Federated via graft
Document federation onboarding, Radicle node bootstrap, token sharing
Three additions to TUTORIAL.md, none of them code:
- "Sharing a token safely": scope it down, never paste into chat/email/
a ticket, prefer the peer writing it themselves over their own SSH
session, use a one-time-secret tool when a third party must relay it,
rotate on any doubt.
- "Adding a peer to an existing federation": the actual steps behind
what's already running in production (a second Forgejo pair joining
an existing series, reusing the same RID/seed rather than standing up
new Radicle infra) — plus what the dashboard's source/replicated cells
mean when a newly-joined side shows them.
- "Adding a brand-new Radicle node to the mesh": rad node connect (the
one-time introduction dynamic peer discovery needs) and rad seed
(seeding policy defaults to block — connected doesn't imply seeded).
- "Future: built-in token exchange": a menu of options graft could
implement instead of leaning on outside tools — a one-time-secret
page with a 6-digit passcode, a QR-code variant for live handoffs, a
magic-wormhole-style PAKE exchange, or skipping the handoff entirely
via a delegated Forgejo token-minting flow. None built; recorded for
when it's worth picking one.
author
jeanbaptiste
date
(Sep 13, 2026, 11:12 AM +0200)
commit
4333571b
4333571b07cdb83f8dd87c1f0d90c78164303d6d
parent
36eadfe5
36eadfe59e888f0788cae9c533a9cd5351529b0b
+95
+95
TUTORIAL.md
+95
TUTORIAL.md
···
41
41
mkdir -p /etc/graft
42
42
echo "$TOKEN" > /etc/graft/my-project.token
43
43
chmod 600 /etc/graft/my-project.token
44
+
chown graft:graft /etc/graft/my-project.token # the user graft actually runs as, not root
44
45
```
46
+
47
+
### Sharing a token safely
48
+
49
+
The token never belongs in `config.yaml` — only its file path does (`token_file`). The token itself lives in its own `chmod 600` file, owned by the user `graft` runs as. That much is enforced by `graft` itself (`config.Load` requires `token_file`, never accepts an inline value). Getting the token *into* that file safely is on you — a few rules:
50
+
51
+
- **Scope it down.** `write:repository` + `write:issue` only — never an admin or org-wide token. A leaked scoped token can only touch the one repo it was made for.
52
+
- **Never paste it into chat, email, or a ticket.** All three are permanently searchable logs the moment the secret lands in them, long after anyone remembers it's still live there.
53
+
- **Prefer the peer's own hands on the keyboard.** The person who owns the Forgejo instance generates the token and writes it directly into `/etc/graft/<name>.token` themselves, over their own SSH session — it never transits through a third party (including you) at all.
54
+
- **When a third party must relay it, use something that dies after one read** — a password manager's one-time-share feature (Bitwarden Send, 1Password Psst), or a self-hosted equivalent (`onetimesecret.com` is open source and self-hostable). Never a plain link with no expiry.
55
+
- **Rotate it if you're ever unsure it stayed private** — regenerating a Forgejo token is free; guessing whether a six-month-old paste is still up is not.
56
+
57
+
See [Future: built-in token exchange](#future-built-in-token-exchange) at the bottom for ideas on `graft` doing this itself instead of leaning on outside tools.
45
58
46
59
## 3. Create the repo on Forgejo
47
60
···
123
136
124
137
It now mirrors both repos every `sync_interval`, in both directions, on its own.
125
138
139
+
## Adding a peer to an existing federation
140
+
141
+
Once `my-project` is mirrored between one Forgejo and one Radicle node, adding a second Forgejo instance to the same repo — a genuine federation, not just a pair — needs neither a new Radicle node nor any change to the two pairs already running.
142
+
143
+
| Placeholder | Stands for |
144
+
|---|---|
145
+
| `git2.example.org` | the second Forgejo instance |
146
+
| `bob` | the account on it that owns the mirrored repo |
147
+
148
+
**The one thing worth understanding first**: a Radicle repository (a RID) already replicates to every Radicle node that seeds it, automatically, regardless of `graft`. Pointing a second Forgejo pair's `radicle.http_base_url` at the *same* seed you're already using doesn't create a new relationship — that seed already has the content. Nothing to do on the Radicle side at all for this case; skip straight to the config.
149
+
150
+
1. On `git2.example.org`, create the repo (`auto_init: true`, same default branch) and a token, exactly as in step 2–3 above, scoped `write:repository` + `write:issue`, saved to its own file:
151
+
152
+
```sh
153
+
echo "$TOKEN2" > /etc/graft/my-project-git2.token
154
+
chmod 600 /etc/graft/my-project-git2.token
155
+
chown graft:graft /etc/graft/my-project-git2.token
156
+
```
157
+
158
+
2. Add a second entry to `config.yaml`, under the same `series` as the first pair (defaults to the pair's own `name` if you never set one — give the *first* pair an explicit `series` too if you didn't already, so both land on the same row):
159
+
160
+
```yaml
161
+
repos:
162
+
- name: my-project-git1
163
+
series: my-project # both pairs share this
164
+
forgejo:
165
+
base_url: https://git.example.org
166
+
owner: alice
167
+
repo: my-project
168
+
token_file: /etc/graft/my-project.token
169
+
radicle:
170
+
rid: rad:z3gqcJUoA1n9HaHKufZs5FCSGazv5
171
+
http_base_url: https://seed.example.org
172
+
rad_home: /home/graft/.radicle
173
+
sync: { git: true, issues: true, patches: true }
174
+
175
+
- name: my-project-git2
176
+
series: my-project # joins the same federation
177
+
forgejo:
178
+
base_url: https://git2.example.org
179
+
owner: bob
180
+
repo: my-project
181
+
token_file: /etc/graft/my-project-git2.token
182
+
radicle:
183
+
rid: rad:z3gqcJUoA1n9HaHKufZs5FCSGazv5 # same RID — not a new one
184
+
http_base_url: https://seed.example.org # same seed — not a new node
185
+
rad_home: /home/graft/.radicle # always graft's own, never the peer's
186
+
explorer_url: https://explorer.example.org
187
+
sync: { git: true, issues: true, patches: true }
188
+
```
189
+
190
+
`rad_home` never changes between pairs — it's `graft`'s own local Radicle identity, the same one it pushes as for every pair it runs. Only the `forgejo` block is specific to the new peer.
191
+
192
+
3. Restart: `sudo systemctl restart graft`. `config.Load` validates every field on startup — a typo surfaces immediately in the logs, not as a silent no-op.
193
+
194
+
4. Confirm: the status dashboard now shows `git2.example.org` as a second row under `my-project`. If it joined an already-active federation, its side may show cells colored differently from a normal commit/issue/patch — "source" (this is where content first entered the federation) or "replicated" (this side holds it without `graft` ever having logged a push here itself, because Radicle's own gossip — or a Forgejo Authorized Integration — got there first). Hover one for the date, the original side, and a link.
195
+
196
+
## Adding a brand-new Radicle node to the mesh
197
+
198
+
Reusing an existing seed (above) needs nothing extra. Standing up a node nobody in the mesh has ever talked to is a two-step manual bootstrap — `graft`'s own Radicle node (`rad_home`) doesn't discover new peers on its own:
199
+
200
+
```sh
201
+
rad node connect <new-node-id>@<new-node-address>:8776
202
+
rad seed rad:z3gqcJUoA1n9HaHKufZs5FCSGazv5
203
+
```
204
+
205
+
Both matter. `rad node connect` is the first handshake — Radicle's peer discovery (`"peers": {"type": "dynamic"}` in `~/.radicle/config.json`) spreads pairs it already knows about, but a node nobody is yet connected to needs one explicit introduction. `rad seed` matters separately: a Radicle node's default seeding policy is to replicate *nothing* it isn't explicitly told to — being connected doesn't imply seeding. Skip either step and the new node stays either unreachable or silently empty.
206
+
126
207
## Troubleshooting
127
208
128
209
**"token does not have at least one of required scope(s)"** — the Forgejo token is missing `write:issue` or `write:repository`. Regenerate it with both.
···
132
213
**"'remote-rad' is not a git command"** — `git-remote-rad` (installed alongside `rad`) isn't on `PATH` for the user running `graft`. Add its directory to the systemd unit's `Environment=PATH=...`.
133
214
134
215
**A patch/PR never shows up on the other side** — check that the corresponding repo isn't still empty on the destination. The first git sync has to land before patches/issues referencing that content can be mirrored.
216
+
217
+
## Future: built-in token exchange
218
+
219
+
Today, getting a token from the peer who generated it into `graft`'s `token_file` is entirely a human problem — see [Sharing a token safely](#sharing-a-token-safely) above. `graft` could own more of that instead of leaning on outside tools. None of this is built; it's a menu of options worth weighing against each other before picking one, roughly cheapest-to-build first:
220
+
221
+
- **Point at an existing one-time-secret tool.** Zero code: document `onetimesecret.com` (self-hostable, open source) as the recommended relay when a direct SSH handoff isn't possible. Lowest effort, but `graft` doesn't control the experience or leave an audit trail of who claimed a token and when.
222
+
223
+
- **A `graft token share` command, minimal version.** `graft` generates a random opaque ID and a separate 6-digit passcode, stores the token encrypted (with a key derived from the passcode, so the encrypted-at-rest copy alone is useless), and serves it once at `https://graft.cyberwild.org/share/<id>` — the page asks for the passcode before it reveals anything, and the record is deleted server-side the moment it's successfully viewed (a second visit gets "already claimed", not the secret). A short TTL (15 minutes) covers the case where it's never claimed at all. Passcode communicated over a *different* channel than the link — a phone call, a second messaging app — so a leaked link alone is still useless. Straightforward to build on top of the state store already in place; the main design care is making sure the "delete on view" step is atomic (one claim, ever, even under a race).
224
+
225
+
- **QR code variant of the same page.** Same backend, but the sending admin displays a QR code on their own screen (in person, or screen-shared on a call) instead of sending a link at all — the receiving admin scans it with their phone. Puts the passcode in the URL fragment (`#passcode=...`) rather than a query param, so it decrypts client-side in the browser and never reaches `graft`'s own access logs. Nicer for a live handoff; no real advantage over the plain link for an async one.
226
+
227
+
- **A wormhole-style PAKE exchange** (the protocol behind [`magic-wormhole`](https://github.com/magic-wormhole/magic-wormhole)): both admins type the same short human-readable code (`7-crossover-clarinet`) into a small CLI, and a key exchange derives a shared secret from the code without either the code or the token ever crossing the wire in the clear — `graft`'s own server, even if fully compromised, never sees the plaintext token at any point. The strongest guarantee of the options here, and the most to build (needs a relay endpoint speaking the handshake, and both sides need shell access rather than just a browser) — worth it mainly if token handoffs become frequent enough to justify the investment.
228
+
229
+
- **Skip the handoff entirely.** The most elegant fix doesn't share a secret better — it avoids minting a long-lived one a human has to move at all. If Forgejo ever exposes a delegated token-creation flow (the peer's admin clicks a link, logs into *their own* Forgejo, and a token scoped to exactly `write:repository` + `write:issue` on exactly that repo is minted and handed to `graft` server-to-server), no plaintext secret ever needs to touch a clipboard, a chat window, or a QR code in the first place. Worth checking against Forgejo's OAuth2/application-token APIs before building any of the options above.