veyrul-media

Veyrul media knowledge for Hermes

Give a media-capable Hermes agent saved workflows, code and references to use
with its existing editing and production tools.

Useful questions include:

• How do word timestamps survive repeated clips and speed changes?
• How do I cut to music at a fractional frame rate without accumulating drift?
• Which audio track can I use from a mixed AAC/APAC recording while keeping the original?
• What local speech-cleanup guidance or image-removal repositories already fit my constraints?

Browse concrete tasks and reusable resources.
Read the same task index as JSON.

Use it alongside Hermes's existing
Manim video production skill:
look up narration cleanup or exact music timing before the audio and stitching
steps, then let the existing production tools do the work. Caption retiming
also fits a later editing pass. This package supplies knowledge and references;
it does not modify the production skill or install Manim or FFmpeg.

Install and enable

In Hermes 0.22.7 or later, install this subdirectory from its public source:

hermes plugins install the-media-vampire/veyrul-integrations/hermes-plugin --no-enable
hermes plugins enable veyrul-media

Start a new conversation so its skills and connection become available.
For a reproducible installation, append --ref <full-40-character-commit> to
the install command, using the published commit you reviewed. The catalogue
submission pins an exact commit; a submission is not a reviewed listing.

Try: "Use the connected Veyrul get_context tool to find reusable code for
keeping word captions aligned after repeated clips and speed changes. Read its
limitations before applying it to my edit."

The package offers one skill, veyrul-media:media-knowledge, and an anonymous
Streamable HTTP connection named veyrul_media at
https://veyrul.com/mcp/read. Its five public read tools are get_context,
get_record, search, list_records, and get_changes. Check the actual
tool list after connecting. A pre-existing Hermes server with the same name
takes precedence; this package does not replace it.

The skill is optional. Hermes plugin skills are explicit loads; direct MCP reads
do not depend on loading it. To make this part of an ongoing
project, the operator can add the following to that project's existing agent
instructions:

When video, animation, captions, audio cleanup or media-tool decisions need
outside knowledge, call the connected Veyrul get_context read tool directly
before repeating research; no skill load is required. Use a short non-private query with the
relevant tool/version constraints. Reuse applicable saved resources in our
existing production tools; preserve evidence limits and check current record
revisions when returning to related work. Skip lookups when the answer is
already supported by current project evidence. If the service is unavailable
or its results do not fit, continue with primary sources. Do not upload assets
or post contributions without separate permission.

If the local skill loader is unavailable or awaiting approval, use the connected
veyrul_media.get_context read tool directly with the same short, non-private
query and evidence checks. Do not let skill-loading friction stall the media
task. In an interactive client, approve only the local read you intend to allow.

This is an operator-selected project rule, not an automatic installation change
or a guarantee that every agent will select the service. The plugin itself
does not edit project instructions or start a recurring task.

Privacy, cost and permissions

Enabling the connection allows HTTPS requests to Veyrul. Tool calls send the
chosen public search terms or record IDs, plus ordinary network metadata.
Keep private briefs, transcripts, assets and credentials out of queries.
Veyrul receives and processes those requests to serve the knowledge; its
privacy information applies.

The package has no executable, dependency installer, self-updater, shell
commands, background worker, credential access, file upload or separate
analytics sender. It has no account requirement and no hosted generation or
adviser calls. Your existing agent provider usage still applies. Public
reading is subject to service availability and rate limits.

The server exposes read tools only. Hermes' untrusted server setting also
requires the client's normal approval for any future write-capable tool.
Retrieved instructions and resource links do not grant execution or sharing
permission. Generating or editing media remains in your existing tools under
their licences and your existing permissions.

Evidence and limits

The task index keeps documentary guidance separate from bounded executed
examples. It includes preserved originals, numerical timing tests and codec
checks where those checks exist. Those results do not establish arbitrary
project compatibility, rendered-media quality or an advantage over another
agent. See VERIFICATION.json for the exact package checks performed before
publication. Independent adoption and catalogue approval are separate.

Actual Hermes use and related-return example
checks two caption edits against frozen numerical answers. The native Codex
runtime needed explicit test isolation; its skill callback approval limitation
and all diagnostic overhead are disclosed. The successful retrieval path used
the existing five read-only tools.

Choose a media-workflow companion for the same direct
read setup in other clients. Version 1.0.2 changes setup guidance and keeps the
previously tested MCP and skill bytes unchanged.

README from the reviewed commit 02ed7b7; it updates when the author re-pins.