`MissionRuntimeProvisioner::from_env` is called per use — on every mission
launch and from the terminal-mission reaper sweep — so the line added in
cdc45bd would have printed on every tick forever. A log that repeats
itself is a log nobody reads, which would have cost exactly the
visibility the line was added to provide.
Co-Authored-By: Claude Opus 5 <[email protected]>
Every mission on gw-04 was already running v0.8.4 — pinned by
CLAWMATES_RUNTIME_IMAGE in .env. The canary is retired: the default tag
`clawmates-runtime:sync` now IS that image, the override is commented out,
and the built-in default is the single source of truth again.
Promoting it exposed why the pin was load-bearing in the first place. The
default tag resolved to zeroclaw 0.8.3 — two releases behind what was
actually running — and the REGISTRY copy of the same tag was a different
image again, 849MB against 2.31GB, without the Rust toolchain. A host that
pulled `sync` rather than retagging it would have lost the on-green test
gate with every probe still reporting success.
A moving tag pointing somewhere old resolves perfectly, starts perfectly,
and runs old code. Nothing anywhere said which image a mission got, so two
things now do:
- mission_runtime logs the image it resolved and whether that came from
the env override or the built-in default, once at startup.
- runtime_preflight probes `zeroclaw --version` alongside the other tools
and prints every tool's VERSION, not just that it is present. A presence
check passes happily on an image two releases behind, which is exactly
what happened here and was found by running the binary by hand.
Rollback is a retag: `clawmates-runtime:pre-v084-default` on gw-04 holds
the previous default, and .env.pre-v084-default holds the previous pin.
Co-Authored-By: Claude Opus 5 <[email protected]>
The frame is `{"type":"tool_call","id","name","args"}` — zeroclaw-gateway
/src/ws.rs. The tap read `tool` then `name`, and `arguments` then `input`.
`name` happened to be in the fallback chain; `args` was not in it at all,
so a container-tier tool call would have been recorded with its name and
NO path — a tool that reads as having touched nothing. `tool` and
`arguments` belong to `approval_request`, which is where they came from.
Also corrects what the histogram was read as saying. A mission turn on
gw-04 carried only chunk/done/session_start, and the first reading was
"there is no tool_call frame". Wrong: `grep -c tool_call` on the deployed
0.8.3 binary returns 46. Container-tier agents are provisioned tool-free
behind the MCP door (§15), so they call nothing — there is nothing to
observe on that tier, and nothing is broken.
That distinction is exactly what the histogram was shipped to make
possible: a tap matching no frame is otherwise indistinguishable from a
mission that used no tools.
Co-Authored-By: Claude Opus 5 <[email protected]>
The plan deferred this as "the only fleet-node binary change". It is not
one. `fcagent` is thread-per-connection — its own comment says so, and
the live log tail has relied on exactly that for the whole length of a
turn, on a second connection. So the host can drain the tap WHILE the
turn's exec is in flight, from the server alone.
The turn and a 20s drain loop now run concurrently. A coding phase shows
its files being touched as it works rather than an hour later, all at
once, and the drain is bounded by a cursor so a repeated poll returns
only what is new.
The cursor counts LINES, not parsed events, and that distinction is the
bug this commit would otherwise have shipped. The hook appends the event
and then a newline of its own, so a two-event tap is four lines; advancing
by event count leaves the cursor two lines short, `tail -n +N` hands back
events already recorded, and the live drain re-records everything it has
already written — worse the longer the turn runs, and silent throughout.
Caught while writing the test, not by it.
`tap_sink` and `VmOutcome::tools` are mutually exclusive by contract: with
a sink, the sink owns recording including the final batch and `tools`
comes back empty. Handing the same calls back on both would double every
file orb's weight with no way for the caller to tell which it was
looking at.
The sink is an unbounded channel to a recorder task, so the VM executor
stays free of the database: it observes, phase_runner records. The task
ends when the sender drops with the phase.
Verified before this change: the microVM tap is real. The `microvm`
scenario passed 6/6 and left ten `tool.call` rows and a `file.touch` on
MICROVM.md, repo-relative, from Claude Code's own PostToolUse hook.
Co-Authored-By: Claude Opus 5 <[email protected]>
Security: the pawns already orbited their destination, so homing them at
the security station gave circling for free. This adds the radial
press-and-retreat — an agent closing on the target and backing off reads
as probing it, where a fixed radius reads as waiting — and holds the
stochastic target release while probing, or the circling breaks up into
stray trips that look like distraction rather than a scan.
Findings are `mission_tasks` rows, one orb each, popped once. There is
deliberately no severity anywhere in the path: the scanner keeps
severity, file and line as substrings inside `title`, so a severity
parsed out of prose and rendered as an orb's RADIUS would be the picture
asserting a measurement the data never contained. Count only.
Benchmarks annotate the station, as text. `delta` has no schema —
compute_delta emits `{kind:"opaque"}` whenever the before/after metrics
were not structurally comparable, which is most drivers. The server
formats the shape it can parse and COUNTS the rest; an unparseable driver
reports "3 sample(s)" rather than an invented improvement, and an opaque
delta says nothing at all.
The finished map: the live label rule gates service/event nodes on
`heat > 0.12`, which is exactly backwards once everything has cooled — a
static map would be unlabelled dots. Frozen, the 25 most-touched nodes
label regardless of heat, phase stations carry a second line counting
what they produced, and the camera is released ONCE so it frames the
result even if the user panned during the run.
Every count in a caption is read off the drawn scene rather than a
parallel tally, so the words and the picture cannot disagree.
Co-Authored-By: Claude Opus 5 <[email protected]>
The World could draw a mission's shape but nothing about the work. The
detail existed only as prose in checkpoint.log and model output, where a
tool name is indistinguishable from an agent *talking about* a tool — so
it was never parsed, deliberately. `mission_events` is the structured
channel that replaces it.
Three taps, one table:
- Container tier: the `_ => {}` at the end of topology_exec's typed frame
stream now matches `tool_call` and reads the tool's JSON ARGUMENTS for a
path. Never the prose summary — a path scraped from a sentence would put
files on the map that no agent opened, and the test proves a Grep whose
summary says "src/main.rs" produces no file touch. The frame name itself
is unverified, so the same commit ships an unmatched-frame-type
histogram: a tap that matches nothing looks exactly like a mission that
used no tools, and this is how one gw-04 run names the real frame.
- microVM tier: a `PostToolUse` hook, the seam vm_stop_gate already proved
fires under `claude -p`. It copies stdin to /root/tap and exits 0
unconditionally — a non-zero PostToolUse hook talks back to the model,
which would turn the observer into a participant. Drained before collect,
since the VM is destroyed moments later.
- Phase transitions: five identical copies of the pending→running UPDATE
became one `mark_phase_running`, and `close_finished_phases` grew
RETURNING. Its CASE decides each phase's status inside SQL from rows the
statement does not change, so it cannot be re-derived afterwards without
writing that CASE twice — without RETURNING it emits zero phase.completed
and reports success.
The settings.json hazard the plan called out: the stop gate wrote the
WHOLE document, so a second hook writer would have silently erased it and
a coding phase would then complete having written nothing — the exact
failure the gate exists to catch. There is now one composer,
`vm_tool_tap::guest_settings`, one writer, and a source-walk test that
fails if anything else writes a settings document.
`mission_events.run_id` carries no FK on purpose: phase_runner DELETEs
topology_runs on retry, and a cascade would erase a phase's whole history
the moment it retried — silently, since a cascade is not an error.
world.rs streams it with a cursor that separates backfill from motion.
Everything already in the table when a subscriber arrives is drawn as
settled history; only what lands afterwards animates. Otherwise opening a
finished mission replays an hour of tool calls as a burst storm.
Bounded twice: 400 events per phase (enforced inside the INSERT, since
two concurrent taps would each read a count below the cap) and a 7-day
retention sweep in mission_gc.
Co-Authored-By: Claude Opus 5 <[email protected]>
The engine already turned `file:src/lib/a.ts` into a real dir chain, but
both copies of that loop rooted it at the origin — so a mission's files
floated beside the map instead of belonging to the work that produced
them. One `fileParent` helper now serves both call sites; splitting them
was how half the files could end up nesting correctly and half not,
decided by whichever code path saw the file first.
Files hang under the coding station when there is exactly one, else the
single running phase, else the origin. `world.touch` carries no phase id,
so with two coding phases any attribution is invented — the fallback is
the honest answer.
Two ordering hazards, both silent:
- the server emitted files BEFORE phases, so on the first pass a file
arrived with no station to hang under and first-write-wins pinned its
tree at the origin. Loops reordered, with a source-walk guard.
- `setFileHome` re-parents trees rooted before the plan landed, for the
reconnect case the ordering alone cannot cover.
`mission.file` with `source: "tool"` is treated as motion (burst, pawn
beams); `"diff"` is end-of-phase truth and only marks the file present
and warm — bursting every file of a captured diff would set the whole
map alight at once on reconnect.
Directories taper in radius and opacity by path depth, so `src` and
`src/lib/live` no longer render as identical dots.
Co-Authored-By: Claude Opus 5 <[email protected]>
`capture_phase_diff_at` parsed `git diff --stat` down to three integers and
threw the filenames away. Nothing downstream could name a single file a coding
phase changed: the World can draw a coding station but nothing underneath it,
and an operator reading a mission sees "11 files" with no way to learn which.
A second `--name-status` call now records the paths into the code_diff metadata
and into `names.txt` beside `diffstat.txt`, so raw evidence survives
independently of the JSONB.
Three ways this could have been wrong, each guarded:
- Different revision or excludes from the `--stat` call would make
`files_changed` and the path list describe different diffs, with no way to
tell which lied. A source-walk test pins both to the same `base_sha` and
the same `excludes`.
- Running after `git reset --quiet` would drop newly CREATED files, since
`--intent-to-add` is what makes them visible to diff at all — and the stat
would still count them, so the list would look merely incomplete rather
than wrong. A test asserts the ordering.
- A rename is `R100\told\tnew` — three fields. Taking field two records where
the file USED to be, naming a path nobody can open, and the bug is
invisible in any repo where nothing was renamed. `changed_paths` is now
shared with auto_merge (which had the same parse) and takes the NEW path,
with tests for renames and copies.
The list is capped at 500 paths with `files_truncated` beside it: a cap that
silently clips is worse than no cap, because "touched 12 files" and "touched at
least 500" would look identical.
Co-Authored-By: Claude Opus 5 <[email protected]>
The clump was structural, not cosmetic. Four causes, each fixed here.
`homes` (an agent's resting node) was written only by `seed()`, so every agent
homed to the mission centre and orbited the same dot regardless of which phase
it was on. `setHome` points each agent at its CURRENT phase, and the existing
physics does the rest for free: the pawn rests at its station, the pawn→home
line tethers it there, and a touch becomes a visible departure and return. No
new motion code.
The mission node was seeded as `level: "team"` (my own bug from the focus
work). `seed()` casts that straight to a Tier and `ensureNode` is
first-write-wins, so it was created as a small teal team dot that the later
`node.activity` could never upgrade. That dot at the centre of the scene was
one line.
`mission`/`phase` replace the retired `repo`/`loop` tiers rather than adding a
parallel set. The backend stopped emitting repo:/loop: ids, which left their
whole landmark treatment — bigger radius, distinct colour, always-labelled, 60s
fade instead of 22s — orphaned on prefixes nothing sends. Missions and phases
need exactly that treatment. The two separately-written prefix→tier ternaries
in onTouch and onNodeActivity are now one `tierFor`: they agreed only by luck,
and whichever path saw a node first fixed its tier forever.
Phase stations spring out at 190 rather than the shared 64, or they pack into a
rosette around the centre and the point — agents moving BETWEEN stations — is
invisible. Idle roam is off under a mission scope: wandering to a random node
keeps an idle workspace alive, but inside one mission it sends agents to files
nobody opened, which reads as work and isn't.
Palette is injected at construction and keyed on template_kind, so a benchmark
run and a security sweep no longer render identically to a research mission.
It also collapses two uncoordinated kind→colour maps that had drifted:
LEVEL_COLOR by tier, and the fireColor if-chain by id prefix.
Co-Authored-By: Claude Opus 5 <[email protected]>
Three bugs in one query, each hiding the next, plus one that made the whole
rich layer dead code.
`active_missions` joined `team_members` on `missions.team_id` — the LEGACY
pointer at the first minted team, superseded by the `mission_teams` junction in
0056. It was an INNER JOIN, and `mission_orchestrator::on_launch` deliberately
mints no team for a microVM mission, so the platform's primary execution tier
was dropped by a join and the World has been showing nothing at all for it. And
it selected only `status='running'`, while missions finish in minutes, so the
scene was empty almost always.
Now: join `mission_teams`, LEFT so teamless missions survive (their `agent_id`
is NULL and no pawn beams at them, which is the truth — nothing on this
platform ran that phase except a VM), and include missions finished in the last
24h carrying `status`/`template_kind` so the client can draw a finished map
instead of animating a corpse. `?mission=` scopes the feed server-side.
The whole phase plan now ships as `mission.phase`, including phases that have
not started: a phase list that appeared only as phases began made a five-phase
mission look like a one-phase mission until it was nearly over. Attribution
reuses `phase_runner::purposes_for` rather than copying it — two copies would
let the picture disagree with the machine about who is working on what, which
presents as a rendering bug and is really a lie.
Deleted the checkpoint tail. It read `topology_runs` keyed by an `agent_runs`
id; mission phases live in `topology_runs` under independently generated ids,
so it ran every poll and matched nothing, for every mission, forever. That is
why no mission has ever shown tool or file activity. Not repointed at
`topology_runs`: its only per-step content is the agent's own prose, and a tool
name in prose cannot be told from an agent talking about a tool. The a2a
`run_events` tail is kept — it genuinely works for the path that writes it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Seeding the name pick with the role index (0..n) started every crew at the top
of the pool and took the next free names, so the first mission after the switch
to per-mission crews hired Aarav, Abebe, Adaora, Adrian, Agnieszka. Unique and
correct, and transparently generated.
Seed from the claw's own uuid instead. UUIDv7 puts its random bytes LAST — the
leading bytes are a timestamp, which would cluster the same way — so the tail
is what spreads five picks across the whole pool.
Co-Authored-By: Claude Opus 5 <[email protected]>
Reverses the reuse added earlier, by operator decision. Reuse hired the
existing claw for a (template, slot) so the roster stayed at one team — but it
also meant every mission was staffed by the same five names, and the workforce
view showed one crew repeated down the page with nothing to tell the missions
apart. Distinct crews read better than a bounded roster.
The cost is the one reuse existed to avoid: claws are lifecycle='permanent'
and nothing reaps them until their MISSION is deleted, so the roster now grows
by the team size per mission. `agent_names::pick` keeps names unique
workspace-wide and degrades to a numeric suffix rather than colliding, and the
pool grew from 70 to 200+ given names so a workspace runs ~35 missions before
the first repeat. `reusable_claw` is kept in cm-db with its tests: this policy
has now flipped twice and the query is the hard part.
Also revives a test that had silently stopped running. An edit stranded
`runtime_data_is_scoped_to_one_mission`'s `#[test]` above its neighbour,
leaving two attributes there and none here — so the neighbour ran TWICE and
this one never ran at all. The total test count was unchanged by the fix
(291 before and after), which is exactly why a count is not evidence: rustc
had said "duplicated attribute" and "function is never used" all along, and
both read as ordinary warnings. The test guards per-mission `/zeroclaw-data`
isolation, i.e. one mission reading another's door token.
Co-Authored-By: Claude Opus 5 <[email protected]>
The roster query named tm.role_slot. That column is on agent_template_link;
team_members calls it plain `role`. These queries use untyped sqlx::query(),
so nothing caught it at compile time and the endpoint 500'd on its first real
request — the 401 an unauthed probe returns looks identical whether the SQL is
valid or not, which is why the payload had to be fetched with a real session
before believing the route worked.
Co-Authored-By: Claude Opus 5 <[email protected]>
Three things, all visible on the agents page.
**The roster looked like it was multiplying.** The sidebar flattened
orgs → companies → teams → agents, which renders a claw once per TEAM it
belongs to. Claws are reused across missions now, so a crew of five that had
run five missions appeared as twenty-five rows of the same five people. The
data was right and the view was lying. `GET /api/workforce` returns the roster
grouped by mission, and the tree renders each mission as a collapsible group,
so the repetition means something: the same colleague under each mission they
staffed. Claws on no mission come back under "Not on a mission" rather than
vanishing. The root now counts DISTINCT people, not rows.
**Agents were named after their jobs.** A team came back as planner, coder,
tester, reviewer, committer — the UI showed the same word twice (name on top,
role beneath) and the roster read as a stack of job tickets. New claws get a
given name from a deliberately wide pool (Amara, Vijay, Tomasz, Meredith…),
unique against the workspace roster AND within the team being minted. The role
is untouched in `job_title`, which is what the mission machinery binds on:
team_members.role_slot and the topology node carry the slot, so nothing
downstream keys off the display name. A reused claw keeps the name it had.
**Two latent reap bugs found while investigating a leak that was not one.**
Containers of completed missions are removed by `spawn_sweeper` after a
30-minute grace, and it works — an earlier report of leaking containers was me
reading that deliberate grace as a bug. But:
- the sweeper cleared the runtime binding even when teardown FAILED, and it
selects on `runtime_endpoint IS NOT NULL`. One transient docker error would
therefore hide a surviving container from the only thing that would retry
it, permanently. It now asks docker whether the container actually
survived: gone means clear, still there means keep the binding and retry —
which closes the orphan path without reintroducing the infinite retry the
original comment was guarding against.
- `set_runtime_binding` discarded rows_affected, so a mismatched workspace
updated nothing and returned Ok. The binding is how the sweeper finds a
container; a silent no-op there leaks one with no record of anything wrong.
Co-Authored-By: Claude Opus 5 <[email protected]>
`provider_alias_for` still documented the judge as deliberately sitting on
`anthropic.judge`/API key, so a subscription throttle would degrade missions
while verification kept working. That credential is an account with a zero
balance — driving the real path returns 400 "Your credit balance is too low" —
so the comment pointed the next reader at something that cannot answer.
`anthropic.default` and `anthropic.judge` are retired from the runtime config
and every agent that named them was repointed onto a live credential. The
independence argument that put the judge there still holds; it is now served by
a different FAMILY rather than a different key — CLAWMATES_VALIDATOR_MODEL is
glm:glm-4.7 on gw-04, and cross_provider_judge already refuses a validator in
the implementer's own family.
Config-side (gw-04, not in this repo): 755 -> 246 lines, 128 -> 14 agent
blocks, after sweeping 19 [agents.claw_<uuid>] corpses — every one verified
against agents WHERE deleted_at IS NULL. Nothing reaped those, and the file is
byte-copied into every mission.
Verified: scout/judge/worker_kimi/worker_glm each answer on their new
provider, and multirole passes 4/4 against the reorganized config.
Co-Authored-By: Claude Opus 5 <[email protected]>
A throttled subscription had nowhere to go. `claude_cli.default` carried no
`fallback`, and neither target alias was declared — they existed only as
commented-out examples. Forcing a 429 with a shimmed `claude` surfaced four
defects that all read as correct config and do nothing:
- a `[providers.models.<f>.<a>.env]` SUB-TABLE is parsed then silently
ignored ("fields must live directly under ..."). This block was already
live for claude_cli.default, so the token injection has been inert. For
the glm alias it would have dropped the z.ai routing AND the clearing of
CLAUDE_CODE_OAUTH_TOKEN — credentials crossing between providers.
- an empty `[providers.models.kimi_cli.default]` is skipped at runtime.
- a claude_cli alias used as a fallback needs a non-empty `api_key` to pass
FamilyProviderFactory's default readiness gate, even though the provider
ignores the key and authenticates through `env`. Absent it the agent dies
at STARTUP, which takes out every mission, not just throttled ones.
- timeout_secs=600 capped every turn under the 3600s TURN_TIMEOUT from
4c418f7, so that raise bought long turns nothing.
The kimi hop then 401'd: `kimi_cli` spawns the `kimi` binary, which rejects a
Kimi Code key. Kimi is reached the way the agent-kimi microVMs already reach
it — the `claude` binary against api.kimi.com/coding (no /v1; Claude Code
appends it). So the hop is `claude_cli.kimi`, and `kimi_cli.default` is left
declared but out of the chain so the finding stays visible.
kimi-home joins SEEDED_PATHS: each claude_cli fallback alias needs its own
HOME with a .claude.json, and separate homes stop two concurrent fallbacks
from sharing one Claude Code session directory.
Measured on gw-04 against the live config, zeroclaw 0.8.4:
throttled primary -> OK (chain fires) happy path -> OK (no regression)
Isolating each hop: kimi alone answers, glm alone answers, and with an empty
fallback the turn fails `rate_limited` at phase=http_response — the control
that makes the other two mean something. An earlier OK came from glm, not
kimi, so the reply alone was never evidence.
Co-Authored-By: Claude Opus 5 <[email protected]>
A turn is an agent LOOP, not one model call. Each call inside it is bounded
separately by the daemon — `claude_cli`'s `timeout_secs`, 600s on gw-04,
verified in the live config — so TURN_TIMEOUT has to cover however many calls
the loop makes, not one of them. It was 700s.
MEASURED: a healthy research turn is ~157s. A throttled one blew the budget with
one slow call plus a second, and the executor killed it at 11m43s with no error
from the daemon, because nothing had failed yet. The operator got
"turn executor failed: turn timed out" and the container holding the reason was
torn down minutes later.
An hour matches the phase's own budget. A genuinely stuck CALL is still caught
at 600s by the daemon and surfaces as a real error; this only stops us killing
turns that are working, slowly.
Co-Authored-By: Claude Opus 5 <[email protected]>
A research_and_code mission failed with:
turn executor failed: turn timed out
turn executor failed: turn timed out
and that is the entire record. Investigating it found: the run produced zero
steps and zero output, it died at exactly 700s (TURN_TIMEOUT), the node→claw
aliases were bound correctly, and the same zeroclaw team path passes in the
`multirole` scenario. So the platform path is fine and the agent simply never
finished a turn — but the one place the reason lived, the per-mission runtime
container, is torn down after the phase and takes its log with it. By the time
anyone looks, all that survives is the string.
The timeout now reads the last 40 lines out of that container while it still
exists, and reports which agent alias and which gateway it was driving.
Best-effort by construction: it runs on a path that is ALREADY failing, so a
docker error there degrades to a note rather than replacing the real failure
with a second one.
`container_name` derives the container from the gateway URL and returns None
rather than guessing, because this feeds a diagnostic — a wrong name would put a
different container's log under a failure and send the reader somewhere else
entirely.
Co-Authored-By: Claude Opus 5 <[email protected]>
Every zeroclaw mission minted a fresh team of claws. They are created
`lifecycle = 'permanent'` and nothing reaps them until the MISSION is deleted,
so the roster grew by a whole team per mission while each member worked exactly
once — "My Workforce" was a list of strangers, and upskilling had nothing
durable to act on.
A mission now hires the claw that already does the job, matched on
`agent_template_link (template_id, role_slot)`, minting only what is missing.
Oldest first, so reuse concentrates on the same few claws and their brains
actually accumulate rather than spreading thinly across a growing pool.
A claw on a RUNNING mission is not offered. Two missions driving the same
ZeroClaw agent and the same `.brain` at once is a data race with a model on the
other end of it, and minting a second claw is much cheaper than reasoning about
that.
A reused claw is NOT re-seeded from the template's brain_seed — that would
overwrite what it learned with its starting point, which is precisely the
accumulation this exists for.
Names are the role now (`planner`), not
`"{mission} · {purpose} · {template} · {slot}"`. That produced
"verify: a repo-less research mission keeps its output · mission · Rust SDLC ·
planner" — unreadable in the roster, the API and every log line at once. Which
mission a claw is on is context a caller can join to; it is not its name.
And the half that makes reuse safe rather than destructive: deleting a mission
now purges only claws no OTHER mission still employs. Without it, tidying up one
mission deletes staff another one holds — presenting as the roster quietly
shrinking rather than as an error. A test asserts the guard exists inside the
reaper AND runs before the purge, because a check after it is decoration.
Co-Authored-By: Claude Opus 5 <[email protected]>
The research-vm scenario failed on its first run, and said exactly why:
mission has no checkout at /var/lib/clawmates-missions/<id>/repo
— a microvm phase needs a repository
`phase_runner` refuses upstream of both places the last commit fixed. Three
guards, written independently, all encoding "a microVM phase implies a git
checkout" — which is why the capture filter could cite it as settled fact.
A repo-BACKED mission with no checkout is still a real fault and still refused;
booting a VM to hand the agent an empty directory would turn a setup failure
into a confusing agent report. A repo-LESS one now gets the empty workspace
made here, so the executor's inject has something to pack.
Co-Authored-By: Claude Opus 5 <[email protected]>
Two halves, and the first was worse than the plan assumed. `run_phase_in_vm`
packed `<missions_root>/<mission>/repo` unconditionally — a directory a
repo-less mission does not have — and then required `/mission/repo/.git` inside
the guest before spending a turn. So a repo-less microVM phase did not merely
go uncaptured: it failed before the agent ran.
A repo-less mission now gets an EMPTY workspace at the same guest path, created
host-side so the collect unpacks back over it with no special case, and the
readiness probe asks for what was actually sent — the directory rather than a
`.git` that was never going to be there.
`mission_outputs` then drops its `runtime_kind <> 'microvm'` exclusion, whose
stated reason ("a microVM mission always has a checkout") is exactly what
stopped being true. Where the files come from now depends on the runtime, and
the difference is not cosmetic: a container mission's output is still inside a
running container, while a VM's has already been unpacked onto the host by the
end-of-turn collect. Asking docker for a VM mission's files would query a
container that never existed.
The recursive copy skips symlinks rather than following them — a link out of
the tree would publish whatever it points at.
`research-vm` is the proof, added to the suite as well as the dispatch: the same
assertions as `research-only` with `runtime_kind: microvm`. A scenario nobody
runs is a scenario that does not exist.
Co-Authored-By: Claude Opus 5 <[email protected]>
Two endpoints the wizard redesign needs.
`POST /api/missions/refine-draft` — the polish button fires while the user is
still typing, before anything is created, so it has no id to route on.
`refine` deliberately requires a saved draft because its Accept writes back;
this one has nothing to write back to and returns the text. Same system prompt,
same model chain. The phase list comes from the workflow recipe rather than the
caller, for the same reason `phases_for_create` prefers it: a client that
guessed would have the model write acceptance criteria for phases the mission
will not run.
`GET /api/missions/{id}/artifacts/{artifact_id}/download` — the file itself.
`artifact_content` caps at 2 MiB and reads as UTF-8, so a large or binary
artifact is unreachable by any means today; this streams the bytes with a
filename attached and no ceiling.
Both artifact routes now resolve through ONE containment check. Two copies of
"is this path under _outputs" is two chances for one of them to be the lenient
one, and the lenient one is an arbitrary read of the gateway's filesystem — so a
test asserts there is a single resolver and that both routes call it.
The download filename was chosen by an AGENT and lands in a header every browser
parses, so quotes, backslashes and control characters are stripped rather than
escaped; the test covers a header-injection attempt.
Co-Authored-By: Claude Opus 5 <[email protected]>
Clearing the fleet's four leftover agents returned 404 on every one. They had
been soft-deleted back in June — correctly invisible in the UI ever since — and
`agents::get` filters `deleted_at IS NULL`, so `workspace_agent` could not find
them. Every route uses it, including `batch-delete`, the one that exists to
HARD-purge. So a soft-deleted agent was unreachable from the application
entirely and its row stayed forever.
`get_any` sees them, and only the purge path uses it: hiding soft-deleted rows
is right for every read, and wrong for the one operation whose whole job is
removing them. Written with `query_as` rather than the checked macro so it does
not force an offline-cache regeneration on every machine that builds this.
`fleet-reset.sh` now uses `batch-delete` for agents rather than
`DELETE /api/claws/{id}`. The latter is a SOFT delete, so pointing a reset
script at it would have quietly added to the pile it was meant to clear.
Co-Authored-By: Claude Opus 5 <[email protected]>
`default` is the generic `rootfs.ext4` and `claude` is the named one, and
`microvm_credential_for` gives them the identical contract — so the list came
back with both under the same label, and whichever a user picked they got the
same thing. Collapsed to the named one where it exists; the generic keeps a
label of its own for a fleet that only has that.
Co-Authored-By: Claude Opus 5 <[email protected]>
The full harness found it — 12 of 13 scenarios green, `roster` red:
roster: the planner sized this mission at 2 member(s) PASS
roster: the approved roster is on the mission (2 nodes, composed) PASS
roster: this run added 1 line(s) for a 2-member roster FAIL
topology_runs.error: turn executor failed: node n1 in a microVM:
vm_create failed: no rootfs for backend "canary-claude" on this node
The roster proposed `verifier@canary-claude`. Placement asked
`online_for_backend` about the MISSION's backend — `claude` — and architect
answered, holding `claude` and `local-ornith`. The graph's first node ran and
delivered, the second could not boot, and the mission finished half-done. The
question placement asked was true and insufficient.
A composed graph runs on ONE node, so that node needs every image its nodes ask
for. `required_backends` collects the mission's plus each
`config.roster.nodes[].attrs.backend`, and `online_for_backends` passes the
whole set to the same jsonb `@>` — containment already means "contains ALL of
these", so the query shape did not have to change, only what it was asked.
This is the failure mode the roster feature creates by existing: its entire
purpose is putting a verifier on a different provider, which is exactly what
makes one node insufficient. Nothing before the full suite had a reason to
exercise it — the composed scenario uses one backend for all five nodes.
`NoCapableNode` now names the set and says why one node must hold all of them.
Co-Authored-By: Claude Opus 5 <[email protected]>
Everything built today — Firecracker missions, the four backends, the local GPU
one — was unreachable from the dashboard. The wizard offered `zeroclaw` and
`local_herdr` and nothing else, so a mission created in the UI could not be a
microVM mission at all, and `local-ornith`/`glm`/`kimi` were API-only. Testing
"our workflows in the UI" would have exercised none of it.
Adds the runtime option and a backend picker, fed by a new
`GET /api/fleet/backends` that returns `mission_roster::available_backends`
verbatim — the SAME list the roster planner is handed, not a second one. Its two
rules are both load-bearing and neither is visible from a node's capabilities
alone: the image must be built on an online node, and the backend must have a
credential contract. `agent-terminal` passes the first and fails the second —
bootable, with nothing for the agent inside to authenticate with — so offering
it would produce a mission that validates, launches, and dies at the agent turn.
Ids are deployment vocabulary, so the picker labels them: a user choosing
between `local-ornith` and `canary-claude` should not have to know which company
each one bills. An empty list says why (no rootfs built) instead of showing an
empty dropdown, and no node is chosen for a microVM mission because
`vm_placement` picks it per phase.
Co-Authored-By: Claude Opus 5 <[email protected]>
`cleanup_sweeper` prunes ROWS. Deleting a row has never deleted a directory,
and `teardown_container` only runs while a mission still exists to tear down —
so a mission removed by any path that skipped teardown left its tree behind
permanently, on the smallest disk in the fleet (150 GB, shared with postgres and
every checkout). 106 mission directories are sitting there now.
Filesystem-first, deliberately: the DB is the PREDICATE, never the enumerator.
Enumerating from the database is exactly how these became invisible — a
directory whose row is gone is the one a row-driven sweep cannot see.
Three reapers, one deletion path. Orphan mission dirs (no row, past a 2h grace),
scratch trees (_bench/_gate/_verify/_merge past 6h — all four have leaked
before), and _outputs past 90d, whose artifact rows are marked only AFTER the
files are gone, because the other order claims artifacts are reaped while they
are still on disk.
The single removal path escalates: the server is uid 65532 and cannot delete
what the per-mission daemon leaves as root, so PermissionDenied falls back to
`root_copy::purge` and shouts if the tree survives even that. A GC that cannot
collect is the thing being fixed, so failures are counted and reported, never
swallowed.
Guards worth naming: `_cargo` is a SHARED cache every mission writes to and
lives under the same root, so an underscore-prefixed sibling treated as an
orphan mission would delete it out from under running work and look like a slow
cargo build. Only a well-formed mission id is ever a candidate — a directory
whose name is not an id can have no row by construction, so without that gate
every unrecognised directory looks orphaned.
Co-Authored-By: Claude Opus 5 <[email protected]>
Deployed, and the report simply did not appear — from the tool built to stop
things failing silently. Two causes, both worth keeping:
There was no timeout anywhere in the probe, so one slow provider swallowed the
entire report. Each link is now bounded at 60s (generous: `complete_or` spends
up to 30s in its own backoff, so a tighter cap would report a merely throttled
link as hung) with `TimedOut` as its own state, and every line is emitted AS IT
RESOLVES rather than collected and printed at the end — a later link that hangs
must not be able to hide the ones already checked.
The first attempt at the timeout awaited the probe and then wrapped the result:
let probe = complete_or(...).await;
timeout(PROBE_TIMEOUT, async { probe }).await
That compiles, reads correctly, and bounds nothing. The timeout has to wrap the
future.
Co-Authored-By: Claude Opus 5 <[email protected]>
Committed and deployed while its test compile was failing: the verify step was
`cargo test | grep -E "^error|test result" && git commit`, and grep exits 0 when
it MATCHES, so finding the error is what let the commit proceed. The library
built fine, so the deploy was sound, but the check that was supposed to gate it
did the opposite of gating.
The error itself was a borrow in a test closure; a plain fn fixes it.
Co-Authored-By: Claude Opus 5 <[email protected]>
fallback chain (6 link(s), 4 usable):
claude-opus-4-8 ok
claude-sonnet-4-6 ok
claude-haiku-4-5-20251001 ok
kimi:kimi-k2.7-code BROKEN: 400 ... role 'system' must not be empty
glm:glm-4.7 ok
local:ornith-fleet:9b UNREGISTERED — resolves to the DEFAULT provider
Neither link was actually broken.
The probe sent an EMPTY system prompt, which Kimi rejects outright. A probe has
to look like the traffic it stands in for, or it measures itself.
The second is the one worth keeping. `resolve_provider` returns a spec unchanged
when it does not recognise the provider, and the part after the FIRST colon when
it does — so the obvious test, "does the model half still contain a colon",
reads correctly and is wrong the moment a model id has one. `ornith-fleet:9b`
has one. The probe reported a provider the server had just finished registering
as UNREGISTERED.
`evaluator::cross_provider_judge` had the identical check, and would therefore
have refused a local judge as "not independent" — silently falling back to a
same-family one, which is the exact claim that path exists to make honestly.
Both now compare against the whole spec.
That bug was written into the codebase before a model name with a colon existed,
was correct at the time, and became wrong when one arrived. Nothing would have
reported it; a boot-time probe of every link did, on its first run.
Co-Authored-By: Claude Opus 5 <[email protected]>
opus -> sonnet -> haiku -> kimi -> glm -> local. The order is capability first,
then independence: three Anthropic tiers on one account (a throttle usually hits
a tier, so stepping down often clears it), then two separately funded accounts
(now an outage, not just a throttle, is survivable), then our own GPU (nothing
left to be down). Every id was probed on this deployment and answered 200.
The preflight is the more important half. Configured is not working, and this
chain has a specific way of lying: `resolve_provider` falls back to the DEFAULT
provider when it does not recognise a provider name, so a typo in `kimi:` does
not error — it quietly runs on Anthropic, and a chain that reads as three
accounts is really one. A reachability-only probe calls that link green.
So `preflight` checks resolution and reachability separately, eight tokens per
link through the REAL call path, and reports four states. `Throttled` is
deliberately not a failure: a 429 means the spec resolved, the credential
authenticated, and there was no capacity this second — the exact condition the
chain exists to route around, and painting it red would train an operator to
ignore red. `Unregistered` and `Broken` are failures, and they get different
words because they need different fixes.
It runs at boot alongside validator_preflight and runtime_preflight, spawned so
it cannot delay startup. A chain is the one piece of infrastructure nobody looks
at until the day it has to work, so it is now checked on the days it does not.
Co-Authored-By: Claude Opus 5 <[email protected]>
`local:ornith-fleet:9b` joins opus -> haiku -> glm as the final link. Every
entry above it depends on somebody else's account staying funded and
unthrottled; this one depends on a GPU in the next room. It is last because it
is the weakest model, and present because a chain whose every link is external
is not a fallback chain, it is one outage in a trench coat.
Three small changes make it work:
- `build_provider_registry` accepts a provider with an empty `api_key_env`.
A model on our own hardware has nothing to authenticate to, and the old
behaviour SKIPPED a keyless provider — leaving the chain quietly one link
shorter than it reads, which is the failure mode this whole area keeps
producing.
- `provider_family` learns `ornith`/`ollama` for BARE names. A qualified
`local:` spec was already answered by the split, but a bare one fell through
to "unknown", and `cross_provider_judge` would then refuse a judge that is
genuinely a different family from the Anthropic implementer.
- A test pins that the last link survives `resolve_provider`'s split-on-FIRST-
colon: `local:ornith-fleet:9b` is provider `local`, model `ornith-fleet:9b`.
Splitting on the last colon would ask for a provider named
`local:ornith-fleet`, and the symptom would be a silent fall back to the
default provider.
Infra: Ollama on tank and architect now binds 0.0.0.0 so the gateway (which has
no GPU) can reach it. `tailscale serve` cannot — Ollama rejects a non-local Host
header as a DNS-rebinding guard and OLLAMA_ORIGINS is CORS-only, so it 403s.
0.0.0.0 still includes loopback, so the microVM vsock pipe is unaffected;
verified on both nodes. This is an explicit trade: Ollama has no auth and its
API can pull and delete models, so it is now reachable from the LAN as well as
the tailnet. The drop-in carries the ufw one-liner to close the LAN side.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude Code pointed at the Ollama already installed on every GPU node. Ollama
has served a native Anthropic-compatible /v1/messages since v0.14, so this is
an env contract rather than a translation layer — the fourth variation on the
same idea as agent-glm and agent-kimi.
The route is NOT the egress proxy, and that is the design. `egress` speaks
CONNECT, takes a destination from the guest, resolves it and decides; every one
of those powers is a liability, which is why it refuses non-443 ports and IP
literals after a unit test caught them being bypassed. Routing a local model
through it would have meant relaxing both.
`local_model` is the opposite shape: there is no destination in the protocol.
fcagent listens on guest 127.0.0.1:11434 and pumps to vsock 9003; the node
splices that onto its own 127.0.0.1:11434 and copies bytes. A compromised guest
cannot redirect it because there is nothing to redirect — it is a pipe, not a
proxy, and strictly narrower than anything an allow-list could express. The
bytes never touch a network, so there is no wire for TLS to protect, and Ollama
stays bound to loopback rather than being exposed on the tailnet.
The socket is bound only for a backend declared to use a local model, so a
`local-ornith` VM reaches the forge through egress and nothing else, while every
other backend's guest port simply refuses. Both halves have negative controls.
`scripts/fleet-model-setup.sh` exists because of one measurement: stock
ornith:9b reported input_tokens=2050 for a 48000-word prompt and answered as
though nothing had been dropped. Ollama's default window is ~2K whatever the
model card says, and it truncates silently — the exact failure an agent turn
would hit and never report. The script pins num_ctx=131072 into a derived tag
and then PROVES both the window and tool calling before declaring success.
Verified on architect: ~65536 words -> 65604 input tokens, stop_reason=tool_use.
Placement needs no new capability key: building the rootfs only on GPU nodes
means `nodes::online_for_backend`'s existing `rootfs @> ["local-ornith"]`
predicate does the affinity, so morpheus never offers the backend.
Co-Authored-By: Claude Opus 5 <[email protected]>
The server runs as uid 65532, so `remove_dir_all` on a mission directory
returns PermissionDenied the moment anything root-owned is left in it — and the
old code logged that at the same level as "file not found" and moved on. The
directory then lived forever.
After the seed-copy fix a mission holds 3281 files owned by 65532 and 26 owned
by root: `.claude.json` and the session jsonl the per-mission ZeroClaw daemon
writes itself, after the copy has been chowned. Twenty-six files is small
enough to keep every mission directory alive without anyone noticing why.
PermissionDenied now falls back to `root_copy::purge`, which deletes from
inside the container as root — the same escape hatch `container_exec` keeps for
exactly this, clearing debris a root process created. And if the directory
survives even that, it says so, because a cleanup that silently failed is the
thing being fixed.
Removing the last 26 properly means running the per-mission daemon as 65532,
which needs `/mission` pre-created in the image with that ownership — the
daemon creates it at boot today and cannot at a lower uid. That is an image
change, deliberately not bundled here.
Co-Authored-By: Claude Opus 5 <[email protected]>
Running the seed copier as 65532 broke mission launch, and broke it quietly.
The seed dir is root-owned with parts at mode 0600 (`.claude.json`,
`clawmates-mcp.json`), so uid 65532 cannot READ them: `cp` failed on the first
unreadable entry, `set -e` abandoned the rest, and the mission came up with a
runtime-data holding `.zeroclaw` and nothing else — no Claude credentials, no
door config. The daemon then never created its agents' workspace, and the phase
failed 200 lines later on "Could not find the file /mission in container",
which points nowhere near the cause.
It was quiet because `seed_runtime_data` polled for the container to STOP and
returned Ok without ever reading its exit code. A copier that died on a
permission error and one that finished cleanly were indistinguishable. It now
reads the status and says what went wrong.
So: root for the read, `chown -R 65532:65532 /dst` for the result. Both halves
matter and they pull opposite ways — root is needed to read the seed, and
65532 is needed because everything else in the missions tree is 65532 and a GC
running as 65532 cannot delete what root left behind.
Co-Authored-By: Claude Opus 5 <[email protected]>
The uid fix landed and the CHECKOUT came back completely clean — 0 non-65532
files under `repo/` after a benchmark run that builds and tests Rust. But the
same mission still held 3247 root-owned files, all under `runtime-data/`.
`seed_runtime_data` spawns a throwaway container to `cp -a` the runtime seed
into the mission's directory and never set `user`, so it ran as root — the
identical absent-`user` omission `container_exec` had, in a container create
instead of an exec. The seed source is 65532-owned and the destination is
created by the server (which itself runs as 65532), so the copy never had a
reason to out-rank either.
This is the tree a gateway GC has to be able to delete, and a GC running as
65532 cannot remove root-owned files — the cleanup-that-cannot-clean-up shape,
found before writing the GC rather than after.
Co-Authored-By: Claude Opus 5 <[email protected]>
`CreateExecOptions` never set `user`. Not a wrong value — an ABSENT one: the
daemon defaults to root, and twelve callers inherited that without any of them
choosing it. That single omission is the origin of four separate patches —
root-owned `target/` directories inside a checkout owned by 65532, `root_copy`
existing at all, and a cleanup that had to re-enter the container as root to
undo its own mess.
The rule is positional and lives in ONE place: an exec whose workdir is inside
`missions_root()` runs as 65532; anything else (preflight probes, image checks)
keeps the daemon default so unrelated call sites cannot break. Twelve callers
each remembering to pass a uid is twelve chances to forget, and the one that
forgets leaves debris the other eleven cannot remove.
Non-root needs an environment the image does not provide. Measured in the
deployed image: uid 65532's HOME (/zeroclaw-data) and /usr/local/cargo are both
root-owned and unwritable, so this would otherwise break every cargo call — the
benchmark runner, the judge's sandbox, the delivery test gate — far more quietly
than the leak it fixes. The missions root IS bind-mounted and writable by 65532,
so HOME/CARGO_HOME move there and the cargo cache is shared across missions
rather than re-fetched per mission. Verified on gw-04: a clean `cargo build` as
65532 with those three variables produces output owned entirely by 65532.
Root remains reachable only through `exec_as_root`, whose name says so, and
which exists solely to clear debris earlier root execs left. `runtime_preflight`
now probes the whole policy at boot, so an image that moves or tightens that
mount fails loudly instead of failing every cargo call for a reason no error
message would connect to a uid. evaluator_tools' inlined fourth copy of the
purge is replaced by `root_copy::purge`.
Co-Authored-By: Claude Opus 5 <[email protected]>
drain-midmission passed 3/3 twice, but the "re-placing this phase" line the
last commit added never appeared in the log. It cannot: `online_for_backend`
filters on `status = 'online'`, so a draining node is not a candidate, never
reaches `unfit`, and the pin simply falls through to ranking — on the old code
as well as the new.
So the scenario passes either way and proves the affinity decision, not the
`TargetUnfit` bug. The path that genuinely used to fail a phase is "the
previous phase's node has since FILLED UP": that puts it in `unfit`, which
returned a non-transient error, which never reached the queue. That is what the
unit test now says, in place of a claim about draining the harness does not
support.
The accidental mission-to-node affinity was real and unconditional either way.
Co-Authored-By: Claude Opus 5 <[email protected]>
`drain-midmission` found this. `choose` treated its `want` argument as a hard
requirement, and the only caller passes `missions.target_node_id` — which is
not an operator's choice, only where the PREVIOUS phase happened to run. Two
consequences, both wrong:
- A node drained or filled between phases produced `TargetUnfit`, which
`is_transient()` says false to, so `phase_runner` FAILED the phase rather
than queueing or moving it. The queue silently did not apply to the second
phase of any mission.
- While the node stayed fit, every later phase went straight back to it
regardless of ranking — accidental mission-to-node affinity, which this
module's own header says must not exist.
Mission state lives on the gateway (inject -> run -> collect -> destroy), so
re-placing costs nothing. The pin is now advisory: preferred while it fits,
and when it does not, the reason is logged and ranking proceeds. `TargetUnfit`
is deleted rather than left unconstructed, so it cannot come back as a
non-transient failure by accident.
The scenario had its own race: it waited for phase 0 to COMPLETE before
draining, but warm phases finish in ~80s against a 10s placement sweep, so
phase 1 was often already placed — and the run then blamed the platform for
running on a node that was not yet drained. It now drains while phase 0 is
still running, which does not disturb a live VM and is the more faithful test.
Co-Authored-By: Claude Opus 5 <[email protected]>
`mission_workspace::missions_root()` is now the only place that answers "where
does mission state live". It had fragmented into five: this function, private
`env::var("CLAWMATES_MISSIONS_ROOT")` copies in security_scan, benchmark_runner
and mission_outputs, and a hardcoded `MISSIONS_HOST_ROOT` const in
mission_runtime that read no env at all.
They agree on the deployed value, so nothing has broken. The risk is entirely
in what comes next: anything that sweeps or reclaims this tree has to be
sweeping the same tree the writers use, and five definitions cannot promise
that — a reaper written against one would silently leave the others' directories
behind forever, which is how the orphans got there in the first place.
A source-walk test fails any module outside `mission_workspace` that reads the
env var itself.
Co-Authored-By: Claude Opus 5 <[email protected]>
The capacity harness scenario, on its first full run, caught what it was
written to catch:
capacity: architect peaked at 6 of 6 slot(s)
FAIL capacity: 'morpheus' peaked at 3 concurrent VM(s) with only 2 slot(s)
capacity: tank peaked at 6 of 6 slot(s)
PASS capacity: the over-capacity missions QUEUED
PASS capacity: all 16 queued/placed missions completed
`capacity_of` inferred the host's own footprint by subtracting the VMs' FULL
8 GiB claim from observed usage — which assumes they have already consumed it.
A VM booted seconds ago holds about an eighth. On morpheus (31757 MiB total,
4314 MiB idle, 2 slots) with 2 young VMs at ~6314 MiB observed, the inference
6314 - 16384 goes negative, clamps to the 2048 floor, and invents 2266 MiB —
exactly enough for a third VM on a two-slot node.
The footprint is only honestly MEASURABLE when nothing is committed, so
remember it then: `nodes.mem_baseline_mib`, sampled by `survey` whenever it
observes an idle node with fresh health. When VMs are committed, take the
LARGER of the remembered reading and the old inference — a node that was once
idle at 4 GiB and is now running a 20 GiB build must not be scored as idle,
which would be the same over-commit arrived at from the other direction. Both
directions have a test; the second is the one that would otherwise rot.
Raising HOST_BASELINE_FLOOR_MIB would have made this one node's numbers pass
and drifted the moment the fleet changed shape.
Co-Authored-By: Claude Opus 5 <[email protected]>
Pulled forward from the observability phase because the capacity harness
scenario needs it. A test that recomputed the slot arithmetic in bash would
drift from `vm_placement` and then agree with itself while the scheduler did
something else — the same shape as every silent-success bug in this codebase.
Returns `survey()` + `rank()` unmodified, and keeps `unfit` as its own list:
"the fleet is full" and "we could not read the fleet" send an operator to
different places, so they must not be summed into one number.
Co-Authored-By: Claude Opus 5 <[email protected]>
The research scenario passed 4/4 and the log underneath it said:
phase_summarizer: ... failed: anthropic 400 Bad Request: "Your credit
balance is too low to access the Anthropic API"
`phase_summarizer` and `mission_refiner` each built their own reqwest POST to
the Messages API with `x-api-key: $ANTHROPIC_API_KEY`. No audit of `.complete(`
call sites could have found them — they never touched a provider — so every
phase summary and every mission-brief refinement on this deployment had been
failing against an empty account while the phases themselves ran fine. The
summarizer even persisted an error row per phase, which is why nothing ever
retried loudly enough to notice.
Both now go through `subscription::complete_with_fallback`, so they inherit the
subscription-first credential choice, the 429 backoff, and the opus -> haiku ->
glm chain. The summarizer records the model that ANSWERED in
mission_phase_summaries.model rather than the one it asked for.
The guard is a source WALK, not a file list: any .rs under cm-api/src that
mentions the Messages API host or `x-api-key` fails the test. A hand-listed set
of files is exactly what let these two hide.
Co-Authored-By: Claude Opus 5 <[email protected]>
`Runtime::judge` returns "governor unreachable (fail-open)" whenever the
provider never answers, and the door caller drops `reason` on every allow —
so a judge model that is rate limited or uncredited turns the governor into a
rubber stamp with nothing anywhere saying so. Fail-open stays (a governor
outage must not halt agents), but it is now loud.
Found while removing the metered key as a dependency: the governor reads
CLAWMATES_JUDGE_MODEL, which was `claude-opus-4-8` — a model that is 429 on
this deployment's subscription. gw-04's .env now points it at `glm:glm-4.7`,
matching CLAWMATES_VALIDATOR_MODEL: funded separately, uncapped, and a
different family from the agent it judges.
Co-Authored-By: Claude Opus 5 <[email protected]>
Two changes so an empty metered account stops being a platform outage.
1. `build_provider` prefers the subscription token over ANTHROPIC_API_KEY.
A bare model name resolves to whatever this returns, so making it the
subscription means no server-side call can reach the metered key by
construction — rather than by a source-grep test that already missed four
call sites once. The metered key remains a fallback and now warns loudly
when it is the one in use; boot no longer requires it at all.
2. `complete_with_fallback` walks a declared chain when a model has no
capacity: opus -> haiku -> glm:glm-4.7 by default, overridable via
CLAWMATES_MODEL_FALLBACK, empty to disable. Measured on gw-04 today: opus
and sonnet return 429 on the subscription while haiku, GLM and Kimi all
return 200, so a capped window no longer means "the planner is gone".
The chain returns the model that ANSWERED, and every caller persists it —
mission_plan_proposals.author_model, mission_team_proposals.author_model, and
the swarm's step role. A plan drafted by the third link and filed as an opus
plan is a silent quality change, which is the failure shape this project keeps
paying for. Two negative controls hold the design: the chain never retries the
model that just failed as its own fallback, and it steps down ONLY for a
capacity failure — walking it on a malformed prompt would ask three models the
same bad question and report the third one's confusion.
Co-Authored-By: Claude Opus 5 <[email protected]>
The test that was supposed to prevent this grepped for the literal
`runtime.complete(` and passed while the phase planner (`mission_plan.rs`),
both swarm calls, and a second enhance path in `claws.rs` still billed the
pay-as-you-go account. They spell the receiver `state.runtime` or wrap the
call across lines, so the receiver name was never the thing to match. The
test now matches the METHOD, and covers all five files.
`complete_or` gains the rule that makes it safe to apply everywhere: a
`name:model` spec is an operator's explicit provider choice — the swarm
worker model is configured exactly that way — and is passed straight to
`Runtime::resolve_provider` untouched. Only a bare name is ambiguous, and a
bare name is precisely what resolves to the default provider. Hijacking a
chosen Kimi or GLM model onto Anthropic would be the same silent-substitution
bug pointed the other way.
`validator_preflight` and the evaluator judge keep calling the runtime
directly, on purpose: both exist to exercise the CONFIGURED spec.
Co-Authored-By: Claude Opus 5 <[email protected]>
The retry landed and still failed: all four attempts returned 429. A bare
16-token probe with the same token, straight from gw-04, also returned 429
with `x-should-retry: true` — the Claude Code subscription itself is limited
right now, and no amount of backoff inside one HTTP request will outlast it.
So stop pretending it is a server bug. New `ApiError::Unavailable` → 503,
carrying the one sentence the operator can act on ("clears on its own; try
again shortly"), instead of an opaque `internal error` that sends them into
the logs. The harness now prints the response body rather than the generic
"the planner produced no usable proposal", which is what hid both walls —
first the credit balance, now this.
Co-Authored-By: Claude Opus 5 <[email protected]>
Moving the roster and planner onto the subscription removed the credit wall
and revealed the next one: the harness went from
`400 credit balance too low` to `429 rate_limit_error`. A one-shot proposal
call had no retry — there is no retry convention anywhere in cm-llm — so a
limit that clears in seconds killed the "propose a team" button outright.
Four attempts, 2/8/20s backoff, and only for errors that can actually clear:
429/5xx/transport. A 400, 401 or 404 returns immediately, because retrying
those is a 30s hang ending in the identical message, which reads to an
operator as a stall rather than a bad request.
Co-Authored-By: Claude Opus 5 <[email protected]>
The roster planner died with `400 — "Your credit balance is too low to access
the Anthropic API"` while every mission on the same machine kept running. Two
Anthropic credentials reach this server and they bill differently:
`ANTHROPIC_API_KEY` (sk-ant-api, metered, runs out) and the Claude Code
subscription token (sk-ant-oat) that every VM already uses.
`Runtime::complete` with a BARE model name — "claude-opus-4-8" — resolves to the
default provider, which is the metered key. Three server-side callers did that:
the roster planner, the Master Planner, and the claw enhancer. Missions were
never affected because `mission_runtime` deliberately sends only the
subscription token into a guest; the server had no equivalent rule.
`subscription::complete_or` is now that rule, and it is the ONE place a
subscription token becomes a provider — `evaluator::subscription_judge` had its
own copy, and two of them is how one ends up with a prefix check the other
lacks.
The `sk-ant-oat` prefix is checked rather than the variable name trusted: an
API key pasted into the OAuth slot would authenticate, work, and bill the
metered account — the same failure again, discovered weeks later.
`web_search` is carried explicitly rather than defaulted. The Master Planner and
the claw enhancer both pass `true`, and a helper that quietly dropped it would
have taken web search away from two features while every test still passed.
`validator_preflight` deliberately keeps `Runtime::complete`: it probes whatever
validator spec is configured (today `glm:glm-4.7`), and forcing it onto Anthropic
would make it prove the wrong thing. A test pins both halves — no other
server-side caller may regress to the metered key, and preflight must keep
probing the configured spec.
258 lib tests.
Phase 1b: wires the capacity model from 3a2d76a into the launch path, and turns
the existing pending-phase loop into the queue.
Placement moves from mission launch to PHASE launch. A node chosen at launch is
chosen once, minutes before the first VM boots and hours before the last — and
re-placing between phases is free, because mission state lives on the gateway
checkout and every VM is inject -> run -> collect -> destroy. Pinning early
bought nothing and cost the ability to react to a node filling or draining
mid-mission. One call site serves both the solo and composed paths so they cannot
disagree; the composed worker reads `missions.target_node_id`, which placement
writes before dispatch.
QUEUEING, with no new machinery: a phase with no admissible node keeps its
`pending` status and creates no `topology_runs` row. `start_pending_phases`
retries every 10s — that loop already was a queue; nothing downstream ever sees a
run that did not happen.
The risk that creates is the one this codebase keeps paying for: a phase waiting
for capacity looks exactly like a phase nothing is working on. So the wait is
RECORDED, not merely logged — migration 0073 adds `capacity_blocked_since` and
`capacity_note`, stamped once and preserved across retries so the wait is
measured from the first refusal. It is bounded at two full turns: a fleet that
frees will free within one, and a phase that waited two hours must say so rather
than sit pending forever looking like a bug.
Mission launch still fails when NO node could ever run the backend — that is not
transient, waiting will not fix it, and `microvm-negctl` asserts such a mission
stays `draft`. Capacity refusals are transient and queue; capability refusals are
not and fail. The two are separate variants precisely so they cannot be confused.
257 lib tests, 20 binaries.
