OXYGENHUB
# 6ba07678ae86 oxygenhub / docs / keyless.md
8.2 KiB · Markdown · last changed transparency.md and keyless.md: accepted, with Andy's decisions 19 days ago renderedsource blame history raw
1 
# Keyless signing: short-lived keys from an OIDC identity
2
3 
Design for oxygenhub#241, written 2026-10-17. **Status: accepted** (Andy,
4 
2026-10-17, the four decisions below as recommended). Nothing here is built
5 
yet. It builds on the hub's
6 
OIDC issuer (hub spec §13.5) and on the transparency log
7 
(docs/transparency.md, #238/#240).
8
9 
## Why
10
11 
Every signing key in Oxygen lives long: a runner's `runner.key`, an
12 
agent's key in its home (`~/.oxy-agents/claude/keys`), a person's key. A
13 
delegation can already expire in minutes (`oxy identity delegate add
14 
--expires 30m`), but the key it names stays on disk afterwards, and each
15 
grant re-issues the owner's identity by hand. A stolen key works until
16 
someone notices and revokes it.
17
18 
Sigstore's answer (Fulcio) is a key that lives only for one job: the
19 
signer proves who it is with an OIDC token, gets a certificate for a fresh
20 
key valid for minutes, signs, and throws the key away. Nothing long-lived
21 
sits on the runner or in the agent's home, and a policy can name **who**
22 
may sign ("this pipeline on main", "Claude, for andymccall") rather than
23 
which key.
24
25 
## What exists already
26
27 
- **Delegation** (spec §14.2, §14.5): an identity lists delegates with a
28 
  scope (`changes`, `refs`, `seal`, `all`) and an expiry, checked against
29 
  the time the signed object claims. The identity issues it itself; no
30 
  other key can.
31 
- **The hub's OIDC issuer** (hub spec §13.5, `oidc.rs`): a Flow job that
32 
  declares `identity "AUD"` gets an RS256 token with `sub`
33 
  `repo:OWNER/NAME:bookmark:B`, the environment and its approval, expiring
34 
  with the job (at most 6 h). The JWKS is at `/.well-known/jwks.json`.
35 
  The hub already uses these tokens for keyless backups to GCS and S3.
36 
- **The hub's key** (hub spec §15.5) signs web edits, wiki changes,
37 
  Diffusion bumps, swarm tickets and SLSA provenance.
38
39 
What is missing is a delegation **issued by someone else** (a certificate
40 
authority), policy rules that name identities rather than keys, and time
41 
a verifier can trust once the certificate has long expired.
42
43 
## The certificate
44
45 
A new object, `certificate`, issued and signed by a hub's key (its
46 
composite Ed25519 + ML-DSA-65 key, so certificates stay post-quantum):
47
48 
```
49 
Certificate {
50 
  version     [0]: 1
51 
  key         [1]: Identity        // the fresh key pair's public identity
52 
  issuer      [2]: KeyId           // the hub key that signs it
53 
  subject     [3]: text            // who proved it: "repo:andymccall/oxygen:bookmark:main", "agent:claude@andymccall"
54 
  claims      [4]: map text->text  // what else the token said: environment, approval, run, job
55 
  scope       [5]: Scope           // changes, refs, seal
56 
  not_before  [6]: u64             // milliseconds
57 
  expires     [7]: u64             // not_before + at most 10 minutes for jobs
58 
}
59 
```
60
61 
It travels with what it signed, like an identity: a push carries the
62 
certificates of the keys that signed its changes and attestations, and a
63 
fetch brings them.
64
65 
## Getting one
66
67 
```
68 
POST /api/v1/keyless/certify
69 
  { "token": "<OIDC token>", "identity": "<fresh identity, CBOR, base64>",
70 
    "proof": "<the fresh key's signature over the hub's challenge>", "scope": "refs" }
71 
→ { "certificate": "<CBOR, base64>" }
72 
```
73
74 
The hub checks:
75 
- the token: its own issuer to begin with (others later, see the
76 
  decisions);
77 
- that the proof shows the caller holds the fresh key (the challenge
78 
  signatures of spec §14.3);
79 
- that the scope asked is one the subject may have.
80
81 
It then signs a certificate expiring at the earlier of ten minutes and
82 
the token's expiry, and logs it (below).
83
84 
- **A Flow job** gets a token as it does today, generates a key in memory,
85 
  certifies it, signs (a receipt, a tag, an attestation of a release), and
86 
  the key dies with the job. `runner.key` remains only to authenticate the
87 
  runner to the hub, not to sign what jobs made.
88 
- **An agent** has no OIDC token of its own. The hub gives it one when its
89 
  owner says so: `oh agent grant claude --scope changes --for 8h` on
90 
  Andy's side, or a hub page with a button, makes an agent token. The
91 
  agent exchanges that token for certificates as it needs them (each ten
92 
  minutes). The owner revokes the grant on the hub, and no key is left
93 
  anywhere. The agent's home holds only the grant token, which signs
94 
  nothing by itself.
95 
- **People** keep their own keys; passkeys (#237) are the separate answer
96 
  for them.
97
98 
## Policy names identities
99
100 
A rule today lists keys. It may also list **subjects from an issuer**:
101
102 
```
103 
rule "main" {
104 
  key 4cd9916c…                                   // Andy, as today
105 
  subject "repo:andymccall/oxygen:bookmark:main" issuer HUB scope refs
106 
  subject "agent:claude@andymccall" issuer HUB scope changes
107 
}
108 
```
109
110 
`HUB` is the hub key's KeyId, so the policy says which CA it trusts for
111 
which subjects. A pattern may end in `*`. The policy's roots still sign
112 
the policy, as now.
113
114 
## Verifying later
115
116 
A certificate is valid for ten minutes. A verifier checking a signature a
117 
year later must know **the signature was made inside those ten minutes**,
118 
and the signer's own claim of time is not enough (that is #240's
119 
problem). Two sources, in order of strength:
120
121 
1. **The transparency log** (docs/transparency.md): the hub logs every
122 
   certificate it issues and every object signed by a certified key. An
123 
   object integrated (witnessed) before its certificate's `expires` was
124 
   signed in time. This is exactly how Sigstore uses Rekor with Fulcio.
125 
2. **The hub's acceptance**, until the log exists: when a push carries a
126 
   certified signature, the hub countersigns its acceptance time on the
127 
   object (a small attestation, kind `accepted`). That trusts the hub's
128 
   clock alone, which the log later removes.
129
130 
A signature whose certificate has expired, with neither proof, is
131 
invalid.
132
133 
## Revocation
134
135 
Short certificates mostly replace revocation: a stolen key is useless ten
136 
minutes later. What can still be revoked:
137 
- **the grant** an agent holds (on the hub, at once);
138 
- **the issuer** itself (a policy without the hub's key trusts none of
139 
  its certificates);
140 
- **one certificate** by its hash (a revocation listed in the policy, as
141 
  keys are now), for the ten minutes it matters.
142
143 
## Order of work
144
145 
1. **The certificate object** in oxy-core: encoding, verification (issuer
146 
   signature, scope, expiry against a given trusted time), carried in
147 
   bundles like identities. Spec §14. About three days.
148 
2. **The certify endpoint** on the hub: token check, proof of
149 
   possession, issuing, logging. About two days.
150 
3. **Flow jobs:** the runner certifies a fresh key per job and signs
151 
   receipts with it; the hub checks receipts by certificate as well as by
152 
   `flow_runner_keys`. About two days.
153 
4. **Policy subjects:** rules naming issuers and subject patterns; `oxy
154 
   policy grant --subject`. About two days.
155 
5. **Agent grants:** `oh agent grant`, the hub's grant tokens, and oxy
156 
   certifying on demand when `author.signer` is `keyless`. About three days.
157 
6. **Trusted time:** the acceptance attestation now; the log's integrated
158 
   time when #238 lands.
159
160 
## Decisions (taken 2026-10-17)
161
162 
Andy chose the recommendation each time: our own composite-signed
163 
certificate, built now with the hub's acceptance time (the log's time
164 
later), agent grants that retire the agent keys, and the hub's own OIDC
165 
issuer first.
166
167 
1. **Certificate format:**
168 
   - Our own CBOR object, composite-signed: post-quantum, and small.
169 
   - X.509, as Fulcio issues: readable by Sigstore's tools, but classical
170 
     only.
171 
   - Recommendation: **our own**, with an X.509 export later if a
172 
     customer needs Sigstore's tools to read them.
173 
2. **Build before or after the transparency log?** Certificates are only
174 
   as good as the proof of when they were used.
175 
   - Recommendation: **build steps 1 to 4 now with the hub's acceptance
176 
     attestation**, and move to the log's time when #238 step 1 lands.
177 
3. **Agents:** a grant on the hub that the agent exchanges for ten-minute
178 
   certificates, replacing the long-lived agent keys and the 8 h
179 
   delegations re-issued by hand.
180 
   - Recommendation: **yes**, and retire `~/.oxy-agents/*/keys` once it
181 
     works.
182 
4. **Other OIDC issuers** (GitHub Actions, GitLab CI, Google service
183 
   accounts) as subjects, for repositories mirrored or built elsewhere:
184 
   - Recommendation: **the hub's own issuer first**; others as a site
185 
     setting listing trusted issuers, later.