Answer prose round 1: 0 🔴, 2 🟡 (0 misread); reword the timer's docstrings, comments and help, rename the no-compaction test

This commit is contained in:
2026-11-19 04:44:37 +01:00
parent 708347382b
commit a2ea673822
4 changed files with 9 additions and 9 deletions
+2 -2
View File
@@ -21,7 +21,7 @@ ANSWER_QUESTION = "answer the question above"
# TALK, what the human types at loop-claude's terminal to open claude for the question, and ASK_ANSWER were picked on 2026-10-17 by a bench of three readers seated as the README's audience, as ~/.claude/CLAUDE.md → "Working loop" → "Bench a syntax with the people who write it." has it.
TALK = "/talk"
ASK_ANSWER = f"Answer ({TALK} opens claude, Enter stops the loop): "
# While the human hasn't answered, the worker's session is compacted CACHE_MARGIN seconds before its hour-long prompt cache expires: compacting reads the context from the cache, while resuming after it expired writes the whole context again. Below COMPACT_FLOOR tokens of context, that write costs less than compacting. A five-minute cache expires within the margin, so it is left as it is. lilleman chose this on 2026-11-11.
# While the human hasn't answered, the worker's session is compacted CACHE_MARGIN seconds before its hour-long prompt cache expires: compacting reads the context from the cache, while resuming after it expired writes the whole context again. Below COMPACT_FLOOR tokens of context, that write costs less than compacting. A five-minute cache expires within the margin, so a session that last wrote one isn't compacted. lilleman chose this on 2026-11-11.
HOUR_CACHE = 3600
CACHE_MARGIN = 600
COMPACT_FLOOR = 60_000
@@ -75,7 +75,7 @@ def answered(answer_here: bool, session: WorkerSession, effects: Effects) -> str
def waited_answer(session: WorkerSession, effects: Effects) -> str | Stop:
"""typed_answer, compacting the worker's session where no answer has come shortly before its prompt cache expires. The claude TALK opens compacts an idle session itself."""
"""typed_answer, compacting the worker's session where no answer has come by CACHE_MARGIN seconds before its prompt cache expires. /talk needs no compaction: the claude it opens compacts an idle session itself."""
entries = read_transcript(session.transcript_path)
now = effects.now()
delay = None if isinstance(entries, Failure) else compaction_delay(entries, now)
+4 -4
View File
@@ -58,16 +58,16 @@ there:
- /talk: opens the worker's claude session with `claude --resume` on the terminal, to read the diff
or discuss first. Once that claude exits, the judge rules on what was typed there.
- Enter alone: stops the loop.
Where no answer has come 10 minutes before the worker's hour-long prompt cache expires, and its
context holds at least 60k tokens, loop-claude compacts the worker's session with
`claude -p /compact` and keeps waiting, so the resume doesn't write the whole context again.
A release ready to cut opens that claude session on the terminal at once, since only the human may
cut it. While claude is open, Ctrl-C reaches it alone, and the loop goes on. Where the project's
config sets notifier, loop-claude also sends the human each question, naming its tmux window as
<session>:<window>, each release ready to cut and each stop as a DM, as bin/notify sends it. Four
stops are left out: Ctrl-C, printing this help, the loop-claude that started the windows once it
attaches or prints how to, and a loop whose chunk finished once it started the next chunks. A DM
that fails is said on stderr, and the loop goes on.
that fails is said on stderr, and the loop goes on. Where no answer has come 10 minutes before the
worker's hour-long prompt cache expires, and the worker's context holds at least 60k tokens,
loop-claude compacts the worker's session with `claude -p /compact` and keeps waiting, so the resume
doesn't write the whole context again.
As a chunk starts, before it is routed, loop-claude runs bin/chunk-preflight --worktree, adding
--item <ID> in a window named for an item, and quotes what it prints in the first prompt
+2 -2
View File
@@ -838,7 +838,7 @@ class Interrupts(unittest.TestCase):
handle.write(json.dumps(said) + "\n")
def fired(self, fake, answer):
"""An input the human types answer at, after every timer started has fired."""
"""An input that first fires every timer started, then returns answer as the line the human typed."""
def typed(prompt):
for _, action in fake.timers:
action()
@@ -895,7 +895,7 @@ class Interrupts(unittest.TestCase):
self.assertEqual([seam for seam, _ in fake.calls if seam in ("memory", "sleep", "stream_claude")], ["stream_claude"])
self.assertNotIn("waiting to start the worker", said)
def test_an_answer_before_the_timer_fires_stops_it_and_compacts_nothing(self):
def test_an_answer_before_the_timer_fires_stops_it_and_doesnt_compact_anything(self):
self.worked(minutes_ago=10)
fake = FakeEffects(inputs=["Use B."])
self.assertEqual(self.put(fake)[0], "Use B.")
+1 -1
View File
@@ -119,7 +119,7 @@ def run_python(args: list[str], cwd: str, timeout: float) -> subprocess.Complete
def start_timer(delay: float, action: typing.Callable[[], None]) -> typing.Callable[[], None]:
"""Runs action on a thread of its own once delay seconds have passed, and returns what stops the timer: it cancels an action not yet started, and waits for one that has started to end."""
timer = threading.Timer(delay, action)
# A loop that exits while it waits doesn't wait for the action.
# A timer still running doesn't keep the process alive once the loop exits.
timer.daemon = True
timer.start()