.agents/shared/test-loop.md
The portable core of autonomous, tested implementation. /perform-task (Claude Code) and
$perform-task (Codex) read it. This file defines shared defaults after one task's implementation
is ready to commit; wrappers own setup, splitting, and spawn/wait mechanics. A wrapper may adapt
commit ownership, task baseline/attempt caps, staging/source restoration, account swapping,
EVIDENCE_DIR, or an optional UI driver. Its named rule wins only at that
adapter point; every other rule here still applies.
#ifdef _DEBUG test code for the current task. Never part of an
implementation commit. Lives as a patch under the task folder between rounds.TASK_DIR — external ai-tdesktop/tasks/<full-task-id>/ directory for this task.WORK_DIR — tracked resumable artifacts under TASK_DIR/work/.TASK_ID — full dated task identifier and required source-commit locator.BASE_REF — local pre-task baseline ref derived from TASK_ID.GREEN_REF — local ref for the current retained implementation attempt.EVIDENCE_DIR — per-run logs and screenshots; defaults to TASK_DIR unless the wrapper passes a
run-specific directory.task.md, including its design
basis when the wrapper records one, plus any referenced images (images/<file> mockups /
screenshots / graphic resources). Images are optional evidence: read them when present, but their
absence is never by itself a planning, implementation, or test blocker. The spec and its cited
repository/baseline sources are one side of test design; the implementation diff is the other.BUILD (build command), EXE (built binary path), MAX_ATTEMPTS (default 4). The test
account lives in out/Debug/ as the portable-data folders described under "Test account" below;
the wrapper has already confirmed the golden one exists (launch gate). All paths are relative to
the current checkout — no worktrees are created; the run happens in whatever repository slot it
was launched from.Precondition: the implementation for this task is ready in the current checkout. The performer
stages exact task-owned paths and commits; leaf implementation agents never stage, commit, or stash.
Move GREEN_REF to that commit. The runner tracks the attempt number as its own
state (attempt starts at 1); the commit message carries no attempt marker.
Never resolve a local task ref into a hash in any artifact or report. Commits
follow "Commit message" below.
TEST_AUTHOR -> RUN -> ASSESS (adversarial — see "Assessing"):
APPROVED -> restore overlay paths to GREEN_REF; delete the test binary; return DONE up.
TEST_FLAW -> fix the overlay only; back to RUN. Does NOT cost an impl attempt.
IMPL_BUG -> spawn impl-fix agent (input = test.md, latest attempt's Root cause / Fix hint);
performer commits a NEW attempt and moves GREEN_REF; re-apply overlay
(--3way, else re-author); RUN. attempt++
UNRECOVERABLE -> delete the test binary; return BLOCKED up with the reason. Stop.
attempt > MAX -> delete the test binary; return BLOCKED up with test.md + "improve" notes. Stop.
On every TERMINAL exit (APPROVED / BLOCKED / UNRECOVERABLE / cap) "delete the test binary" means the
step in "Leave no test binary behind" below.
Early-escalation rule: if two consecutive ASSESS rounds produce the same failure signature (same step fails the same way after a fix), stop and return BLOCKED — do not burn the rest of the attempt budget chasing it.
UNRECOVERABLE conditions: the app reaches a login screen / AUTH_KEY_DUPLICATED and re-copying the
test account does not recover it, or a crash has no usable diagnostic after one retry. Missing
test_TelegramForcePortable is a global environment hard stop, not a task Block. A file-lock build
error (LNK1104, C1041, access denied, file in use) is likewise a repository hard stop: do not
retry or work around it; ask the user to close the app and debugger.
GREEN_REF to the resulting commit.test.md) is the only fix-agent handoff. Give it the latest Attempt/Run section,
especially Root cause / Fix hint and Failure signature. Reserve wrapper-owned result.md for the
final AI result or exceptional blocked boundary; never create result<n>.md.Impl commits must read like the repository's own history and carry only the durable task locator.
Match the style of recent git log subjects.
Task: <TASK_ID>.Autotask:, attempt marker, Co-Authored-By:, or any
tool/assistant attribution. The attempt number is runner state, never part of the message.The debug build runs in portable mode out of out/Debug/. Three sibling folders matter:
test_TelegramForcePortable — the golden test account, prepared by the user. Read-only SOURCE,
never modified by tests. (Its presence is the launch gate; the wrapper aborts if it is missing.)TelegramForcePortable — the LIVE folder the app actually uses (its presence is what puts the
build in portable mode). A marker file named testing directly inside it marks it as a
disposable test copy; a live folder WITHOUT the marker is the user's real data.real_TelegramForcePortable — the user's real data, preserved so manual use survives. Once it
exists, NO flow step may ever delete, rename, move, overwrite, or write into it.SETUP — run at the START of every test run, with NO app instance alive. Idempotent, and a pure no-op between runs and between consecutive tasks (the marked test copy is simply reused).
test_TelegramForcePortable. Its absence is the only portable-account setup blocker.TelegramForcePortable/testing exists, the live folder is already the reusable test copy:
touch none of the three folders and proceed straight to testing.TelegramForcePortable exists without the marker, it is the user's real data: move it to
real_TelegramForcePortable when that is absent. If real_... already exists, the unmarked
live folder is the user's manual restore of that same preserved data — recursively delete the
live folder only.test_TelegramForcePortable to TelegramForcePortable, then create the marker file
TelegramForcePortable/testing (any content, e.g. echo 1). Never rename, modify, or delete
the golden folder.Any live/real folder combination is never a blocker. After SETUP the live folder is a marked test
copy, the golden folder is untouched, and real_... may or may not exist.
NO CLEANUP — the flow performs no folder operations after testing, ever. The marked test copy
stays live, so the next run or next task starts with SETUP as a no-op and the folders are never
copied, moved, or deleted between testing phases. The flow never restores real data to live: when
the user wants manual use they copy real_... to TelegramForcePortable themselves (keeping
real_... in place), and the next SETUP handles that unmarked live folder by step 3.
Deletion guard — the only folder the flow may ever delete is a live TelegramForcePortable that
either carries the testing marker or coexists with real_... (step 3). If the test account
breaks mid-loop (login screen, AUTH_KEY_DUPLICATED), delete the MARKED live folder, re-run SETUP
for a fresh golden copy, and retry once; if it is still broken the run is UNRECOVERABLE. Never
delete or alter test_... or real_... under any circumstances.
Serialize app runs. Never have two Telegram.exe instances alive against this account at once —
concurrent reuse of one auth key can trigger a server-side session reset. Before SETUP, launching, or
rebuilding, kill any straggler of THIS checkout's binary only — the one whose full executable
path is EXE (out/Debug/Telegram.exe in this checkout). Match on the full path; do NOT blanket-kill
every Telegram.exe on the machine. The user may be running a system-installed client or another
checkout's build against unrelated accounts — those use different auth keys, never conflict with this
account, and MUST be left alive. On Windows, scope the kill by path:
$exe = (Resolve-Path "$EXE").Path
Get-CimInstance Win32_Process -Filter "Name = 'Telegram.exe'" |
Where-Object { $_.ExecutablePath -eq $exe } |
ForEach-Object { Stop-Process -Id $_.ProcessId -Force }
taskkill /IM Telegram.exe /F is forbidden here and anywhere else in this loop — it is image-name-wide
and takes down the user's unrelated clients. Every "kill stragglers" / "taskkill" step below means
this path-scoped kill.
Avoid account-fatal calls; cloud data is otherwise fair game. The overlay must never trigger logout / session-termination / account-deletion, and must not wipe the account wholesale. Tests that genuinely need those use a separate burner account, not this one. (If a permanent destructive-call fuse is later added to the debug build, this is enforced in code; until then it is the test-author's responsibility.) Everything short of that is allowed: this is a test-server account, so freely CREATE content in any chats (messages, drafts, tables, media) and freely DELETE or clear content that test runs created — including leftovers from previous runs and sessions (e.g. clear the self-chat rich compose cloud draft before a run instead of designing around accumulated junk; the live test copy is reused across runs and tasks, so local AND cloud state accumulate — reset whatever state the test depends on at the start of the run). Don't delete anything the user placed on the account by hand unless the task says so.
The single most important rule: tests are derived from what THIS task changed — not from generic project navigation, and not reused from a previous task. Different change → different checks. If two tasks produce the same screenshots and the same assertions, the second test is a no-op. Before writing any overlay:
Design-Basis: or
equivalent cited sources, and every referenced image when present. (b) The change under test —
git diff <BASE_REF>..<GREEN_REF> (the complete task diff) and
<WORK_DIR>/plan.md. List every concrete thing the
diff changed and every surface the task (description + "Observable result") says it affects.
The diff proves what shipped; it is not independent authority for what the design should be.Visual: appearance;
add screenshot comparison only when its presentation is separately in scope. By change type:
<WORK_DIR>/visual.md, cited current/legacy analogues, style-token or resource identity, and
the pre-task baseline. Confirm a baseline delta whenever the task requires one. If the target
still matches the old state when a change is expected, that is a FAIL, not a pass. A
Visual: layout task must also satisfy every numeric design-contract line (sizes, spacings,
alignment); supplied artwork is optional and never a prerequisite for that contract.<WORK_DIR>/test.md BEFORE running (format under "Test report"), so the
design is explicit and Actual/Result can be filled in per check afterward.When the wrapper marks a task Visual: layout, "looks right" is not a vibe — it is a small
computation, and the test MEASURES it. The wrapper's design-spec phase writes the contract to
<WORK_DIR>/visual.md; impl builds to it; this loop verifies it. (Tasks marked Visual: appearance
use the ordinary visual/asset check above. Unmarked non-visual tasks use their applicable text or
behavior checks; only legacy unclassified visual changes use the visual/asset branch.)
Build the contract from the strongest available design evidence: explicit request relationships;
supplied references when present; current or legacy task-adjacent UI; then the closest established
desktop component/style token while preserving unspecified behavior. A mockup gives relationships,
never desktop pixels. With no mockup, repository anchors and the written requirement provide those
relationships. The strongest anchor is an existing widget: "the count badge IS the dialogs-list
unread badge" pins font + height + padding to st::dialogsUnread* and is self-correcting — far
better than "a blue circle ~24px". Cite each source and record every inference; never invent a
reference or arbitrary geometry merely to fill the contract.
Write it as an ORDERED DERIVATION: each step resolves one quantity the next consumes, so impl and test are both mechanical. Example — a glyph-on-rounded-square icon + title + count, in a bubble:
Anchor: T = st::<title>.font->height ; Badge := the dialogs unread-badge metrics
1. glyphH = 1.4·T ±2px — white glyph box height (from T)
2. square = glyphH ÷ (2/3) ±2px — accent rounded-square side ; iconR = square·0.28
3. margin m (equal on square's top/left/bottom) ; bubbleH = square + 2·m ±1px
bubbleR = bubbleH/2 ; iconR : bubbleR must read as in-sync (icon proportionally smaller)
4. titleY = (bubbleH − T)/2 ±1px — title vertically centered in the bubble
5. badge = Badge (font+height+padding) ; vertically centered ; margins top=right=bottom equal ±1px
Then the RELATIONSHIP checks that catch what existence-checks miss — each falsifiable: square ≤ bubbleH (no overflow/overlap), the square's three margins equal, the two corner radii in sync, the
badge identical to a real chat-row unread badge. Note each source-to-desktop adjustment and which
token or metric grounds it; describe mobile→desktop conversion only when a mobile reference exists.
How TEST verifies it (numbers over eyes):
font->height and the
QRect of each piece (glyph, square, bubble, title, badge) — and assert each derivation line
arithmetically within tolerance. Live-widget geometry is the primary oracle; it deterministically
catches "icon taller than the bubble", "square overflows", "badge oversized / cramped". Where a
rect can't be logged, measure it from a tight crop by colour (accent square, badge, bubble outline
are separable).Visual: layout check APPROVES only when the measured geometry satisfies the contract; any line
out of tolerance is an IMPL_BUG (report measured-vs-target) and loops like any other.The overlay is ad-hoc, authored fresh against the CURRENT implementation, injected at the
highest level that still exercises the change (often a direct data-layer call like
item->applyEdition(...) rather than a faked MTP response). It is also the complete runtime driver
of first resort: prefer programmatically triggering every required action and judging the saved logs
and captures afterwards over any external desktop driver, whether or not one is available. Drive the
whole task-specific flow inside the Debug binary by invoking application actions or posting Qt input events on the event loop, waiting for
observable state, logging assertions, capturing the rendered target in-process, and quitting. A
locked macOS session does not reduce required coverage and is never a testing blocker. The overlay
must:
#ifdef _DEBUG blocks.live-data (use real account data) · live-mutate (really create an entity — prefer a
throwaway target, clean up after) · inject (build fake local state without the network) ·
mock-api (intercept specific requests, return canned responses — for payments/destructive).
Prefer inject over live-mutate to avoid account/server accumulation and flake.<EVIDENCE_DIR>/test_log.txt (open Append|Text, flush after each write) and
save screenshots to <EVIDENCE_DIR>/screenshots/. Delete the old log at the first step.QWidget::grab() or a renderer-owned image after layout and paint have completed. A widget or
test-window grab plus logged geometry is primary visual evidence; never wait for unlock merely to
obtain a desktop screenshot.<name>_{old,new}.png. Without target
artwork, save the baseline/reference-component crop when available and log the contract anchors,
style/resource identities, and measurements. Never fabricate an _new image.TEST_STEP: <desc> · TEST_RESULT: PASS: <what> / TEST_RESULT: FAIL: <what> - <details> ·
SCREENSHOT: <full path> · TEST_COMPLETE (immediately before quit).QTimer at scenario start that force-quits (Core::Quit(), and if
needed std::abort after a flush) at a hard wall-clock cap (default 120s). This guarantees the
app never hangs holding a lock on the exe — independent of the runner's own timeout.TEST_COMPLETE then Core::Quit().Telegram's custom widgets (Ui::InputField, Ui::FlatLabel, Ui::RpWidget, boxes, buttons, …)
do NOT declare Q_OBJECT — they have no own meta-object. So QObject::findChildren<T*>() does
not filter by type for them: with no distinct meta-object it matches the nearest moc'd base
(QWidget), i.e. it returns every child widget blindly cast to T*. The moment you use one as
T (e.g. call InputField::setFocused() / rawTextEdit() on what is really a VerticalLayout) you
get a raw SIGSEGV — the debugger shows this with the wrong dynamic type. A clean rebuild does NOT
fix it; it is a real bug in the overlay, not a stale build.
findChildren<Ui::SomeCustomWidget*>(). Instead enumerate findChildren<QWidget*>()
(QWidget is Q_OBJECT, so that call is sound and returns all descendants) and
dynamic_cast<Ui::SomeCustomWidget*>() each, keeping the non-null results — C++ RTTI identifies the
real type regardless of Q_OBJECT. A reusable helper:
template <typename T>
[[nodiscard]] std::vector<T*> FindWidgets(QWidget *root) {
auto out = std::vector<T*>();
for (const auto w : root->findChildren<QWidget*>()) {
if (const auto t = dynamic_cast<T*>(w)) out.push_back(t);
}
return out;
}
Q_OBJECT types (QWidget, QLabel, QLineEdit, …) are safe to pass directly to
findChildren<T*>().The Windows launcher changes the working directory to the exe folder before the app runs, so a
relative overlay log path (<EVIDENCE_DIR>/test_log.txt) silently fails to write (QFile won't
create missing parents) — the run looks "clean" but produces no evidence. Create and resolve
EVIDENCE_DIR to an absolute path up front (or bake its absolute path into the overlay) so flushes
actually land; likewise for screenshots.
<WORK_DIR>/test-overlay.paths; no
unrelated or untracked source path may be used. After building, save the overlay with
git diff --binary HEAD > <WORK_DIR>/test-overlay.patch and verify the patch is nonempty and
reapplicable. Restore only the inventoried overlay paths to GREEN_REF; never hard-reset the
repository. The overlay never enters an impl commit.git apply --3way <WORK_DIR>/test-overlay.patch. This succeeds ~90% of the time when the tail change was small.<WORK_DIR>/test.md (which
records injection point, fake values, and assertions) rather than fighting conflict markers.
Scenario steps that only call public APIs should live in their own block so they never conflict;
only true in-situ injections land inside impl files.On macOS, a locked graphical session disables external UI driving only. Launch EXE normally,
run the in-binary overlay flow, collect its logs and widget/window grabs, assess them, and clean up.
Do not try to unlock the session and do not return BLOCKED because the lock screen is present.
Build with BUILD. A single changed TU compiles fast; only the overlay-touched files + link
rebuild between rounds. Proactive path-scoped cleanup may run before the build. If the build reports
LNK1104, C1041, access denied, or file in use, follow AGENTS.md: stop immediately, do not
retry or attempt a workaround, and ask the user to close the app/debugger.
Codegen does not track resource mtimes. If the task changed only a resource the style codegen
consumes (an icon .svg, etc.) without touching a .style, an incremental build will NOT re-pack
it and the binary keeps the OLD asset. Before building such a task force regeneration — touch the
referencing .style (or clean the codegen output) — so the change actually ships. A render that
shows no difference from before is the symptom of skipping this.
Run: run the SETUP steps (Test account) -> create EVIDENCE_DIR -> launch EXE with
-testagent in the background, redirecting stdout to <EVIDENCE_DIR>/app_stdout.txt and stderr
to <EVIDENCE_DIR>/app_stderr.txt (this flag prevents modal crash hangs, and stderr captures
assertion text) -> start a hard wall-clock deadline (~90s) from launch -> poll
<EVIDENCE_DIR>/test_log.txt every ~5s -> on each SCREENSHOT: read the image and judge it -> detect
TEST_COMPLETE (success) or process death (crash) or no new output for the watchdog cap, or the
hard deadline elapsing (hang) -> path-scoped kill of any straggler (Test account → "Serialize app
runs") -> save the binary overlay patch -> restore only inventoried overlay
paths to GREEN_REF (the patch must be saved before this restore).
On Windows, launch and capture both streams like:
$exe = (Resolve-Path "$EXE").Path
Start-Process -FilePath $exe -ArgumentList '-testagent' `
-RedirectStandardError "$EVIDENCE_DIR/app_stderr.txt" `
-RedirectStandardOutput "$EVIDENCE_DIR/app_stdout.txt" -PassThru
-testagent)A Debug build normally turns a failed std::vector bounds check, a bad iterator, an assert(), a
pure-virtual call, or abort() into a modal Abort / Retry / Ignore dialog. That dialog blocks
the process forever — the agent sees no TEST_COMPLETE, no process death, just a hang until the
watchdog cap, and learns nothing about the cause. -testagent removes those dialogs. With it
set, the binary:
abort() message box (no button to press, never hangs);<EVIDENCE_DIR>/app_stderr.txt, tagged [testagent];-testagent implies -debug).Do NOT key the crash decision on exit code. Breakpad handles the crash and the process usually
exits 0 — exactly as tdesktop's own crash detection assumes. The reliable crash signals are: the
process is gone WITHOUT a TEST_COMPLETE marker, AND a fresh non-empty
<workdir>/tdata/working exists. So always pass -testagent, and on a crash gather diagnostics
in this order before deciding the verdict:
<EVIDENCE_DIR>/app_stderr.txt — the [testagent] assert: … line gives the failed expression and
file:line (e.g. vector(1931) : … vector subscript out of range). Usually enough to localize.<workdir>/tdata/working — the crash report the reporter wrote: the Assertion: /
CrtAssert: annotations, the failed file:line, and Caught signal … / minidump id. Plain text;
read it directly. <workdir> is the launch -workdir (in portable test runs,
out/Debug/TelegramForcePortable/).<workdir>/tdata/dumps/*.dmp — the minidump (full stack, needs symbols to read; note its path
in test.md, don't try to symbolize inline).A crash is an IMPL_BUG (the implementation tripped an assertion / dereferenced out of range), not
a TEST_FLAW, unless the overlay itself is what reached out of bounds — quote the [testagent] line
and the tdata/working excerpt in test.md as evidence, and feed the expression + file:line to the
impl-fix agent as the Root cause / Fix hint. Only a crash with NO usable diagnostic after one retry
is UNRECOVERABLE.
A run that never reaches TEST_COMPLETE and never dies is a hang. Two independent guards catch it:
-testagent force-enables the built-in DeadlockDetector — a
ping thread that, if the main/event loop stops responding (a genuine deadlock or an infinite loop
on the UI thread), raises Unexpected("Deadlock found!") from a side thread. That crashes through
the same reporter, so the frozen main-thread stack is captured in the minidump and the process
exits on its own (key on the tdata/working report, not the exit code) — same diagnostics path as
a crash above. No agent action needed beyond reading tdata/working / the dump. Detection is
within ~30–90s of the stall.Core::Quit(). For that the runner enforces a
hard wall-clock deadline (~90s) from launch and, when it elapses, does the path-scoped kill
regardless of output. No legitimate auto-test runs anywhere near a minute, so this cap is pure
backstop — but it is what guarantees the agent can never wedge forever.Classify by which guard tripped: a DeadlockDetector crash with a real main-thread stack in app code
is an IMPL_BUG; the external cap firing is almost always a TEST_FLAW (the overlay didn't
drive to TEST_COMPLETE/quit) — re-author the overlay — unless the captured stack/log shows the
implementation itself wedged, in which case it is an IMPL_BUG. Two external-cap kills in a row with
the same signature → BLOCKED (early-escalation rule).
The on-disk EXE (out/Debug/Telegram.exe) always contains the compiled overlay after a test run.
Restoring source does not rewrite the binary. When the loop reaches a TERMINAL verdict (APPROVED,
BLOCKED, UNRECOVERABLE, or attempt cap), after the final path-scoped kill and exact-path source
restore, delete the built EXE so no overlay-laden test binary is left for
the user to launch by mistake:
Remove-Item -Force "$EXE"
A clean, feature-ready binary is one BUILD away on demand. (Delete only on terminal exit — between
attempts the next round rebuilds the overlay, so the binary is reused there.)
ASSESS decides APPROVED / TEST_FLAW / IMPL_BUG. Default to not approved; a check passes only on positive, specific evidence — in the captured pixels or the log — that the change is present AND correct.
visual.md geometry, named style/resource identities, or a
cited current/legacy analogue also qualify. Missing mockups alone never means the oracle is missing.<WORK_DIR>/test.md) — human-readable, append per attemptThe file the human opens to see how testing went. The test-author writes checks before running;
ASSESS fills Actual / Result and the verdict. Create one ## Attempt per implementation commit and
append one ### Run per execution. A TEST_FLAW adds a Run under the same Attempt; an IMPL_BUG fix
starts the next Attempt. Never overwrite history.
# Test report — <project>/<letter>: <title>
## Attempt <n>
### Run <m> — strategy <...> — driver <overlay|hybrid> — verdict <APPROVED|TEST_FLAW|IMPL_BUG|UNRECOVERABLE>
- Evidence directory: <EVIDENCE_DIR>
#### Test 1 — <aspect of THIS change>
- Expected: <observable effect the change should produce>
- Oracle: <what would make this check FAIL>
- Oracle source: <task fact / visual.md line / repo analogue / supplied image / baseline>
- Observed via: <surface + how captured: tight crop, geometry log, runtime state>
- Actual: <what is literally visible / logged>
- Screenshots: <after.png and any real reference crops; none only for a non-visual check>
- Result: PASS | FAIL
#### Test 2 — ...
#### Verdict reasoning
<1-3 lines tying the checks to the verdict>
#### Root cause / Fix hint (only if IMPL_BUG — the impl-fix agent reads this)
#### Failure signature (one line, for early-escalation comparison)
TASK: <TASK_ID>
STATUS: <DONE|BLOCKED>
VERDICT: <APPROVED|NOT_APPLICABLE|reason if blocked>
ATTEMPTS: <n>
TOUCHED: <repo paths or none>
DISCOVERED: <none|present in result.md|inline concise follow-ups when the wrapper has no result.md>
NOTES: <one or two lines, or none>
Detailed reasoning stays in external AI task artifacts. The chat reply is only this block.