Verifies the AIUI-01 gating question against source (no tool-calling
anywhere in this codebase today; Pine's HA intents are read-only Q&A,
not an action-executing loop), surfaces a live unauthenticated
Claude-proxy exposure (port 3142) and a same-origin iframe sandbox
gap not previously named, and maps existing Cashu/Nostr primitives
onto the Routstr integration.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Records the two release-process traps for the next cut: the manifest is
committed before signing (so the fleet would refuse the OTA), and
gitea-vps2 is the same server as gitea-ai with a dead token.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
AIUI is embedded and styled but not functional: the chat cannot act on the
node and its content views are not wired to real data. Phase 13 scopes making
it work — Pine's human-language intent->action capability reachable from typed
chat, conversational settings, and the peer-files/music/movies/node-content
surfaces rendered live.
The gating requirement is AIUI-04: a user-granted capability sandbox. An LLM in
the browser is now adjacent to wallet keys, macaroons and node identity, so
secrets stay server-side behind scoped tokens, capability grants default closed
and stay revocable, destructive operations need a human confirmation, and
peer-supplied text is treated as untrusted input to the model context. This must
not widen the Phase 10 hard-refuse gates.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Bookkeeping left uncommitted by an earlier session. It records work that
is already shipped — bc9a210c routed Paid Files pictures/videos into the
app lightbox with a visible wait — so the checklist and the mapping table
were simply lagging the code.
Landed as its own commit rather than swept into the release: it is not my
edit, and the release script refuses to run with a dirty tree.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Includes what was NOT verified — the torrc block is deployed but dormant,
since regenerate_torrc only fires on a Tor services change and the change
is inert until bitcoind gets an -onion flag in Phase 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Per the operator: every option umbrelOS surfaces must be reachable in the
UI, Knots-only options surfaced separately from the ones Core shares, and
network mode a setting whose DEFAULT is Tor rather than clearnet.
Scoped as a phase rather than done inline because bitcoind's arguments are
currently hardcoded in three places (first-boot-containers.sh,
container-specs.sh, apps/bitcoin-knots/manifest.yml) — the same
triplication that produced the lnd-ui HTTP 000 defect. There is nowhere
for a UI to write, so BTCSET-01 is a settings model those three render
FROM, not another restatement.
Two constraints recorded up front so they are not discovered late:
- Knots-only flags gated to Knots is a CORRECTNESS requirement — offering
one on Core yields a node that refuses to start.
- Several options are not freely reversible: txindex forces a reindex,
prune is destructive and needs a full resync to undo. On a node that is
somebody's wallet backend those must be labelled and gated, not
silently applied. Any change at all restarts bitcoind, interrupting
LND, electrs and the fedimint gateways.
Inbound onion is explicitly out of scope: it needs Tor's ControlPort,
which is deliberately disabled for security, so the node reaches .onion
peers but stays unlisted. The UI must say so rather than imply otherwise.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Deployed 6b3693dc to archi-dev-box off a clean tree and exercised the
real RPC surface. lnd.getinfo returns a real identity_pubkey through the
field that did not exist before this plan. mesh.lightning-peers returns
an empty array with success — the FED-05 empty edge proven on hardware,
not just in a unit test. Both refusal paths of mesh.send-lightning-info
observed live, including T-01-13's "no broadcast form".
The finding worth carrying to 01-06: this node's LND advertises no URI
(uris: []), so send-lightning-info correctly refuses rather than
shipping an empty advertisement a peer would store as an undialable
target. The receive and list halves work; the SEND half is inert on any
node whose LND has no externally reachable address. The picker must not
assume the local node always has something to share.
Corrects this summary's own earlier claim that nothing was exercised on
hardware — that was true when written and is not now. The mesh leg
proper (an advertisement crossing real RF into a peer's lightning_uri)
remains unproven and is still called out as such.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Task 3, completing 01-04.
mesh.lightning-peers returns the peers that have advertised a Lightning
URI: filtered, deduplicated, deterministically ordered, and an empty
array rather than an error when nobody has — "nobody yet" is a normal
state on a fresh node, not a fault.
mesh.send-lightning-info advertises this node's own URI to ONE chosen
peer. There is deliberately no broadcast form: this discloses the node's
payment endpoint, and who learns it is the operator's choice rather than
a side effect of being in radio range (T-01-13). It refuses to send when
LND advertises no URI, instead of sending an empty one a peer would
store as an undialable target.
The list-building and target-parsing logic is extracted into pure
functions because this file has no handler test harness and the
handlers need a live mesh service. That keeps the three contracts that
actually matter provable rather than merely readable:
- dedup is keyed on identity_pubkey_hex() — the AUTHENTICATING key,
lowercased — never the firmware routing key, so a radio contact and
its federation twin collapse to one entry (T-01-11)
- "newest advertisement wins" compares PARSED RFC3339 timestamps, not
strings: 09:30-01:00 is later than 10:00Z while sorting earlier as
text, and there is a test that fails if that is ever string-compared
- ordering is name-then-contact_id and asserted byte-identical across
eight rotations of the input, because a HashMap's iteration order is
not stable and a picker that reshuffles between reads means an
operator can click a different node than the one they aimed at
The peer allow-list is untouched: server.rs has an empty diff and
is_peer_allowed_path still occurs 13 times (T-01-15).
Verified: cargo test -p archipelago 1087 passed / 0 failed; clippy
--all-targets clean in every touched module (two useless_format lints in
the new test code fixed, not waived).
The SUMMARY records one deviation honestly: Task 1's tests were written
alongside its implementation rather than before, so no pre-implementation
failing output exists. A mutation test was run in its place — disabling
the pubkey validation fails 3 of the 5 tests — which proves the
assertions bind, and the mutation was reverted and verified gone.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Controlled test on archi-dev-box with operator approval. The daemon was
stopped first so the reconciler could not repair the state before the
re-exposure was confirmed — without a confirmed 200, the later 401 would
be consistent with the state never having been broken at all.
1. stale conf installed + container restarted -> POST /bitcoin-rpc/
returned 200 with a real block height and Allow-Origin: *
2. daemon started 20:00:36, nothing else touched
3. 20:02:19 reconcile rendered the conf and logged the expected warn
line naming bitcoin-ui/archy-bitcoin-ui, then restarted it
4. POST -> 401, Allow-Origin origin-scoped
5. conf byte-identical to the pre-test known-good, container healthy
Both halves are now proven on real hardware: a05956c4's template (the
gate works) and f6b5245b's delivery path (the gate reaches a container
the reconciler had been skipping).
Also records the operator's decision AGAINST credential rotation — no
macaroon, no Bitcoin RPC password — with the trade it accepts stated
plainly, so it is not silently re-litigated later.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Names the four open items explicitly, including the one that is easy to
lose: the reconcile fix is deployed but unexercised, so the node's 401
proves the template and not the delivery path.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The node is closed and verified 401 with origin-scoped CORS. But an
unrelated bitcoin-ui rebuild at 18:36 cleared the stale conf before the
reconcile fix was deployed at 19:06, so the 401 proves a05956c4's
template and NOT the delivery path f6b5245b adds.
Window 14 closed (exposure gone, verified). Window 15 opened for the
delivery path, which is deployed but never exercised — bitcoin-ui is
still in the uninstall marker, so this node depends on that untested
path the next time its config has to change.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
system.stats on archi-dev-box returns host_secrets with verdict 'per-node'
and three evidence lines (machine-id anchor 2026-04-09; every SSH host key
and the TLS key newer than the anchor). Previously proven against the file
contract in unit tests only.
Honest limitation: this is one node, not the dev pair — archy-x250-dev has
been offline for two days, so the second node is unreachable, not skipped.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Verified live on archi-dev-box, code fix committed in f6b5245b but not
deployed. Deliberately logged as open rather than fixed: the defect that
matters to an operator is the running node, not the source tree.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
All five KEY-05 layers landed. Records the two things a reader would
otherwise get wrong:
- the 2 boot_reconciler test failures in the full-suite run are
parallel-load flakes (4/4 pass in isolation), not regressions;
- layer (b)'s clippy gate is LIVE but not yet EFFECTIVE, because 42
pre-existing warnings already fail the same -D warnings step.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
SUMMARY for 10-04, plus three WINDOWS.md entries (11-13) so the unverified
items stay visible at ship time: the rotation never exercised on real
hardware, host_secrets never observed in a live system.stats, and the C-3
finding itself — three live nodes still on shared SSH host keys, two of them
also sharing a TLS private key, none of them rotated.
STATE.md and ROADMAP.md deliberately not touched: both carry other agents'
uncommitted work in this shared tree and the orchestrator owns them for this
wave.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Deployed the frontend to archi-dev-box only (--frontend-only, no fleet,
no alpha-tester, no Tailscale, no OTA, no release) and proved all six
UIFIX fixes are in the bundle the node actually serves.
- Resolved the live chunk set from sw.js first: /opt/archipelago/web-ui/
assets keeps every prior deploy's hashed chunks, so a naive disk grep
returns hits from dead chunks and would have produced a false pass
(threat T-01-83, and it was a real trap here).
- Fetched each live chunk over HTTP from http://archi-dev-box and grepped
it: all eight probe strings for UIFIX-01..06 PRESENT.
- Real Chromium boot check on the node at 1440x900 and 390x740: app
mounts, 0 console errors, 0 page errors, 0 failed requests.
- archy-x250-dev recorded as an explicit gap: offline, last seen 2d ago,
no MagicDNS record; still has neither this plan set's nor phase 2's
frontend.
Task 2's six numbered checks are all recorded NOT VERIFIED. They need an
authenticated session on the node (UI returns 401 / redirects to /login,
and no credential was guessed against a node holding real funds), and two
of them are not testable as the node stands: it owns exactly one
purchased item (image/jpeg) and has zero video and zero audio content
anywhere, so the purchased-video, purchased-music and picture-in-picture
checks have nothing to open.
No source file modified, no fix applied inline, and STATE/ROADMAP/
REQUIREMENTS deliberately left untouched - UIFIX-01..06 are NOT closed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Task 1 done. Tasks 2 and 3 are blocked on unmet preconditions and were
NOT auto-approved: no fleet node runs 10-01's gate (installed binary
predates 879de59e and lacks the refusal string), and no second machine
was available to probe from. Nothing moved UNVERIFIED -> VERIFIED.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Written by the previous session's executors for 01-17, 10-01, 10-03 and
10-05, all of which are complete and whose code is already committed. The
session was cut off by a dropped SSH connection before these were staged,
so they were sitting untracked. Recording them so the phase history is not
lost.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The on-disk format was detected by sniffing the first byte for `[` or `{`.
Encrypted blobs begin with a random 12-byte nonce, so roughly 1 in 128
saves produced a valid encrypted file whose first byte was 0x5B or 0x7B;
those were misread as plaintext JSON, failed `String::from_utf8`, and the
store became permanently unreadable. This was surfacing as a flaky
`test_list_credentials_no_filter`, but it is a real data-loss bug: a node
whose ciphertext happened to start with one of those bytes could not load
its credentials.
Writes now carry a fixed `ARCHYCRED1` marker, which cannot collide with a
random nonce, so detection of the current format is exact.
Legacy unmarked files are detected by SUCCESSFUL AEAD DECRYPTION rather
than by another byte sniff. A verifying Poly1305 tag under the node key is
a cryptographic discriminator (~2^-128 false-positive rate), strictly
stronger than any structural guess — which is why the deferred item's
suggested "keep the first-byte sniff as the legacy fallback" was not the
shape adopted. Plaintext JSON remains the last resort, and is still
reachable on a node that has no node key at all.
An undecodable file now errors instead of returning an empty store, so a
transiently unreadable file is never silently replaced by an empty one
that the next save would commit to disk (CLAUDE.md: migrations never
destroy data). Legacy files upgrade on write, never on read.
Tests drive the collision deterministically via an explicit nonce rather
than waiting on the 1-in-128 draw, and cover all three on-disk
populations, the read-path-does-not-rewrite guarantee, and tamper
rejection. 28 passed, 0 failed.
Closes the 10-01 deferred item.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Backend already exists: server.set-name runs hostnamectl set-hostname and
regenerates the TLS cert with a SAN for the new name. This is a UI step.
Records the hazard that decides the design: renaming changes both the mDNS
.local name and the TLS cert, so a rename mid-flow can drop the user's session
in the middle of onboarding — potentially between seed generation and seed
verification. Placement is therefore a design decision, with three options laid
out (last-before-Done, first, or collect-early-apply-late).
Also flags RFC-1123 slugification (users will type "Dorian's Node"), whether
the rename propagates to the Reticulum display name and mesh surfaces, the
reconnection UX, and whether the step is skippable.
Sequenced after the in-flight regenerate_tls_cert atomicity fix, since renaming
is exactly the path that fix protects.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
98 operational files still carry 146.59.87.168 — the domain was only adopted
for the git remote, not for container registry references. Bulk is app
manifests' image: lines, plus .gitmodules, both CI workflows, the signed
catalog.json, and two Android companion files with compiled constants. The 117
hits in .planning/ are historical records and stay.
Not a find-and-replace: the domain serves Gitea over HTTPS:443 while images are
pulled from :3000 over plain HTTP, and podman treats host:3000 and domain as
different registries — so every node re-pulls under the new name and any node
that can't resolve or trust the new host fails to pull. It also invalidates the
signed catalog (needs a re-sign ceremony) and the APK ships compiled constants.
Rollout order: registry serving on the domain → manifests → catalog re-sign →
APK rebuild. Steps 2-4 are actively breaking until step 1 holds.
Analysis from the concurrent agent's session before it ended; recorded so it is
not lost.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Companion build 0.5.27 (versionCode 47) shims navigator.clipboard natively, so
in-app copy/paste is fixed with zero web changes — but the contract must not be
clobbered (no unconditional re-define, no Object.freeze).
Still open web-side: main.ts's fake readText() makes SendBitcoinModal's Paste
button render and silently no-op in plain-HTTP browsers; 30 writeText call
sites across three inconsistent patterns, ~10 of which toast 'Copied!'
regardless of success; scanner prewarm/torch/constraints/no-reinit.
Also records three factual corrections to docs/qr-scanner-snappiness-handover.md
(ZXing not ML Kit; FORMAT_QR_CODE + KEEP_ONLY_LATEST already in place; do NOT
drop to 720p — 1080p is a deliberate 0.5.22 fix for dense bolt11 QRs).
Routed at Phase 11: the signed-PSBT paste affordance and the scanner items are
the same surface as WALLET-05.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The F-10a scope correction committed hours earlier asserted semantics its
evidence did not support. The KEY-05 planner caught it against the code:
- mesh/x3dh.rs:100/:114 are u32 prekey IDENTIFIERS (spk_id, otk_id), not key
agreement material. The X25519 secrets come from
crypto::generate_x25519_ephemeral() at :99/:113 and were never in scope.
- session.rs's 16 raw matches read as 16 production token sites; #[cfg(test)]
begins at :470, so it is 4 production + 12 test.
- wallet/bdhke.rs is 2 production of 4 (#[cfg(test)] at :143) — and those two
ARE genuine key material: generate_secret() :133 and
random_blinding_factor() :139.
The Medium rating still holds, on narrower grounds: bdhke's two production
sites plus storage_crypto.rs:39's AEAD nonce. It no longer rests on x3dh.
Struck rather than silently rewritten. F-10 was corrected on the grounds that
understatement misleads the next reader; overstatement does the same, and this
table managed both within a day.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adds 10-06-PLAN.md covering KEY-05 (F-10a / R-16). 10-01..10-05 untouched.
Six tasks, sequenced so CI stays green at every intermediate commit:
classify all 43 call sites with file:line evidence; a tracer that wires the
sealed KeyGenRng allowlist, the degenerate-entropy predicate and the CSPRNG
readiness ledger end-to-end through the mnemonic seam; two migration tasks;
a blocking human checkpoint for cargo-deny scope and legitimacy; then the
gates are enabled last and observed failing a real build.
The clippy ban is a compile failure under the existing -D warnings CI step,
so core/clippy.toml is deliberately not committed until every site --
including test code, since --all-targets counts it -- has migrated.
Wave 2: shares seed.rs with 10-05 and api/rpc/auth.rs with 10-01.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The checkpoint on archi-dev-box proved rotation alone doesn't close FED-07:
the credential file went unique while the RUNNING container kept serving the
compromised one, because the Quadlet path rewrites a unit without restarting
it and fedimint-gateway is classified restart-sensitive, so drift was detected
and deliberately ignored on every tick.
Rotation now records the app id, and the drift check consumes that flag to
recreate even a restart-sensitive app, with a WARN naming the reason. This
mirrors the published-port carve-out a few lines above, which already makes
the same trade for the same reason: a container that is already broken (there)
or already compromised (here) is not protected by leaving it running.
Restart-sensitivity protects working services. A gateway answering to a
credential published in this repository is not working, it is compromised, and
gateway admin can drain Lightning liquidity — indefinite exposure loses to a
few seconds of restart. Rotating-but-only-alerting was rejected: the
monitoring system fires on metric thresholds only, so it would have needed new
event-alert plumbing to deliver something strictly weaker.
Re-verified on the same node, same scenario: rotation at 06:39:23, recreate at
06:39:27, PID 3923125 -> 148426, running credential now matches the file,
container healthy with the same name and ports, gatewayd.db intact at 18 files
with IDENTITY present, 32 containers untouched, no repeat rotation.
3 new tests. Also lands the missing 01-19 and 01-20 SUMMARYs: both had code
committed 2026-07-31 but no summary and no roadmap tick, so they read as
unstarted. Phase 1 is 11/20. FED-09 carries 15h of Tor uptime and 0
permission-fixes across 542 doctor runs on archi-dev-box.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The audit recorded F-10 as two call sites in container/secrets.rs. The real
defaulted-RNG surface is 41 sites across 15 files: session.rs (16),
pine_ha.rs (6), wallet/bdhke.rs (4 — ecash key material), mesh/x3dh.rs (2 —
key-agreement material), storage_crypto.rs (1 — AEAD nonce), +10 more.
Nothing is broken today: rand::random()/thread_rng() are ChaCha12 seeded from
getrandom(2). What changes is blast radius — F-10's Low rating rested on
'per-app credentials rather than the master key hierarchy', which does not
survive the true scope. Re-rated Medium as F-10a.
Records why the original audit missed it: F-10 was reached by tracing the
manifest-secrets path, and no step enumerated defaulted-RNG use across the
crate independently of the traced paths.
F-10's original text is left unedited so the correction is auditable rather
than retroactive. R-13 superseded by R-16; tracker item replaced.
Adds KEY-05 to Phase 10: sealed allowlist trait at key-gen seams, clippy
disallowed-methods ban (compile-time, CI-enforced), cargo-deny on duplicate
rand majors, degenerate-entropy runtime check, persisted CSPRNG-readiness
verdict. Also retires the false 'impl CryptoRng for CountingRng' at
seed.rs:656.
Records the user's execution gate: Phase 10 does not start until the
concurrent Phase 1 agent is finished and their changes are synced. KEY-05 is
unplanned — the existing 5 plans predate it and a 6th is required.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Handoff for the reactive demo-day session that followed 09-06/09-07
(both already complete). Covers: security audit (6 IDOR fixes across
the botfight repo), Cashu payout claim UI, existing-bot AI-config UI,
botfights 1.2.11 built+deployed to both demo nodes, catalog
signed+published.
Also: discovered and fixed 4 botfight-repo commits that were local-only
and never pushed to origin — pushed as part of this handoff step
(botfight @ d00e792..10d4209 -> origin/main).
Co-Authored-By: Claude <noreply@anthropic.com>
The executor was instructed to leave docs artifacts to the orchestrator; this
commits them: the research that drove the audit, the task summary, and the
archi-dev-box test-node todo raised during the same session.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Ran Task 2's blocking checkpoint on a real node. The rotation works and was
proven end to end: it fired ~15s after restart, wrote a fresh unique
credential (0600, service-owned), logged exactly one line naming the .pw path
with no value in it, left every other secret and the gateway's data untouched,
and kept the container's name and ports.
But the assumption the plan rests on is WRONG, and the checkpoint is what
caught it. 25 minutes after rotating, /proc/<pid>/environ showed the running
gatewayd still using the PRE-ROTATION credential while the file and podman
secret held the new one. The orchestrator explains itself in its own logs:
Quadlet unit drift-synced — file rewritten, .service NOT restarted
(operator restart picks up new config)
container drift detected during boot reconcile;
leaving running restart-sensitive app untouched
Two deliberate guards: the Quadlet path never restarts a unit it rewrites, and
fedimint-gateway is classified restart-sensitive so drift is detected on every
tick and then ignored — logged at 15:51, 15:53, 15:54, 15:56 and counting.
So on a real affected node the credential file becomes unique while the
gateway keeps answering to the compromised one until an unrelated reboot, and
the operator reading the .pw gets a password the gateway rejects — T-01-77
inverted. FED-07 is NOT closed and this plan alone cannot close it.
Not hand-rolled around, per the plan's own instruction. The fix needs a design
decision: whether a compromised credential is the case that should override
restart-sensitivity, or whether rotation must raise an operator-facing
"restart required" alert instead of logging into the void.
Incidentally disproved: restarting archipelago does NOT kill containers here
(29/29 then 31/31 survived; "Adopted 31 existing container(s)"). The service
is system.slice/KillMode=control-group while containers live in
user-1000.slice/…/libpod-*. The CLAUDE.md SIGKILL rule predates Quadlet.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
FED-07 was marked Complete when 01-11 landed, which was premature: the
requirement text explicitly includes "existing installs with the default
password get a migration path", and that migration has never been exercised on
a node. Corrected to code-complete/verification-pending.
Checkpoint step 1 was run read-only on archi-dev-box: the node is CLEAN (hash
present, 600, service-owned, not the shipped default) and has NO gateway
container — the app is installed but nothing runs and its data dir is empty.
So rotation cannot fire naturally here, and the steps that matter most (data
survives the recreate, new credential authenticates, old one rejected) have
nothing to exercise without installing and seeding first.
Deferred deliberately rather than run unattended: 30 containers are up with
4-8 days uptime (IndeeHub, Immich, BTCPay, netbird, strfry, …), the
archipelago system service is active, and restarting it SIGKILLs containers
until Quadlet is the default.
The todo carries the full context plus two adjacent findings: fedimint-gateway
is missing from handle_package_credentials (so a rotated password has no UI
retrieval path), and photoprism ships a fixed admin password in its manifest.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
01-11 stopped new installs from ever taking a shipped credential, but did
nothing for the nodes that already did — those gateways still answer to a
password published in this repository.
rotate_compromised_gateway_credential() detects an EXACT match against the
denylist and replaces the pair; absent, unique, or merely unrecognised values
are left alone and return false. That distinction is the point: an operator
who deliberately set their own credential also has an "unrecognised" one, and
rotating it would be the same class of harm as leaving the default in place.
It hangs off resolve_dynamic_env beside ensure_generated_secrets, gated on the
gateway's app id, so an affected node heals on its next reconcile tick. There
is deliberately no teardown here: the new hash changes the resolved secret env,
which changes secret_env_hash, which the drift check reads as a container-label
mismatch — so the platform's own recreate path rebuilds the gateway around its
unchanged data directory, ports, volumes and name.
Rotation is self-terminating (the value written is not on the denylist, so the
next tick is a no-op) and errors propagate rather than being swallowed, because
the atomic write leaves the previous credential intact on failure.
Bcrypt generation was factored out of ensure_one into write_bcrypt_pair, which
both generation and rotation call — 01-11's SUMMARY claimed such a helper
existed but the arm was still inline, and rotation cannot reuse
ensure_gateway_credential because its idempotent fast path returns early
exactly when the file is present, which is the case rotation acts on.
Also fixes cargo fmt drift left by 42652547 in install.rs.
Verified: 6 new tests, secrets suite 16/16; full suite 1008 passed with one
known wall-clock flake (green 4/4 in isolation). NOT verified on a node —
Task 2's blocking checkpoint has not been run, so FED-07 stays open.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The two-scan dance (node shows unsigned PSBT as animated QR, signer signs,
node scans the signed PSBT back, finalize + broadcast) is ~80% plumbed and 0%
usable. Verified gaps:
- No UI: lnd.create-psbt / lnd.finalize-psbt and their rpc-client.ts:417
wrappers are called by nothing but unit tests.
- No animated-QR encoder: qrcode/qrloop are deps and the inbound path
(useAnimatedQRDecoder + WalletScanModal) works, but nothing encodes a PSBT.
- Wrong format for real signers: qrloop is Ledger's; Passport/SeedSigner speak
BC-UR (ur:crypto-psbt), Coldcard Q speaks BBQr. BC-UR is the priority given
the existing Passport-Prime-compatible SeedQR work.
Gated on 10-05: create-psbt funds from LND's own wallet, so until LND is
watch-only against the external signer the offline device signs inputs whose
keys the node already holds.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
First-run wallet-type chooser, seed handling reusing the shipped SeedQR +
seed-words components, and evidence-based umbrelOS LND UI parity.
Gated on Phase 10's 10-05: the set of wallet types WALLET-01 can offer is a
direct consequence of the watch-only verdict that plan produces, and 10-05 also
deletes the dead Core wallet path so this phase never represents it in the UI.
Records the already-shipped inventory (channels panel, send/receive/scan/
settings modals, SeedRevealPanel, LndSeedBackupPrompt, utils/seedqr.ts) so the
parity matrix closes real gaps instead of rebuilding existing surfaces.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
D-03 named NodeIdentity::key_exists as one of the two gate signals. Server::new
(server.rs:63-72) calls load_or_create on both branches, and load_or_create
(identity.rs:48-51) generates and writes a random temporary node key when none
exists — so key_exists is true on any node that has booted once, onboarded or
not. A gate keyed on it would refuse seed.generate on a fresh node and brick
onboarding fleet-wide.
The flaw came from the audit's own suggested remediation (§214-221) and was
repeated in the planning brief; the planner caught it against the code.
D-03's intent (two signals, OR-ed, fail safe on drift) is unchanged. Corrected
signal set: is_setup() / is_onboarding_complete() / seed_exists(), pinned by a
test rather than a comment.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Closes the three exploitable findings from the 2026-07-31 entropy/seed audit.
Wave 1 (parallel):
- 10-01 KEY-01: shared onboarding gate refuses seed.generate/seed.restore/
seed.save-encrypted/backup.restore-identity/auth.setup on a provisioned node,
plus an auth.onboardingComplete guard and retry-budget-derived rate limits.
Independently shippable (D-11): no depends_on, no shared files.
- 10-03 KEY-02: first-boot secret regeneration retries with backoff then fails
closed; rootfs tar ships identity-free so failure degrades to "no key".
- 10-05 KEY-03: delete the uncalled bitcoin.init-wallet-from-seed xprv-import
path (D-07b); make LND's PSBT round trip first-class with a key-origin report.
Wave 2:
- 10-02 (deps 10-01) KEY-01/KEY-04: on-node C-6 exposure measurement, live
refusal proof, fresh-node onboarding non-regression.
- 10-04 (deps 10-03) KEY-02/KEY-04: fleet detection of image-baked host secrets,
guarded rotation behind a D-06 decision checkpoint, C-3 two-node verification.
Planning-time scoping correction recorded in 10-01: D-03 names
NodeIdentity::key_exists as the on-disk "onboarded" signal, but server.rs:63-71
writes a temporary node_key on every boot, so that signal is true on fresh nodes
and would brick first-boot onboarding. D-03's dual-signal intent is preserved
with is_setup / is_onboarding_complete / seed_exists.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Core's wallet is outdated and used by nothing, so bitcoin.init-wallet-from-seed
is deleted outright rather than migrated: uncalled, authenticated and
password-gated, never ran on archi-dev-box, and its only job is deriving and
stringifying the master BIP-84 xprv.
D-07's parity-proof migration and its one-way checkpoint are withdrawn — there
is no wallet to migrate. A small discovery check folds into KEY-04; a wallet
found there is a finding to stop on, not an auto-migration trigger.
PSBT is already solved by LND and already implemented: lnd.create-psbt
(WalletKit FundPsbt) and lnd.finalize-psbt (finalize + broadcast), both
rate-limited, on LND v0.18.4-beta. KEY-03 becomes: delete the Core path and
make that flow first-class, tested and documented, including that the PSBT
carries the BIP-32 key-origin data a hardware signer needs.
Records the standing honesty constraint that Lightning channel/revocation/HTLC
keys are not air-gappable at all, and defers the BDK+ElectrumX cold vault to
its own phase (D-07c).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Bitcoin Core's wallet is legacy and unused: bitcoin.init-wallet-from-seed has
no caller outside its dispatcher registration, the wallet UI is LND-only
(lnd.sendcoins/estimatefee/getinfo), archi-dev-box has no bitcoin/wallets/ dir
so the handler's named descriptor wallet was never created there, and the
endpoint is authenticated + password-gated so F-13 was never remotely
reachable.
F-13 is therefore latent, not live. D-07's migration premise is unproven, so
KEY-03 is re-scoped discovery-first: check the fleet for any wallet this
handler created before planning any migration. Migration + checkpoint stay,
conditional on discovery.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reactions, replies, read-receipts, edits, deletes, forwards and channel sends
shared one bare `{ ok: true, sent: true }` case, so none of them rendered on
the demo — the UI derives reaction chips and reply quotes from the message
store, and there was nothing in it to derive from.
Each now mirrors its daemon counterpart. Reactions/replies/receipts push typed
messages carrying the { sender_pubkey, sender_seq } target key Mesh.vue's
reactionIndex and replyTargetPreview read. Edits rewrite the text and set
edited_at; deletes tombstone IN PLACE (plaintext, typed_payload.deleted,
message_type 'delete') because that is what mesh/mod.rs apply_local_delete
does — it does not remove the row.
Edits and deletes go through a per-session overrides overlay keyed by
sender_seq, because mesh.messages rebuilds its seed array on every read, so
in-place mutation would only ever work for messages sent this session.
mesh.refresh and mesh.reboot-radio stay acknowledgements on purpose — the
daemon's handlers have no message-store effect either — with a comment saying
so, so a later reader does not "fix" them into divergence.
Also completes the phase bookkeeping for 01-02/03/11/12/13/14/15 and lands the
orphaned 01-12/01-14 SUMMARYs.
Verified: parity harness 17/17 live assertions; full frontend suite 102 files
/ 822 tests green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Both paid-tick surfaces now show the EQ-segment ring instead of a CSS ripple
burst (send modal) and a plain circle (scan modal), via a new `badge` size
variant at 160px/192px with matching --viz-radius. A transform scale of the
compact variant was ruled out in 01-UI-SPEC.md because it would scale segment
stroke width and blur along with the geometry.
Both surfaces use the identical composition — a badge-sized relative container
with the checkmark core absolutely centred over the ring — so the two ticks
cannot drift apart visually.
ScreensaverRing also gains the prefers-reduced-motion guard it never had, on
the component rather than per call site, so the screensaver and
SystemDangerZone variants are covered too.
Also records live-browser verification for 01-13's scroll cue (three
viewports, plus a geometry probe of the hide condition).
Verified: 5 new tests; full suite 102 files / 822 tests green; npm run build
clean with viz-ring-badge present in the built bundle. The live visual
no-clipping observation Task 2 asks for is explicitly NOT done — recorded as
deferred to plan 01-07's consolidated sign-off.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
On a short viewport the seed-confirmation tickbox sits below the fold inside
the step's scrolling area while Continue stays pinned and disabled in the
fixed footer — onboarding reads as broken rather than incomplete.
The cue is a sticky-bottom scrim and glass pill inside the scroll region, and
its visibility comes from real geometry: scrollHeight vs clientHeight for
overflow, then a getBoundingClientRect comparison of the tickbox's bottom
against the container's. On a tall screen the element does not render at all,
so those screens are unchanged. Rects rather than offsetTop because offsetTop
is relative to the nearest positioned ancestor — here the outer card, not the
scroll container.
It is wayfinding only: activating it scrolls the tickbox into view and never
sets confirmed, focuses Continue, or auto-ticks, which a test pins.
Listener setup was moved onto both onMounted paths — the sessionStorage
restore path returned early, so a user navigating back would have had no cue.
Verified: 6 new tests plus the full frontend suite green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Entering picture-in-picture read as the lightbox being dismissed, and the
session died with it: both Teleport and KeepAlive move their subtree on
deactivation, which the PiP spec treats as removal.
MediaLightbox now listens for the video's own enterpictureinpicture event —
so PiP entered by the browser's native control behaves identically to the
toolbar button — and follows a fixed order: adopt, animate, then emit close.
Adopting first is what makes the element survive the unmount the emit
triggers; the invariant is documented in place so a refactor cannot reorder
it innocently.
The backdrop animates a handoff on the PiP path only, closing on transitionend
with a bounded 350ms fallback for browsers that skip the transition and for
the reduced-motion path where the duration is zero and the event never fires.
Verified: 5 new tests plus the full frontend suite green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Six code paths configured the Lightning gateway with a bcrypt hash committed
to this repository — and one deploy path with a plaintext password literal —
whenever the per-install secret was missing. Anyone holding a copy of the repo
held the admin credential for every gateway that ever took a fallback.
container::secrets now owns the credential end to end: ensure_gateway_credential
(idempotent, delegates to ensure_one's bcrypt arm) and gateway_bcrypt_hash,
which returns Err when the secret is missing/empty and when the stored value is
on the KNOWN_DEFAULT_GATEWAY_HASHES denylist — so this codebase cannot hand
back the compromised value even to a node already carrying it.
get_app_config was widened to Result so a credential-less install cannot reach
podman run at all; configure_fedimint_lnd takes the resolved hash instead of
re-reading with its own fallback. The four shell paths stop generating
credentials entirely (dropping the htpasswd host dependency) and skip container
creation with a printed reason rather than substituting anything.
Naming converges on the manifest's fedimint-gateway-hash/.pw, with legacy
fedimint-gateway-password values copied forward rather than regenerated so no
node loses a working unique credential. Plan 01-16 owns rotation of installs
already carrying the default.
Verified: cargo build clean; cargo test -p archipelago 999 passed (2
boot_reconciler timing tests failed under concurrent load, green in isolation,
untouched by this diff); bash -n clean on all five scripts; the compromised
literal now appears exactly once in the tree, as the denylist.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
