lilleman
|
e55b8826fb
|
Merge remote-tracking branch 'origin/main' into item-310
# Conflicts:
# bin/agent_cycle/interrupt.py
# bin/agent_cycle/run.py
# bin/agent_cycle/states.py
# docs/reading-cost.md
# todo.md
|
2026-11-19 04:47:53 +01:00 |
|
lilleman
|
06f27feb3e
|
Answer stability round 1: 0 🔴, 6 🟡 (1 misread); name a checkout by its real path, pick a chunk's repo once, test the step line, the repo's model and ladder, and a clone
|
2026-11-18 22:50:30 +01:00 |
|
lilleman
|
aaa080a201
|
Answer product-owner round 1: 0 🔴, 2 🟡 (0 misread); say loop-claude hands only the Bash ceiling
|
2026-11-18 15:54:24 +01:00 |
|
lilleman
|
a7b2fd0ae1
|
Work in progress: Bash output ceiling and refusing Reads of session transcripts
|
2026-11-18 08:28:38 +01:00 |
|
lilleman
|
e60e891251
|
Compact a waiting worker's session shortly before its prompt cache expires
CI / tests (pull_request) Failing after 3m16s
|
2026-11-17 13:00:25 +01:00 |
|
lilleman
|
879dae4960
|
Merge remote-tracking branch 'origin/main' into item-306
CI / tests (pull_request) Successful in 3m6s
|
2026-11-17 02:50:24 +01:00 |
|
lilleman
|
b8a1a13ad0
|
Answer prose round 1: 0 🔴, 8 🟡 (0 misread); the help texts say where the runs go, the decision gets its own dated paragraph, and the comments reread
|
2026-11-17 02:33:40 +01:00 |
|
lilleman
|
eb64154384
|
Answer prose round 1: 0 🔴, 7 🟡 (0 misread); name the transcripts and review files the tool takes, scope the readers' claim, and rewrite the flagged comments
CI / tests (pull_request) Successful in 2m59s
|
2026-11-17 02:33:15 +01:00 |
|
lilleman
|
c7248165db
|
Answer stability round 1: 0 🔴, 1 🟡 (0 misread); the comment on a usage limit's reset time says the stream holds it, and the no-json message quotes the last line only
|
2026-11-17 02:33:04 +01:00 |
|
lilleman
|
b9876c17fe
|
Gzip loop-claude's archive and the review folders 30 days after each file was last written, and read gzipped files by their plain name
CI / tests (pull_request) Failing after 2m59s
|
2026-11-17 02:00:32 +01:00 |
|
lilleman
|
5aabbc87bf
|
Archive each review-replay and judge-replay run's transcript beside its result
CI / tests (pull_request) Failing after 2m52s
|
2026-11-17 01:49:49 +01:00 |
|
lilleman
|
64eb8fc7d7
|
Name the text-file helpers for what they hold, and keep what counts as a quote beside the judge
|
2026-11-13 02:49:07 +01:00 |
|
lilleman
|
14e6756a53
|
Answer prose round 1: 0 🔴, 6 🟡 (0 misread); rewrite each <verb> nothing to the negation rule, tighten the new docstrings and comments
|
2026-11-12 20:33:32 +01:00 |
|
lilleman
|
5c37027046
|
Answer architecture round 1: 1 🔴, 8 🟡 (0 misread); one reader per message, prices and Spent block moved, one table of larv-review commands
|
2026-11-12 20:05:52 +01:00 |
|
lilleman
|
d449c552c6
|
Answer stability round 1: 0 🔴, 2 🟡 (0 misread); keep the whole typed answer that holds the judge's quote of a reversal, and test a failing rev-parse
|
2026-11-12 10:25:30 +01:00 |
|
lilleman
|
3c4cedd389
|
Record with each archived verdict the setup that ran it, what each model spent by kind, and the human's reversals
CI / tests (pull_request) Failing after 2m12s
|
2026-11-12 09:57:56 +01:00 |
|
lilleman
|
09159d7158
|
Answer prose round 1: 0 🔴, 7 🟡 (0 misread); make the entry kinds' docstrings match the code, untangle after_mark and checked, and rename three tests to what they check
|
2026-11-09 00:52:37 +01:00 |
|
lilleman
|
7a5df16882
|
Answer stability round 1: 0 🔴, 5 🟡 (0 misread); read tool results from the API requests Claude Code logs again, test the fields every kind carries, mixed blocks, reads by tool and cwd, and indented or mixed security prompts; drop a cwd fallback that changes no commit found
|
2026-11-09 00:49:37 +01:00 |
|
lilleman
|
b83c1ddf19
|
Parse each transcript entry once into a typed value naming its kind, and have every consumer read those types
|
2026-11-09 00:12:13 +01:00 |
|
lilleman
|
0a0a2207c9
|
Open claude on a terminal with CLAUDE_CODE_SANDBOXED set, and stop writing ~/.claude.json
|
2026-10-30 21:31:44 +01:00 |
|
lilleman
|
c583900199
|
Answer product-owner round 1: 0 🔴, 1 🟡 (0 misread); stamp the transcript and trust lines through say
|
2026-10-28 18:47:39 +01:00 |
|
lilleman
|
b1a90abff4
|
Answer stability round 1: 1 🔴, 8 🟡 (0 misread); read only a regular file, only as far as the shown lines, skip cut-short output, and test the surviving mutants
|
2026-10-28 14:14:53 +01:00 |
|
lilleman
|
488d9b0eee
|
Answer principles round 1: 0 🔴, 2 🟡 (0 misread); kill the surviving mutants and split repeat-read's tests to one behaviour each
|
2026-10-28 13:05:00 +01:00 |
|
lilleman
|
248e46f616
|
Answer architecture round 1: 2 🔴, 2 🟡 (0 misread); move transcript parsing into claude_code and keep repeat_read pure
|
2026-10-28 12:52:34 +01:00 |
|
lilleman
|
14736e8e6e
|
Answer principles round 1: 0 🔴, 9 🟡 (0 misread); name the 200k window's compaction point, read a compaction summary as Claude Code's, remove item 231, refresh items 234, 235, 52 and 102, file item 237
|
2026-10-27 00:32:10 +01:00 |
|
lilleman
|
21d507f0c5
|
Answer principles round 1: 0 🔴, 1 🟡; give the true reason score-outcomes keeps its name, name archived_kind and three locals by what they hold
|
2026-10-23 11:09:51 +02:00 |
|
lilleman
|
387a5ce506
|
Answer architecture round 1: 0 🔴, 2 🟡; finish "record", record the three meanings as a decision, name rule-hook's check once
|
2026-10-23 10:54:24 +02:00 |
|
lilleman
|
6f98a1855b
|
Answer prose round 1: 0 🔴, 5 🟡; tighten the escalation mark's comments, item 221's detail, and fold three decision entries into item 190
|
2026-10-23 05:22:15 +02:00 |
|
lilleman
|
6733615e1c
|
Answer principles round 2: 0 🔴, 0 🟡; approved, file item 221 and fix two blank-line nits
|
2026-10-23 05:09:15 +02:00 |
|
lilleman
|
f781ac611b
|
Answer principles round 1: 0 🔴, 1 🟡; rename the module escalation_mark.py so it no longer shares the escalation field's name
|
2026-10-23 05:07:30 +02:00 |
|
lilleman
|
f0e14f3fdf
|
Hold the escalation mark in one module, and carry a ruling's step through the cycle's states
|
2026-10-23 04:45:03 +02:00 |
|
lilleman
|
e7dbff791a
|
Answer prose round 1: 0 🔴, 1 🟡; reword the recorded-answer paragraph, drop the repeated [loop] definition and the double possessive. The since_step comment's 🟢 is declined: the suggested wording claims the mark stays 0 until a step-up, which the code doesn't show
|
2026-10-22 01:29:15 +02:00 |
|
lilleman
|
eb17a4080d
|
Spell out the judge's prompt in judge.py and call the worker the worker in it
|
2026-10-22 01:18:08 +02:00 |
|
lilleman
|
023380f5a5
|
Answer prose round 1: 0 🔴, 13 🟡; split the exit-2 line, stop calling the claude session the interrupt, reword the interrupt's definition, the release_ready comment, the changelog and items 180 and 182, and file items 183 and 184
|
2026-10-17 23:55:59 +02:00 |
|
lilleman
|
e7f5443688
|
Answer principles round 2: 0 🔴, 0 🟡; name the typed answer's two parts in Typed
|
2026-10-17 23:41:59 +02:00 |
|
lilleman
|
4b2d4a1734
|
Answer principles round 1: 0 🔴, 6 🟡; correct the interrupt's definition, show the judge the question beside the typed answer, ask in one call returning Talk, add a Talking state, give the worker its own question past the cap, assert the prompts' facts, and keep stepped_from as the step's one owner
|
2026-10-17 23:32:21 +02:00 |
|
lilleman
|
831b9c63b6
|
Answer an interrupt on loop-claude's terminal and resume the worker headless with the answer
|
2026-10-17 23:04:28 +02:00 |
|
lilleman
|
89a7485684
|
Answer prose round 1: 0 🔴, 5 🟡; say the README's rung finds what the project's rung finds again, untangle the replay's help, sort the Tools list, word the decision's title plainly
|
2026-10-15 21:29:56 +02:00 |
|
lilleman
|
2752c2dae3
|
Answer stability round 1: 0 🔴, 7 🟡; quote a judge's stdout that is no json, name each left-out round and why, test the string prompt, the miss sum, the args sent to claude and the skip paths, settle the judge's hard place in item 122
|
2026-10-15 21:19:31 +02:00 |
|
lilleman
|
c2a4757d56
|
Answer architecture round 2: 1 🔴, 5 🟡; return a gone checkout's git failure as a value, derive a round's seconds from its stamp, let claude_reply own the flags it parses, test the uncovered paths, warn to sandbox the replay, decline injecting the shell's filesystem
|
2026-10-15 20:57:22 +02:00 |
|
lilleman
|
4f722d94f4
|
Answer architecture round 1: 1 🔴, 2 🟡; share one rung search between both replays, searching down from the project's rung, name the sixth git runner in item 114, decline recording the reviewed commit
|
2026-10-15 20:52:50 +02:00 |
|
lilleman
|
acdba1abf4
|
Answer architecture round 1: 0 🔴, 12 🟡; break the import loop, share the interrupt hand-over and formats, measure under a lock 20 cycles past the last measurement, go on off a terminal, and add a reset
|
2026-10-15 00:49:02 +02:00 |
|
lilleman
|
b1913ed65d
|
Restructure bin/ so the project reads at the comprehension floor
|
2026-10-09 14:12:16 +02:00 |
|