it could access and what is missing. Provide an authorized local recording or transcript if you have one. It should not bypass access controls or claim to have inspected frames it could not retrieve.

How do you use a local recording instead?

Use Clipy to read this video file and extract the feedback relevant to our project: [file].

Give the agent access to the file. If you also have a transcript, include it. Let the agent handle the import and ask it to report any gaps.

How is this different from migrating a Loom library?

Context import answers: “How can my agent use the feedback in this recording?” Library migration answers: “Where will these videos be stored and played in the future?” You may want both, but one does not automatically accomplish the other.

A context document can contain source metadata, narration, selected visual evidence, and references to the original recording. It need not contain a complete independent copy of the media. If your goal is to stop depending on the original provider, inspect the migration product’s actual storage and playback behavior.

Builder’s Clips announcement describes a Loom import that downloads public media, rehosts it, and imports available transcripts. That is a different promise from Clipy’s transcript-first downloader offer. Both may help with agent context, but their handling of the original media matters if you are planning a migration.

For Clipy’s broader migration and product comparison, start with the Loom alternative page. For this guide’s task, keep the scope smaller: import the authorized recording’s usable context, verify its completeness, and give Claude a clear implementation request.

How do you verify that the handoff worked?

A successful command is only the first check. Ask Claude to identify a request from the beginning, middle, and end of the relevant feedback. Compare those items with the source. This can reveal a truncated transcript, incorrect timing, or a missing final correction before code changes begin.

Check visual references separately. If the reviewer says “this field,” ask the agent which field it identified and what evidence supports that identification. Do not accept a confident answer based only on a page title when several targets are possible.

Then make the acceptance criteria explicit. A brief such as “improve the settings page” is too broad to verify. “Keep Weekly selected after Save” creates an observable check. Preserve unknowns until the repository or the reviewer resolves them.

After implementation, use the same flow shown in the source. Review the resulting diff and the relevant behavior. The agent’s report should distinguish what it exercised from what it inferred. A demonstration of one passing interaction is useful evidence of that interaction, not proof that every browser, device, and edge case passes.

Keep the recording reference attached to the task while reviewing. It gives you a way to resolve disagreement about what was requested without depending on an increasingly compressed summary.

What are common questions about Loom recordings and Claude Code?

You can provide the link, but successful use depends on the agent’s tools, permissions, and access to the underlying content. Confirm whether it retrieved narration and relevant visuals. A response about the page title is not evidence that it understood the recording.

Does Clipy’s web offer extract frames from a Loom?

The web downloader’s context offer uses the narration and metadata. For visual evidence, ask your agent to use Clipy to read the video.

What if the Loom recording has no captions?

Give the agent the video link or an authorized local file and ask it to read it with Clipy. If you have a transcript, include that too. Ask it to save the context to your Clipy library only when you want it kept there; the agent should confirm what was read and saved.

What does importing a Loom give the agent?

Context import turns a Loom into a transcript and frames your agent can read. Record new bugs with Clipy to add cursor, console and network evidence.

Does a synced context document become a public Clipy recording?

No. Context import and native recording sharing are distinct paths. Use the returned document identifier and authenticated context-document access. Do not construct a /video/ URL from a context-document ID or change the source’s privacy settings merely to make it easier to fetch.

Where should you start with your own recording?

Give the agent one Loom recording you have permission to use. Ask it to read the video with Clipy, identify the requested change, and flag any evidence it could not retrieve. If you want it saved for later, ask the agent to add the context to your Clipy library.

The examples above are illustrative, not measured customer outcomes. Product behavior is described from the current documented import paths; external project descriptions remain their authors’ claims. Before asking Claude to edit, establish what the source actually supports, clarify the missing parts, and name the behavior you will use to judge the result.