Imported from magnus919/agent-skills (peertube/SKILL.md). Install upstream with npx skills add magnus919/agent-skills --skill peertube. Copyright stays with the author (MIT).

peertube — PeerTube federated video from the terminal

Browse any PeerTube instance — a federated deployment, not a single API — from the terminal: instance stats, latest videos, full video detail, comment threads, channels, accounts, and instance-local search. Authenticate with OAuth2 only for your own account commands. Every command is read-only except login/logout.

Setup

1. Choose the instance to talk to. Every command is per-instance; the API shape is identical everywhere, but accounts, tokens, rules, and catalogs are not:

export PEERTUBE_SERVER="https://<INSTANCE_HOST>"   # e.g. https://tilvids.com

To search the whole fediverse instead of one instance, point the same variable at the public search index: PEERTUBE_SERVER=https://sepiasearch.org (same API shape — see references/search-and-discovery.md).

2. Nothing else is required to browse: videos, search, channels, comments, and instance info are anonymous reads.

3. (Optional) Log in only for your own account commands (me, my-videos):

scripts/peertube login --username <NAME> --prompt

How authentication works

PeerTube uses plain OAuth2 with per-instance client credentials: the CLI anonymously fetches the client pair from GET /api/v1/oauth-clients/local (singular local), then exchanges your username/password for a bearer token at POST /api/v1/users/token (grant_type=password, form-encoded). The token rides Authorization: Bearer <token>, lives for the instance's configured lifetime (read expires_in from the response — do not assume a fixed number), and is refreshed automatically when it expires. The token file is written owner-only to ~/.config/peertube/token.json keyed by server URL. Do not commit tokens — they are account credentials; revoke with scripts/peertube logout (POST /users/revoke-token) when done. Current production instances mask client_secret in the API response; the CLI detects this and explains the workaround. Details and wire-level error signatures: references/auth-and-tokens.md.

Essential Commands

server — instance stats and identity (anonymous)

scripts/peertube server              # name, description, user/video/view counters
scripts/peertube server --json

Composes GET /config/about + GET /server/stats (canonical paths — there is no /instance/stats).

videos — browse the instance's uploads (anonymous)

scripts/peertube videos                       # latest 15, offset pagination
scripts/peertube videos --limit 50 --offset 50
scripts/peertube videos --sort -views --json  # popular first

Pagination is start/count offsets (max count 100) — the API has no page parameter.

search — find videos on THIS instance (anonymous)

scripts/peertube search --query "linux"            # instance-local (searchTarget=local)
scripts/peertube search -q "docker" --limit 20 --json
PEERTUBE_SERVER="https://sepiasearch.org" scripts/peertube search -q "linux"   # fediverse-wide

The bundled CLI performs instance-local search only (searchTarget=local). For fediverse-wide search, point PEERTUBE_SERVER at SepiaSearch — same commands, wider index. Search results carry channel.host/url, the origin instance of federated hits.

video — full detail for one video (anonymous)

scripts/peertube video --id <UUID>          # numeric id, UUID, or shortUUID all work
scripts/peertube video --id <UUID> --json | jq '{name, description, views, url}'

comments — top-level comment threads (anonymous)

scripts/peertube comments --id <UUID>                 # GET /videos/{id}/comment-threads
scripts/peertube comments --id <UUID> --limit 30 --json

channels / channel / account — creators (anonymous)

scripts/peertube channels --limit 20 --json           # instance channel list
scripts/peertube channel --handle framasoft@framatube.org   # name or name@host
scripts/peertube account --name chocobozzz@framatube.org

channel shows metadata plus the channel's uploads (offset-paginated).

me / my-videos — your account (requires login)

scripts/peertube me --json | jq '.role.label'
scripts/peertube my-videos --limit 50 --json

login / logout — OAuth2 session management

scripts/peertube login --username <NAME> --prompt        # hidden prompt
echo "<PASSWORD>" | scripts/peertube login --username <NAME> --password-stdin
scripts/peertube login --username <NAME> --otp <CODE>    # 2FA-enabled accounts
scripts/peertube logout                                   # revoke server-side + delete file

Global flags

scripts/peertube --json videos                     # flag before or after the subcommand
scripts/peertube videos --json
scripts/peertube --dry-run search --query test     # request plan, zero network
scripts/peertube --verbose videos --limit 2        # trace requests on stderr
scripts/peertube --server https://tilvids.com server   # per-invocation instance override

--dry-run emits {"dry_run": true, "method", "path", "params"} (login adds form_fields names only, never values) — use it to verify a jq chain before running it live. --help and --dry-run never require credentials.

Pipeline recipes

Search, then inspect the top hit

scripts/peertube search --query "linux" --limit 5 --json | jq -r '.videos[0].uuid'
scripts/peertube video --id "$(scripts/peertube search -q linux --limit 1 --json | jq -r '.videos[0].uuid')" --json

Page through a channel's uploads

scripts/peertube channel --handle framasoft@framatube.org --limit 100 --offset 0 --json | jq -r '.videos[].name'
# loop: advance --offset by the returned count until you reach .total (no page param exists)

Instance report card

scripts/peertube server --json | jq '{name: .instance.name, videos: .stats.totalLocalVideos, users: .stats.totalUsers, views: .stats.totalLocalVideoViews}'

Log in, check quota, log out

scripts/peertube login --username <NAME> --prompt
scripts/peertube me --json | jq '{username, role: .role.label, quota_bytes: .videoQuota}'
scripts/peertube logout

JSON and jq

--json output keys are stable snake_case wrappers around raw API objects: videos (the API's {total, data} list objects), channels, threads (+ total_not_deleted), instance + stats, channel, dry_run/method/path/params for plans. Video objects keep PeerTube's own field names — uuid, shortUUID, name, duration (seconds), views, publishedAt, account{name,displayName,host}, channel{name,displayName,host} — so jq selectors transfer directly to raw curl against /api/v1. Example: jq -r '.videos[] | [.name, .views, .channel.displayName] | @tsv'.

Known Gotchas

• Instances are independent (federated, not one API) — accounts, tokens, rules, enabled features, and catalogs differ per instance. A token from instance A 401s on instance B; the CLI keys the token file by server URL. Content federated onto an instance still belongs to its origin (channel.host, video url).
• Search scope is two different things — searchTarget=local searches the instance's own catalog; search-index (or SepiaSearch's base URL) searches the fediverse via an external index. Omitting searchTarget gives the instance's own scope on current servers, not the fediverse. The bundled CLI is instance-local unless you point it at sepiasearch.org.
• page does not exist — collections paginate with start/count (max 100). Clients sending page= silently re-read the first page forever.
• The comments route is /comment-threads (hyphenated) — /comments and /commentthreads are not routes (they 400 on current servers).
• Instance metadata paths are mixed — stats at /server/stats (operation titled "instance stats"), about at /config/about, config at /config. No /instance/* metadata paths exist.
• Production masks client_secret — oauth-clients/local answers "********************************" on current production instances; a token request with the masked value 400s. The CLI detects it and explains the front-end-asset workaround. response_type=code appears in old quick-start curls but is not part of the current token schema — the CLI omits it.
• Token lifetimes are instance-configurable — read expires_in per response; the CLI persists the absolute expires_at and refreshes automatically. Store tokens owner-only, never commit them, revoke on logout (deleting the file alone leaves the session live).
• 2FA needs an OTP header — x-peertube-otp on the token request; the CLI maps a bare 401 to "pass --otp".
• Rate limits — default 50 calls/10 s per IP (token endpoint tighter); on 429 read Retry-After and back off. Errors use RFC7807 application/problem+json bodies, and unknown routes answer 400 (not 404) — read the body.
• duration is seconds; ids are triple (id, uuid, shortUUID — all accepted by detail endpoints); role is an object {id, label}; videoQuota is bytes.
• Anonymous vs authed — browsing/search/comments/instance-info need no token; /users/me* and mutations always do.

When to use

Use this skill for read-only interaction with PeerTube instances: browsing and filtering videos, instance-local or fediverse-wide search (via SepiaSearch), video detail and comments, channel/account exploration, instance stats, and managing your own account session with OAuth2 (login, profile, my videos, logout).

When not to use

Do not use this skill for YouTube, Vimeo, or other platform uploads or any video editing/transcoding (route to those platforms' own tooling and ffmpeg); for installing, hosting, or administering a PeerTube server (instance administration is out of scope — the bundled CLI is read-only plus login/logout); or for generic ActivityPub/Mastodon federation questions (use a Mastodon or ActivityPub skill).

Reference Files

Available Scripts and Prerequisites

• scripts/peertube — the bundled Python CLI (--json, --dry-run, --verbose, --server override). Imports only the standard library and requests.
• scripts/test_peertube.py — offline test suite (pytest + unittest compatible); all HTTP is mocked, zero network egress.
• Requires Python 3.8+ and requests. Any reachable PeerTube instance (or SepiaSearch) works; no credentials exist or are required by default. No service is started by this skill.