The decoder died mid-task and it took four separate findings to explain, each
of which read as something else:
1. OOM-KILLED, REPORTED AS A CLEAN EXIT. `OOMKilled: true` with **ExitCode 0**.
So `--restart on-failure` would treat a memory kill as a successful finish
and leave the agent down -- the policy has to be `unless-stopped`.
2. THE JOB CAP WAS SET AND THEN REMOVED THREE LINES LATER. build-reborn has
always exported CARGO_BUILD_JOBS, but a raw `cargo test --release -p
sylpheed-formats` never reaches the wrapper. Adding `-e CARGO_BUILD_JOBS` to
the launcher did not help either: the entrypoint recomputes and exports over
it unconditionally. An explicit value now wins, and says so in the log.
3. THE MEMORY CONSTANT WAS WRONG. `mem_gib * 2 / 3` assumes ~1.5 GB per job;
release rustc on this workspace needs ~2 GB, and 4 jobs in 6 GB is what died.
Divisor is now 2.
4. `--continue` CANNOT RESUME AN ABRUPT DEATH, which is the only kind we get.
It resolves through ~/.claude.json's per-project `history`/`lastSessionId`,
and MEASURED mid-session both are None -- they are written at a graceful
shutdown. A killed container never writes them, so `--continue` answered
"No conversation found to continue" with 33 MB of transcripts in the volume
beside it. Persisting .claude.json did not help, because the fields were
never populated in the first place; that attempt is removed rather than left
in looking useful.
The TRANSCRIPTS are durable and named by session id, so the entrypoint reads
the id off the newest one for its cwd and passes `--resume <id>`. Verified
on both agents: each reattached to its exact prior session and appended to
the same file rather than opening a new one.
The /loop prompt is still passed alongside `--resume`, so the loop is RE-ARMED
rather than merely restored -- a resumed conversation with no wake-up scheduled
answers once and stops, which looks like resuming and is not.
Restarting into the same death is guarded at the other end: a start less than
120 s after the previous one begins FRESH instead of continuing back into
whatever killed it. That fired correctly during this work.
On resume the agent is told it was restarted, that its in-progress work is
uncommitted in the tree, that any build or capture it had running did not
finish and its absence is not a result, and which wrapper to prefer over a raw
release build.
Credentials were seeded only when the container's copy was MISSING. So when a
session expired, the file still existed, the copy was skipped, and restarting
changed nothing -- the one recovery path a human has, re-logging in on the host,
could not reach the containers at all.
Now re-seeds whenever the host's copy is newer. Newer-wins rather than
always-copy, because a container refreshes its own token mid-run and that copy
may legitimately be the fresher of the two.
Found when both sessions expired: host credentials at 16:30, containers holding
14:20 and 14:24.
The decoder had neither, and reported the gap precisely: four scripts in this
repo READ /work/xenia-rs/sylpheed.db and nothing produces it, so the whole static
PPC route was consumers with the producer missing.
Both exist on the host and are now mounted read-only: the 586 MB database (25 481
functions, 851 classes with RTTI, EH tables, imports, 1.8M indirect-dispatch
candidates) and the decompressed image.
The image is the more useful of the two. It is a FLAT VA DUMP -- file offset =
VA - 0x82000000 -- so reading a known address needs no XEX decrypt, no LZX, and
no booted emulator. The decoder had independently recovered the same bytes by
dumping /dev/shm/xenia_memory_* and validating against the GamePart table, which
is good work and a sound method, but it noted itself that needing a running
emulator is a bad dependency for something the entire static corpus rests on. It
does not need one.
Also recorded that an earlier claim the .pe was STALE was tested and refuted, so
nobody re-litigates it, and that instructions.raw is an INT rather than hex.
Written down as reference material, explicitly NOT a deliverable: they are
read-only, they come from outside the repository, and a fresh checkout elsewhere
has neither. Reimplementing the producer belongs in sylpheed-formats, and until
it exists every static finding rests on an artefact this project cannot rebuild.
The audio work was three parts and I shipped two. The transcode-fidelity method
and the pinned 5.1 downmix landed; the null sink -- the only one that answers
"what does the GAME play" -- I deferred to "the next natural rebuild window" and
then rebuilt both images four times without doing it.
pulseaudio-utils is now in both, with tools/audio-capture wrapping it: a null
sink is a real device as far as an application is concerned, so Canary and Godot
open it normally and parec records what they emit.
This unblocks the decoder's Q8. The cue-to-event bindings are currently a name
match against the authors' own identifiers -- a plausible guess, not a
measurement -- and capturing what the game plays on a menu move converts them.
`audio-capture run` reports the peak level and warns when the capture is silent,
because silence is the failure that looks like success: a WAV of exactly the
right duration, full of zeroes, because the application opened a different sink.
A duration check alone passes it, which is how a confident wrong number gets
made.
Three defects, all mine, found by checking instead of assuming.
**They never exchanged a word.** SendMessage=0, ListAgents=0 across both new
sessions. PROTOCOL.md specified in detail what a message MAY and MAY NOT do and
never said how to send one or that the other agent was addressable -- they knew
that last time only because the human told them directly, and rebuilding with
fresh volumes wiped it. Policy without mechanism is prose. Now documented with
the two addresses, a worked example, and an instruction to introduce themselves
on the first iteration rather than waiting to have a question.
**The decoder lost the disc and the ISO.** They used to arrive inside the project
mount and silently stopped when /work became a clone. Silently is the word: the
disc-gated tests SELF-SKIP without SYLPHEED_DISC and report green, so a whole
test suite would have passed while measuring nothing. Both are now mounted
explicitly, the ISO at a stable path so run-canary does not depend on host
directory names.
**Both agents shared one Claude state directory.** They share the host's
~/.claude, and once both working directories became /work they resolved to the
same projects/-work/ -- two supposedly independent agents writing to one place,
which undoes the point of separate checkouts. Each now has its own volume, seeded
once from the host with credentials only, so a token refresh writes locally and
neither can corrupt the host's auth.
Also widened the pacing rule. It banned ScheduleWakeup by name; the decoder then
scheduled itself an hourly cron job -- not harmful, but the same instinct that
ended a run yesterday, through a door I had left open. Now: no self-scheduling by
any route.
Mount audit after the changes: shared and intentional are the exchange volume and
the read-only credential seed. Everything else -- repo, Claude state, cargo,
target, canary, disc, ISO -- is per agent or one-sided.
Two bugs, both mine, both found by starting the thing.
**Volume mount points must exist AND be owned by the agent before USER agent.**
Docker seeds a named volume from whatever the image has at that path, ownership
included, and creates a ROOT-OWNED directory when the path is absent. Either way
the agent cannot write, and the failure surfaced far from its cause: "clone
FAILED", with no permission error anywhere in sight. The port's own Dockerfile
already carried a comment explaining this trap, which I then walked into for
/work and /exchange.
**The clone guard checked for a .git directory, not a usable HEAD.** A clone
interrupted partway -- the container was removed while one ran -- leaves a .git
with no commits, and a presence check then skips the retry forever and hands the
agent an empty repository that looks like a checkout. It now verifies HEAD, and
clones via a temp directory so a partial result never lands in /work at all.
Also: the port launcher's path defaults still assumed the old repo root, so it
mounted no disc; and the stale /reborn notice is gone now that there is one
repository.
Verified running: both agents cloned c58196b, `share` on PATH from /work/tools,
/exchange agent-owned, canary at /canary for the decoder, disc at /disc for the
port.
The last structural fix for the collision class that has bitten three times. Both
containers now clone the repository into their OWN named volume instead of
bind-mounting a human's working tree, so an agent's local git config cannot
capture a human's commits, a credential helper cannot leak a container-only path
onto the host, and a `git add -A` cannot sweep another party's in-flight files.
Cloned once at startup and never auto-pulled: pulling under a running agent
moves files out from under whatever it is mid-edit, which is the same bug again.
Accepted knowingly: Claude Code keys per-project memory off the working
directory, so moving off the host path starts that memory empty. The corpus in
docs/ is the memory that matters and it travels with the clone.
Other changes:
* docker/agent -> docker/decoder; the launcher is sylph-decoder. Roles, not
"the agent", now that there is more than one.
* /reborn is gone -- one repository now, so the port reads HANDOFF from its own
checkout rather than through a live read-only mount of someone else's tree.
* Canary mounts separately at /canary; it stays a fork tracking upstream.
* A shared `sylpheed-exchange` volume at /exchange, with tools/ on PATH so
`share` is available in both.
* The decoder's credential file gets the .host-copy treatment the port already
had -- `credential.helper=store` rewrites by rename-over-target, which is
EBUSY on a bind mount and reports a fatal that is not one.
* Budget split deliberately: decoder 5 cpu / 6 GB, port 3 / 4, leaving room for
the planned Referee. "Half the host" was right when there was one agent.
Prompts move to docs/agents/ and are rewritten around the protocol: the oracle
is the running game, dynamic RE stays with the decoder, each iteration must
attempt to refute one claim of the other, and neither may verify its way out of
its own role.
Merges the Godot port into the reverse-engineering repository, preserving both
histories -- 1019 commits of corpus plus the port's 31, brought in by subtree
merge and then moved into place so git can follow each file across the rename.
The reason is not tidiness. The two-repo split forced the exporter to depend on
the decoders by pinned revision, and that created a whole class of failure that
now disappears: a sha reachable only from a topic branch, orphaned by a
squash-merge, breaking a fresh checkout silently at build time. It also forced a
live read-only mount of one agent's working tree into another's container, which
is why a contract file could move mid-iteration. With a path dependency, a
decoder change and the exporter change it requires land in the same commit or
not at all.
Canary stays separate: it is a fork tracking upstream.
New structure for the long term:
docs/game/ how the game is NAVIGATED -- menus, modals, prompts, alerts,
and in-game flight. Written so nobody rediscovers it. Mostly
open questions on purpose; the in-game tutorials are the
resource for the flight half.
docs/port/MODDING.md
modding as a constraint on the exporter TODAY, not a later
feature: one logical asset in one file (the disc splits nearly
everything, and resolving that is the exporter's job), names a
person recognises, PNG/OGG/OGV/JSON only, base-and-overrides so
re-exporting is always safe, provenance in every file.
data/base + data/mods
generated tree and drop-in overrides, both gitignored
exchange/ transient inter-agent files, deliberately outside history
docs/agents/ the team protocol
Both the README and the navigation doc lead with the correction that cost the
most: the oracle is the real game under Xenia Canary. Reborn's renderer is a
hypothesis under test, it has been wrong, and treating it as ground truth
propagated into three documents and both agents before a human caught it.
Scripted modding stays possible without being built: no screen name is hardcoded
in GDScript and there is no native code in port/, which is what Godot Mod Loader
needs to be able to substitute behaviour later.
Two fixes to push-work and the policy that goes with them.
**Tags.** The port's exporter depends on sylpheed-formats BY REVISION, so a
commit of ours is part of its build -- and the commit it pinned lived on one
auto/* branch and nowhere else. Deleting that branch orphans it; squash-merging
it is worse, because squash creates NEW commits, so main appears to contain the
work while the pinned sha becomes unreachable and the port stops building for a
fresh checkout. Silently, at their build, long after the breakage.
push-work now pushes with --follow-tags, which publishes annotated tags
reachable from the pushed commits, and MISSION.md says to tag whatever the port
needs. formats-pin-2026-08-29 at 76653ca is the first, created after the fact.
**Credentials.** This script set credential.helper as --local config, which
PERSISTS in the repository. The repo is a bind mount the host also uses, so the
host's git inherited a path that exists only inside the container and every host
push failed with "unable to get credential storage lock". Now applied with -c to
the single push. Same fix already landed on the port side; this copy still had
the bug and would have re-leaked on the next push.
Committed with a pathspec so the agent's in-flight work is untouched.
A run ended with a clean exit 0 while the display title read "Loop interval
optimization", leaving four files uncommitted in the tree. Nothing crashed --
the loop was ended, and ending the loop ends the run: the container exits and
there is no next iteration.
The prompt said the agent did not NEED to arm a wakeup. It never said not to
stop one, and an agent that reads about a pacing control will reasonably try to
use it. Now explicit: do not call ScheduleWakeup at all, and if the cadence is
wrong, say so and leave it to a human -- the interval is set outside the prompt.
Recorded with the date and the symptom, because "the container exited cleanly"
looks like a finished job rather than a self-inflicted stop.
The `loose` footer announced "cannot push -- no git credentials are mounted"
unconditionally. It was written before the credential mount existed and then
went stale, so with credentials correctly mounted it told the operator their
work would be lost -- the exact failure the mount was added to prevent.
It now reports what is actually true, and keeps the real warning for the case
where the file genuinely is missing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Corrects the split I got wrong in 965c3f7. The container is for reverse
engineering, now focused on the menu port; a SEPARATE agent builds the port from
its investigation results. My previous version had the container agent writing
the exporter and the Godot project, which is not the intent.
This lands on the research/engineering line that was already in the estimate:
the container agent takes the research half, the port agent the engineering half.
docs/port/MISSION.md is now a list of open QUESTIONS (Q1-Q9) rather than build
milestones, ordered by what blocks the port earliest -- the keyframe time unit,
which build is which screen state, paint order for the six screens, button ->
GamePart, navigation semantics, the boot sequence driver, transitions, menu
audio bindings, and video binding. Each is done when a written result with
evidence exists, not when something compiles. S1, the Ready Room probe, stays
gated at one iteration and a go/no-go.
Most of these are BEHAVIOUR questions -- timing, transitions, what a d-pad press
does at the end of a list -- so the mission and the loop prompt both push hard
on measuring the oracle rather than reasoning from the file.
docs/port/FORMAT.md is deleted. The export schema is the port agent's design and
was not mine to specify. It is replaced by docs/port/HANDOFF.md, the single page
the port agent reads: a status table, what is already settled and can be relied
on today, and the facts that will trip the port up (the WMV3/WMA Pro intro, the
Static.slb size over-declaration, the voice-vs-music downmix, JNGL_001).
The derived/authored idea survives as the thing it always was -- a finding, not
a design. Every answer must be classified DECODED, MEASURED or UNDECODABLE-with-
reach, and never a fourth thing, because measured and undecodable both mean the
port agent is authoring that value and has to know it. Labelling a guess as a
decode would put it into the port wearing the badge of a measurement.
Reverts the Godot install from the RE container, its AGENT.md section, and the
export/ gitignore entry -- none of that belongs on this side of the wall.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sets a new mission: boot the real disc through developer splash -> intro video
-> title -> main menu -> submenus in Godot 4, interactively, with no gameplay,
no 3D and no emulator.
docs/port/MISSION.md defines it -- eight gated milestones, each finished by an
ARTIFACT rather than by compiling, plus the Ready Room as an explicitly gated
stretch goal with a one-iteration probe that decides go/no-go. GP_READY_ROOM is
1106 entries with 6 recoverable names and is ISL-scripted, so it is either a
week or a quarter, and the agent must not start it on its own authority.
Architecture, per the user's decision: the Godot project is INDEPENDENT of the
Rust viewer and never reads a disc format. An offline Rust exporter converts the
disc into open formats; Godot reads only those. No GDExtension, no Rust in the
Godot project, and sylpheed-viewer is off limits -- it stays the human's
verification tool with its static-data rule intact.
docs/port/FORMAT.md specifies the open format, versioned, because modding is the
port's second goal and that makes the layout a deliverable rather than a temp
directory: JSON over XML (Godot parses JSON natively; its XMLParser is SAX),
names never hashes, provenance in every generated file, and unknowns listed
rather than guessed.
The discipline the whole thing rests on is the derived/authored split. `export/`
is regenerated wholesale and never hand-edited; `authored/` is hand-written and
survives a re-export. Three things this milestone needs are NOT on the disc in
any decoded form -- which button does what, paint order, and menu sound cues --
so they live in `authored/` with a stated `why`. Deleting an authored entry
because the exporter can now emit it IS the measure of progress.
`export/` is gitignored: it is generated from the user's own disc and this stays
a clean-room repo.
Container: adds a pinned Godot 4 (windowed under Xvfb for screenshots, plus a
headless wrapper). ffmpeg already carries libtheora, which is the video target --
Godot 4 plays only Ogg Theora natively and the disc's ADV.wmv is WMV3/WMA Pro.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The autonomous agent's loop prompt had grown to ~9 000 words of accumulated
findings, refutations and traps. That is a symptom, not a style: it was doing
the job the corpus should do, in the one place that dies with the container.
Three concrete failures followed from it, and each gets a structural fix rather
than a louder instruction.
1. IT REPEATED WORK IT HAD ALREADY DONE.
The "do not revive" list and the method traps existed ONLY in the prompt, so
nothing pointed a fresh iteration at them. Extracted verbatim into two
tracked files:
docs/re/REFUTED.md 105 claims tested and dead, grouped by subject so a
grep for your noun finds the neighbourhood
docs/re/METHOD.md the traps already paid for -- controls, inference,
searching, reading data, runtime
Both are linked from INDEX.md, and the loop prompt now opens by requiring
them to be read. This is the fix for "re-derived something already known":
the knowledge is now where the next iteration looks, not in a context window.
2. IT FORGOT TO ARM THE NEXT WAKEUP.
The prompt tried to solve this by shouting at itself in the first line. The
real fix is to stop asking: `loose` now defaults to a FIXED interval (45m),
so the harness owns the cadence and a forgotten ScheduleWakeup cannot end the
run. SYLPH_LOOP_INTERVAL= (empty) restores self-pacing.
3. IT COULD NOT PUBLISH, SO THE WORK ONLY EXISTED IN THE CONTAINER.
New `push-work`, plus a read-only credentials mount
(SYLPH_GIT_CREDENTIALS, default ~/.sylph-git-credentials). It pushes the
CURRENT branch only, refuses anything that is not auto/*, and never
force-pushes -- so the consolidated line stays a human's decision and a
confused iteration cannot rewrite history. The loop prompt now requires a
push on every iteration that commits, rather than at the end of some longer
arc, which is exactly when a container dies.
The prompt itself drops from ~9 000 words to 85 lines and, more importantly,
stops accumulating: findings go in the corpus, and the prompt points at it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Remote Control, so a detached run is not a one-way trip:
./sylph-agent remote -> https://claude.ai/code/session_...
./sylph-agent attach -> the container's own terminal
`loose` now starts Claude Code with `--remote-control <name>`, registering the
session with your account so you can chat with it from claude.ai or a phone.
The name is passed EXPLICITLY because the flag's value is optional — a bare
`--remote-control` swallows the /loop prompt that follows as the session name.
Off with SYLPH_REMOTE=0, renamed with SYLPH_REMOTE_NAME.
The pty is forced to 200x50. A detached `docker run -t` gives 80x24, and Claude
Code hard-wraps to the terminal width — which truncated the Remote Control URL
to ".../session_01..." in the one place you actually need to read it, and made
`docker logs` nearly unusable besides.
`remote` waits up to six minutes and reports progress: registration lands a
minute or two after launch, so answering "not found" immediately would be
answering a different question than the one being asked.
Verified: a loose run registers and `./sylph-agent remote` prints the full URL.
`attach` is NOT verified end to end from here — `docker attach` refuses a piped
stdin ("cannot attach stdin to a TTY-enabled container"), which is a property of
this harness rather than of the container, so it needs a real terminal to try.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`./sylph-agent loose [task]` starts Claude Code detached with
--dangerously-skip-permissions, running /loop on loop-task.md: work the RE
backlog one item at a time, commit to auto/* branches, never push, record
withdrawn results rather than deleting them. `logs`/`attach`/`stop` to watch and
end it. Runs -d WITHOUT --rm so the transcript survives the container exiting —
for an unattended run that is the only record of what happened.
Two things had to be fixed for an agent to survive being left alone.
MEMORY CONTINUITY. The project is now bind-mounted twice: at /work, and at its
own host path. Claude Code derives its per-project state key from the working
directory, so running at /work handed the agent an empty project instead of the
accumulated one. Verified: a loose run now reports MEMORY=yes and reads back the
same branch and backlog as the host.
FOUR INTERACTIVE GATES, each a silent permanent hang with nobody at the keyboard
-- no error, no log line, just a container that looks healthy and does nothing:
theme picker hasCompletedOnboarding + lastOnboardingVersion. Re-fires
whenever the container's Claude Code is a different version
to the host's, which is the normal case.
folder trust projects.<path>.hasTrustDialogAccepted
bypass disclaimer answered in a pty by bin/claude-autonomous. It has no config
key by design -- it wants a person to accept once, and the
person did so by launching this.
fullscreen upsell fullscreenUpsellSeenCount. This one fires MID-SESSION, after
the pty wrapper has already handed over, so it cannot be
answered the same way.
Config key names were read out of the shipped binary's own strings, not guessed.
The pty wrapper matches SINGLE WORDS. Claude Code draws its UI with
absolute-column escapes between words, so the prompt arrives as
`Yes,\x1b[13GI\x1b[15Gaccept` and a multi-word pattern never matches -- failing
in a way indistinguishable from the wrapper not running at all. It stops
matching once the session is live so nothing later is answered by accident.
~/.claude.json is now mounted read-only at a staging path and copied in, so the
container cannot rewrite the host config. Credentials stay shared read-write in
~/.claude, which is what token refresh and memory continuity need.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Builds and runs both halves of the project -- Canary as the oracle, Reborn as
the port -- plus the dynamic-RE toolkit. Claude Code runs with
--dangerously-skip-permissions as an unprivileged `agent` user, because that
flag is refused under root.
Capped at half the machine, computed at launch: --cpus nproc/2, --memory half
of MemTotal with --memory-swap equal to it (no swap escape hatch -- a swapping
build thrashes the host, which is the failure the cap exists to prevent), and
build parallelism derived INSIDE the container from available memory rather
than core count, since a full-parallel build of this tree has OOM-killed the
host outright.
Three things the old box got wrong are fixed rather than reproduced: a real
toolchain (so rebuild_canary.sh's hand-relinking is obsolete), numpy and Pillow
(whose absence silently disabled every image oracle and looked like a logic
bug), and a display owned by PID 1 (so Xvfb no longer "dies on its own every
few minutes" -- it was being reaped because nothing owned it).
Verified end to end, not by inspection: the image builds, sylph-doctor is green,
`build-canary` links xenia_canary inside the container, and that binary then
runs -- guest memory and the JIT code cache appear in /dev/shm within 4 s,
1 205 log lines, gmem.py reads guest RAM, pad.py drives the file pad, and
screenshot captures the display.
Five environment defects found and fixed on the way, each of which fails in a
way that points somewhere else entirely:
* /dev/shm is `noexec` under Docker. Xenia maps its JIT code cache out of an
shm file, so it died with "Unable to allocate code cache generated code
storage / Cannot initalize processor" -- which reads as an address-space
clash, not a mount flag. Now `--tmpfs /dev/shm:rw,exec`.
* An unknown xenia flag HANGS rather than errors: ParseLaunchArguments calls
ShowSimpleMessageBox before logging is initialised, and that SDL dialog
blocks on XIfEvent forever. `--audio` (which the RE notes recommend) is not
a cvar in this tree; the symptom was a 10x10 window and an empty log.
* Named volumes come up root-owned unless their mount point exists in the
image, so the first cmake configure failed on pkgRedirects.
* Ubuntu 24.04 ships its own uid-1000 account, colliding with the host user.
* Ubuntu's spirv-opt has no --canonicalize-ids, so the shader step dies ~500
objects in; the launcher mounts the host's LunarG SDK instead of baking one
in, which also keeps shader output byte-identical to a host build.
Known limits, stated rather than papered over: on an NVIDIA host without the
NVIDIA Container Toolkit there is no hardware Vulkan (/dev/dri alone does
nothing for NVIDIA), and under lavapipe the emulator runs correctly but was not
observed to reach a rendered frame within a couple of minutes. gdb needs `sudo`
inside the container because the host's yama ptrace_scope outranks SYS_PTRACE.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
