2026-08-23 - 2026-11-23
Overview
674 Pull requests merged by 1 user
Merged
#693 Record the spread of the reads and the turns beside the tokens' in the task-bench decision
Merged
#692 Print the mean of the bugs' own changes as task-bench's All change
Merged
#691 Stop the bench with a Failure on a run that ended on an api_error without a status
Merged
#689 Calibrate task-bench on fejkdata and record how many tasks and runs a bench takes
Merged
#688 Cut release 0.15.0
Merged
#686 File item 512: make stability-reviewer check that composed lookup keys cannot collide
Merged
#685 Limit goal 3's backup of a bench's transcripts to the time before its gather
Merged
#684 Distil each task-bench run into a digest and gather plan items from it
Merged
#683 Add task-bench, which has Haiku fix planted bugs at a base and a head and prints what each side took
Merged
#682 Run each command a changed doc line says prints something and hand the real output to the product-owner and prose reviewers
Merged
#680 Move a do-epic parent issue whose sub-tasks a person filed once they have landed or are done
Merged
#678 Mutate the Go lines a branch changes with Gremlins in mutate-changes
Merged
#681 File item 511: run the commands a changed doc line says print something
Merged
#679 Record that a Claude home of its own loads the setup and signs in by a long-lived token, but not under --bare
Merged
#673 Let a project's config name a working branch per epic that chunks branch off and merge into
Merged
#677 Block items 469 and 479 on item 472, whose open branch changes the same files
Merged
#676 Plan task-bench: the MVP and Go in mutate-changes in 0.15.0, follow-ups in 0.16.0
Merged
#675 Plan the standalone install as releases 0.16.0 to 0.21.0
Merged
#674 Run mutate-changes on a script its tests import by file name, and name a crash inside mutmut in one line
Merged
#670 Start a do-epic chunk's issue in Jira as its team's board expects
Merged
#672 Move item 474 to 0.15.0 so mutation tests run in docs-and-tools before TW-13503
Merged
#671 Load no MCP server in the bench's headless claude calls that are denied tools
Merged
#669 Load no MCP server in a headless claude call that is denied tools
Merged
#666 Start an epic's runnable issues in order of score, scoring each open issue that holds none
Merged
#668 File item 475: have stability-reviewer report a lookup outside data can make fail without naming what was missing
Merged
#667 File item 474: run mutate-changes where tests import a script by its file name
Merged
#665 File item 472: chunks merge into a working branch per epic, which a person merges into main
Merged
#664 Move item 469 to 0.15.1 so 0.15.0 ships sooner
Merged
#660 File findings, release reviews and added items as the issue types the settings name, and refuse a type the Jira project lacks
Merged
#659 Name the team's label for each repo in do-epic's finding and release-review prompts
Merged
#662 File item 471: ship jira-rest as a claude-config tool
Merged
#661 File items 469 and 470: move do-epic's parent issues and start issues by score
Merged
#657 Record do-epic's repo picks and sub-task summaries in the team's own spelling
Merged
#656 Add do-epic --dry-run, which prints the Jira writes planning would make and the chunks that would start, and makes none
Merged
#658 File item 467: cut 0.15.0 before its release review, and run that review in 0.15.1
Merged
#655 Have stability-reviewer check a separator split against a value holding the separator
Merged
#654 Cut release 0.14.0 with the judge's transcript fix, and move its open items to 0.15.0
Merged
#653 File item 466: stability-reviewer checks a separator split against a value holding the separator
Merged
#651 Show the judge the transcript's tail whatever the size of the records, test the forge timeout and decide the merge stop from the pull requests found
Merged
#650 File items 464 and 465 from item 457's mutation run and paired ruling
Merged
#645 Pass over an item whose pull request awaits a person's merge
Merged
#647 Review the judge's fix for quoting the worker's task as evidence, and close the gaps that review found
Merged
#644 Let do-epic's config name the statuses that unblock an issue and leave the moves to a team's automation
Merged
#649 File item 463: show the judge the transcript's tail whatever the size of the records
Merged
#648 File item 462: run under mutate-changes the tests mutmut's copy leaves out
Merged
#643 Resume a chunk that awaits a person's merge once its pull request no longer merges into its base
Merged
#646 Take the scoring run's values from seat-prompts.md where the working rules hold none
Merged
#642 Cut release 0.13.0 with the judge's fix, and move its open items to 0.14.0
Merged
#641 Let the judge quote the worker's task as evidence
Merged
#640 File item 459: a sentence about several named things holds for each
Merged
#639 Add the project setting merge = "human", under which chunk-pr merge leaves the pull request ready and a do-epic chunk parks on your merge
Merged
#637 Read a team's own repo labels as do-epic's picks, and leave an issue with a person's sub-tasks unsplit
Merged
#638 Let do-epic --file take --blocked-by and --parent, so a change a repo ships as two pull requests gets an issue for each
Merged
#636 Clear 0.13.0 for do-epic's fixes and file items 448 to 455
Merged
#635 Cut release 0.12.0
Merged
#634 Release review of 0.12.0
Merged
#633 Judge a worker that stalls before its result with a headless claude, and report the verdict without a kill
Merged
#611 End a worker that lingers after its final result by stopping the quiet background shells it left
Merged
#632 Remove item 310 again, which merging main into item-442 brought back, and file item 445 to refuse such a plan
Merged
#631 Run do-epic end to end against a sandbox Jira epic with do-epic-e2e
Merged
#630 Plan an epic's open issues before do-epic starts chunks
Merged
#587 Compact a waiting worker's session before its prompt cache expires
Merged
#629 Add a do-epic window that DMs the human and waits for a human-task issue that blocks the epic
Merged
#628 Tell an epic chunk's worker how its repo ships a change, and bound the interrupt test's helper shell
Merged
#626 Print each chunk that asks a question and since when each parked chunk awaits its gate in do-epic --status
Merged
#622 Move an epic's issue to done in Jira when its chunk finishes
Merged
#627 Describe do-epic in the README's tools and setup
Merged
#625 Finish an epic: review each repo it changed, have the human confirm each goal, post a summary and close it
Merged
#624 Park a chunk whose pull request waits on a long gate, and resume it when the gate opens
Merged
#623 File a do-epic chunk's findings and decisions in its epic, and give its judge the epic's records
Merged
#619 Run do-epic <EPIC-KEY>: a tmux session per epic and a window per chunk
Merged
#621 Give the stalled-fetch test a timeout that a loaded runner fits in
Merged
#620 File item 439: the stalled-fetch test fails on a loaded runner
Merged
#617 Refuse an endless wait in rule-hook
Merged
#618 Bring nine lessons from a client setup into the principles, the reviewers, larv-review and the project config
Merged
#616 Re-rank Jira's backlog to match an epic's ranking, moving only the issues the loop filed
Merged
#584 Launch comprehension seats as an agent type of their own, without CLAUDE.md
Merged
#613 Split an epic item that changes several repos into a sub-task per repo
Merged
#612 Read each epic issue's scores, blockers and whether the loop filed it, from both backends
Merged
#615 Fail the tests that skipped at run time, naming what they miss, and give CI's tests docker
Merged
#614 File item 429: lessons from a client setup for the principles, reviewers and larv-review
Merged
#606 Forbid a test that decides at run time whether to check anything
Merged
#609 Move loop-claude's cycle, judge, interrupt, tmux and routing into agent_cycle
Merged
#610 Score an epic's items by one chain of goals and order them by score, then blockers
Merged
#607 Refuse a Touches: list touch-predictions cannot read in todo-item add and edit, and in rank_todo.py --check
Merged
#608 Keep the routing ladder on the call of 2026-11-18
Merged
#605 Hand each worker a Bash output ceiling and refuse a Read of a session transcript
Merged
#599 Pick each epic item's repo by repo-rule.md and record the pick as a label on the issue
Merged
#604 Make the Touches lists of items 400, 401, 416, 420, 421 and 422 readable
Merged
#603 Record lilleman's ruling that the offline classifier may decide finished
Merged
#601 Cancel a pull request's older CI runs when it gets a new push
Merged
#585 Replay the archived judge verdicts offline through code rules and a classifier trained on the archive
Merged
#598 Find an epic repo's checkout and project config by its name, and clone a missing one from the place of the epic's other repos
Merged
#600 File items 420, 421 and 422
Merged
#593 Read do-epic's settings and repo rule from its config folder, refusing to start when either is missing or misspelt
Merged
#595 Write a Jira epic's description by section, and change its Goals and Audience only once the human approves
Merged
#597 File item 419: run the external auto-reviewer until its review of the head has no findings
Merged
#596 Add MIT license for Larv IT AB
Merged
#594 Read and file an epic's child issues and sub-tasks through a plan backend interface, with a Jira backend and an in-memory fake
Merged
#591 Reflect on each outside reviewer's fix and change the setup so review would have caught it
Merged
#592 Rerun the pass that raised a 🟡 when its fix changes what a test asserts or whether it runs
Merged
#590 Drop item 405: answered, the sandbox is larvit.atlassian.net's SCRUM project
Merged
#589 Plan do-epic into release 0.12.0 as items 401 to 418, and record its design
Merged
#588 File items 398, 399 and 400, and move 339, 341 and 298 to 0.13.0
Merged
#586 Block item 341 on item 311, whose chunk builds the replay's Change
Merged
#583 Print one section of a markdown file, picked by a fragment of its heading, with markdown-section
Merged
#582 Audit 0.9.0's chunks: file 1 scripted step, and the audit again for 0.15.0
Merged
#581 Bind a project's rung to its ladder in ProjectRung, and step an L 3 worker one rung up and an L 4 worker two up
Merged
#580 Have larv-review round-title print a round commit's counts from its report, and rule-hook refuse a title that differs
Merged
#579 Move item 309 to 0.13.0, since agreement has 6 of the 10 commits it needs scored twice
Merged
#578 Drop item 319: its bench did not favour a fresh session over resuming
Merged
#577 Move item 395 to release 0.13.0, as lilleman chose
Merged
#576 Set a Jira issue's blocked-by links from todo-item edit --blocked-by on an issue the loop filed
Merged
#574 Read a Touches list from a Jira description, as the codec spells it, and reword the expiry and goal citations of six decisions
Merged
#575 Bench a model step's handoff, and keep resuming the session
Merged
#573 Leave a Jira description or a Confluence page as it is where a save hands back the text a read showed
Merged
#572 Gzip loop-claude's archive and the review folders 30 days after each file last changed, and read gzipped files by their plain name
Merged
#571 Archive each review-replay and judge-replay run transcript beside its result
Merged
#570 Predict touches for release 0.12.0 items
Merged
#568 Archive each judge run's prompt and reply beside its verdict, and replay judge-replay on the archived prompt
Merged
#569 Move the forge helpers into bin/forge/, which chunk_preflight, chunk_pr, deploy_release and larv_review import
Merged
#567 Write no changelog bullet for a plan kept in Jira: todo-item finish there needs no changelog flag and refuses --changelog
Merged
#566 Make the Touches lists of items 303, 306, 287 and 387 readable: mark only named parts small
Merged
#565 File item 394 and move items 303, 306, 304, 287, 387, 338 and 288 into 0.12.0
Merged
#564 Cut release 0.11.0
Merged
#563 Record that the tools send no OTLP traces, logs or metrics, and loop-claude's archive is the trace of each cycle
Merged
#562 Release review of 0.11.0
Merged
#561 Convert ADF to markdown and back through adf-codec's lossless pair
Merged
#560 Run each chunk in its own tmux window, up to parallel-chunks at once, and mark a window that waits
Merged
#559 Rank an issue the loop files in Jira by its score among the issues the loop filed, a security finding first
Merged
#558 Edit a Jira issue the loop filed, and score a person's with a comment saying why
Merged
#557 Record that the loop reads no other mark of a cut release yet
Merged
#556 Keep working when someone else cuts a release that still holds items
Merged
#555 Key project-record's copies per worktree
Merged
#554 Merge a project's chunks one at a time, finishing each one's plan items in its turn
Merged
#553 Read the routing ladder from routing-ladder.json, which a project may replace with a file of its own
Merged
#552 Key loop-claude's session per slot, so two loops of one project run side by side
Merged
#551 Drop item 385: finish fails where no direct transition leads to the done status
Merged
#550 Move a finished item of a Jira plan by the transition and to the status the config names
Merged
#549 Hand reviewers the review standards from the folder a project's config names
Merged
#548 Store a record on a Confluence Cloud page, and never overwrite a save made since the loop read it
Merged
#547 Record the answer to item 384: a plan in Jira states its formula and bar in config.toml
Merged
#546 File a plan item as a child issue of the project's Jira epic
Merged
#545 Pick the chunks that may run at once by predicted parts, and record each prediction beside what the merged chunk touched
Merged
#544 List the plan for reviewers from the store that holds it
Merged
#543 Name each project's notifier in its config, with Mattermost and Slack as backends
Merged
#542 Let a project's config allow its tooling on the host and set its readability floor
Merged
#541 Take a project's goals and audience from a Jira epic's description where its config names one
Merged
#540 Finish item 378: name a Jira plan's chunks by the issue key through the branch and title templates
Merged
#539 Convert every description of a plan read in one adf-codec container run, and take a heavy slot around the build of the codec's image
Merged
#538 File item 380: define the plan in Vocabulary whatever holds it
Merged
#537 Read a project's plan from the child issues of a Jira Cloud epic
Merged
#536 Convert markdown to ADF and back through a pinned adf-codec shim in a container
Merged
#535 Route the plan through one plan-store interface and todo-item
Merged
#534 Record lilleman's ruling on the routing calls of 2026-11-14: L 3 one rung up, L 4 two up
Merged
#533 Write prose in the language a project's config names, and title commits, pull requests and branches by its templates
Merged
#532 Make chunk-pr merge wait for the human approval a project's config requires
Merged
#531 Propose changes to a client's CLAUDE.md and AGENTS.md to its team, and ask the principles for files in folders before banners and for a findable entry point
Merged
#530 Read and write each record through the store its config names
Merged
#529 Remove loop-inbox, so an interrupt is answered at loop-claude's terminal only
Merged
#527 Refuse a Bash command that recursively searches a folder holding an env-file
Merged
#528 File item 376: ask for files in folders before comment banners, and a findable entry point
Merged
#526 Name the tmux session and window in each interrupt's DM, in item 375
Merged
#525 Drop item 361: superseded by items 371 to 375, which remove loop-inbox and run chunks in parallel in tmux
Merged
#524 Refuse a Read, Grep or Bash call that reads ~/.claude/.env or a config's env-file
Merged
#523 Read a project's credentials from the env file its config names
Merged
#522 Cut release 0.10.0
Merged
#521 Release review of 0.10.0
Merged
#520 Commit a chunk's changes to the config's folder in its own repo
Merged
#519 Read and write the plan, AGENTS.md, the records and the changelog at the paths a project's config names
Merged
#518 Read a project's goals and audience from the file its config names
Merged
#517 Hand reviewers the technical principles from the folder a project's config names
Merged
#516 Read a project's settings from the config that names its checkout
Merged
#515 Plan 0.10.0 as the client MVP, 0.11.0 as the rest of the pluggability work, and revise the goals and audience for client work
Merged
#514 Word two README goals with the negation after the verb
Merged
#513 Call a scoring run only for an item whose tagline opens with the comprehension floor phrase
Merged
#512 Cut release 0.9.0
Merged
#511 Release review of 0.9.0
Merged
#510 Move every planned item one release later, and the unreleased work's items into 0.10.0
Merged
#509 Start five reviewers without CLAUDE.md, gated by a replay with a control
Merged
#508 Commit the stability skip count that item 308s decision rests on
Merged
#507 Record lilleman's answers to items 340 and 331
Merged
#506 Skip a stability round 2 whose every 🔴 named a surviving mutant or a failing test
Merged
#505 318 - Qualify a review-replay rung by the project's rung's own recall
Merged
#504 Record with each archived verdict the setup that ran it, what each model spent by kind, and the human's reversals
Merged
#503 Add run-tests, wait-for and larv-review prepare-scoring; have chunk-pr sync, set the title on ready and wait for checks on merge; let deploy-release move past a model pick
Merged
#502 Research the tooling agents read with, show an item with todo-item show, and save calls with command-patterns --calls
Merged
#501 Start principles-auditor without CLAUDE.md, and gate every other reviewer on a replay
Merged
#500 Bench the advisor tool on six archived L 1 and L 2 chunks: Sonnet alone passes as often as Opus at medium for less, so workers stay without it
Merged
#499 Run the judge of an Opus 5.5 project on Opus 5.5 at xhigh in every chunk
Merged
#498 Audit 0.8.0's chunks: file 4 scripted steps, and the audit again for 0.10.0
Merged
#497 Replay the judge on the Fable 5.2 rungs a review climbs to, and keep it on Opus 5.5 rungs after each missed an interrupt
Merged
#496 File the deploy-release checkout refusal on a reordered settings.json
Merged
#495 Move item 328 back to 0.9.0
Merged
#494 Cut release 0.8.0
Merged
#493 Rebuild the routing ladder for Fable 5.2
Merged
#492 Drop item 313
Merged
#491 DM the human on Mattermost when loop-claude needs them
Merged
#490 Items 270, 317: checks before each stability round, one-line reviewer descriptions
Merged
#489 Finish item 284: file items 325 to 327 from the 0.7.0 audit, and the 0.8.0 audit as item 328
Merged
#488 Run mutation tests on a chunk's changed lines before review with bin/mutate-changes
Merged
#487 Have loop-claude read the items a chunk takes from the worker's Chunk items: line
Merged
#486 Fold AGENTS.md into where each part fits, and load CLAUDE.md once in this repo
Merged
#484 Distil the decisions AGENTS.md lists into seven principles, and check each new decision against them
Merged
#485 Move items 302, 303, 304 and 306 to 0.9.0
Merged
#483 Ban a verb with nothing as its object in the prose rules
Merged
#482 File items 314 to 321 and revise items 307 and 312
Merged
#481 File items 305 to 313 from the token research
Merged
#480 File item 304: archive replay transcripts
Merged
#478 Check the files a commit would commit against rule-hook's file rules, and sort .gitignore's allowlist block
Merged
#479 File item 303: archive the judge's prompt and reply
Merged
#477 Plan item 302: call a scoring run only for an item that targets the comprehension floor
Merged
#476 Record the comprehension check: PR #475 at 52da15e means 6.38 overall, run late for item 301; refresh floor item 300
Merged
#475 Reword the blocks-no-merge decision titles to say no score blocks a merge
Merged
#474 Plan item 301: reword the blocks-no-merge decision titles, a question for the human
Merged
#473 Refuse chunk-pr merge while a scoring run the plan calls is unrecorded, and call one for a release cut with none
Merged
#472 Move items 284 and 270 to 0.8.0
Merged
#471 Plan items 293 to 299 for 0.8.0, raise 264, 268 and 270, and drop item 269
Merged
#470 Drop item 52, shift 0.8.0 and 0.9.0 one release later, and plan the Fable 5.2 ladder for 0.8.0
Merged
#469 Cut release 0.7.0
Merged
#467 Release review of 0.7.0
Merged
#468 Move item 291 to 0.8.0
Merged
#466 File item 291: an exception in Sane defaults for a no-config library's clock and randomness
Merged
#465 Move item 1 to 0.9.0
Merged
#464 Drop item 277
Merged
#463 Give the test commands their own README section
Merged
#462 Read the branch's reflog in rule-hook overrides when a path shares its name
Merged
#461 Remove a merged chunk's worktree and branch and fast-forward the main checkout with chunk-pr clean
Merged
#460 Pick each replay's commits by the newest author time, in place of the first such commit in log order
Merged
#459 Record that the maintain weight of 0.5 stays fixed until a project's ranking shows it distorted
Merged
#458 Keep the presentation of loop-claude, with the script that prints it to PDF
Merged
#457 File item 290: read the branch's reflog in rule-hook overrides when a path shares its name
Merged
#456 Keep the reviewer-merge trials, record that each pass keeps its own reviewer, and file the review-replay commit-choice defect
Merged
#455 Ask code health to hold for years and debugging, and goal 1.4 to keep lowering cost
Merged
#454 File item 287: move the forge helpers into one module
Merged
#453 Let an item name a maintain goal by the confidence a test, gate leg or probe adds, weighed 0.5 per point
Merged
#452 Quote the detail of each item that may join a chunk in chunk-preflight, so a worker stops opening todo.md for it
Merged
#451 Finish items 12, 272 and 283
Merged
#450 Run the suite in CI on every pull request; deploy-release --cut reports each step and deletes the release branch; add command-patterns
Merged
#449 Write the branch diff and places into the review folder and print each reviewer's prompt header with larv-review reviewer-header
Merged
#448 Open, mark ready and merge a chunk's pull request with chunk-pr, which picks tea or gh from the remote
Merged
#447 Report L by the share of chunks with a 🔴 in score-outcomes where most hold none
Merged
#446 File and finish todo.md items with todo-item add and todo-item finish
Merged
#445 Let a worker fix a quality finding whose fix changes only prose on its own rung
Merged
#444 Count only built items as shipped, check this repo's branches with their own rank_todo.py, and test the replays' model, floor and checkout
Merged
#443 Finish the preflight comparison: reading cost fell by half, rounds and 🔴 did not rise, nothing to file
Merged
#442 Record the worker decision's check: rounds per pass fell, reading cost rose, and file two items for it
Merged
#441 Parse each transcript entry once into a typed value naming its kind
Merged
#440 Audit 0.2.0's chunks for steps a script can take: file six items, and the audit again for 0.7.0
Merged
#439 Have the judge reject an interrupt that re-asks what the human answered earlier, unless it quotes what that answer left open
Merged
#438 Show the size of the code each finished item names beside its S in score-outcomes
Merged
#437 Run each table-driven test case in its own subTest and split the tests that check several behaviours
Merged
#436 Pass the judge a Brief, decide read_seats from its inputs and return a failed bench run as a line
Merged
#435 Decode a reload's handed JSON with a decoder for LoopRun and its Route
Merged
#434 File or refresh one comprehension-floor item whenever a scoring run is under 7.0
Merged
#433 Let a worker take its own recommendation on a choice no rule reserves for the human
Merged
#432 Group bin's root modules into packages named by purpose
Merged
#431 Move loop-claude's help out of loop/main.py and wrap it at 100 columns
Merged
#430 Set F in todo.md's score formula to 32, measured from loop-claude's archive
Merged
#429 Split the routing figures by concern and keep their state as fields; check the decision titles AGENTS.md lists and todo.md cites
Merged
#428 Parse each comprehension history entry once into votes, scores and listings
Merged
#427 Split the todo.md scoring decision into one paragraph per choice
Merged
#426 Name the project's rung project in judge_replay, subagent_rung and review_replay
Merged
#424 Drop LEGACY_WORKER so route_from reads only JSON worker lines
Merged
#425 File item 274: let an item name a maintain goal it raises confidence in
Merged
#423 Keep the routing steps, drop item 170 and file item 273
Merged
#422 Name code health in item 155's Moves
Merged
#421 Refresh item 155 with the reading cost after item 229's chunk and move it to 0.7.0
Merged
#420 Keep the rule for which commands reach run-heavy's docker beside that docker in capped_docker.py
Merged
#419 File item 272: clearer deploy-release --cut output and delete the merged release branch
Merged
#418 Cut release 0.6.0
Merged
#417 Have deploy-release --cut cut the next release through a pull request, then tag and deploy it
Merged
#416 File items 261 to 271 and plan 0.6.0, 0.7.0 and 0.8.0
Merged
#415 Release review of 0.6.0
Merged
#414 Count a dated CHANGELOG heading as cut without a tag, read only regular files in rule-hook, stamp asks in ISO UTC, give workers the reading rule
Merged
#412 Count the judge's cost in the routing figures
Merged
#413 Name the rules no marker overrides via rule-hook --help in the rule-hook decision
Merged
#411 File item 259: chunk-preflight ignores a changelog's release heading, and its help misstates an untagged branch
Merged
#410 Keep the routing steps after the call on Sonnet 5.5's review rounds
Merged
#409 Make chunk-preflight create the item's worktree only with --worktree, which loop-claude passes
Merged
#408 Open claude on a terminal with CLAUDE_CODE_SANDBOXED set, and stop writing ~/.claude.json
Merged
#407 Refuse a security-review or claude-security Skill call below the project's rung
Merged
#406 Remove done items 250, 253, 256 and 258 from the plan
Merged
#405 Move repeat-read's reads into repeat_read.py, split waiting.py by role, name an untagged release's review, fix release-review findings in-release
Merged
#404 Move six items to 0.6.0 and shift 0.6.0 and 0.7.0 up one release
Merged
#403 Plan item 258: have a release review fix its findings in its own release
Merged
#402 File items 256 and 257 for chunk-preflight's skipped first-release review and its stray worktrees
Merged
#401 Cut release 0.4.0
Merged
#400 Deploy releases to ~/.claude with deploy-release and develop in ~/code/claude-config
Merged
#397 List every running loop in loop-inbox, with the items its chunk takes
Merged
#399 Run the architecture and principles passes only in a release review before each cut
Merged
#398 Take Haiku 5.5 off the routing ladder and review prose-only changes lightly
Merged
#396 List a loop's question in loop-inbox only while the loop holds the session's lock
Merged
#394 Show every waiting loop-claude in loop-inbox and take each one's answer there
Merged
#395 Count a file read in chunk-preflight's worktree in reading-cost's most-read list
Merged
#393 Read the comprehension history's changes from the checkout's top in larv-review
Merged
#392 Count each finished chunk's later fixes in score-outcomes and the routing figures
Merged
#391 Open every line loop-claude prints with the machine's local time
Merged
#390 Give each review pass at most two rounds per chunk, with no third round and no ask for one
Merged
#387 Block a read of file lines already in the agent's context, unless a compaction dropped them or the file changed
Merged
#389 Reword the note that prose findings stay
Merged
#388 File items 246 and 247: cap review at two rounds, count escaped defects; record that prose findings stay
Merged
#386 Run the tests once, on the latest stable Python, in one heavy slot's four workers
Merged
#382 Report the routing figures in dollars, cache reads included, from the cost each archived run records
Merged
#385 File item 245: run the tests once, on the latest stable Python, across the CPUs
Merged
#384 File item 244: stamp each line loop-claude prints with local time
Merged
#383 File item 243: answer every waiting loop-claude from one terminal app
Merged
#381 Record lilleman's choice to try one principles auditor despite the trial's two misses
Merged
#380 Run one principles-auditor per round for the goals and every principles file that applies
Merged
#379 Have the worker enter the chunk's worktree before the security pass, so the skill scans the branch's diff
Merged
#378 Record in the comment at UNFINISHED_LIMIT that the human confirmed 6
Merged
#371 Stop loop-claude and ask the human after six unfinished verdicts running
Merged
#377 Launch a round's reviewers in up to three waves, each in one message
Merged
#376 Give the worker's claude a 200k auto-compact window, from item 231's evaluation of context cost, and file items 235 to 241
Merged
#375 Move each release forward one, and file items 231 to 234 into 0.4.0
Merged
#374 File item 230: an ambitious refactor whenever a scoring run is under 7.0
Merged
#373 Cap the memory of each container a heavy command makes under run-heavy at its slot's share
Merged
#372 Block a docker run or docker create that sets no memory limit, and read docker's TLS and buildx --builder options as taking a value
Merged
#370 Move the run-heavy rule's command parsing into its own module
Merged
#369 Renumber the decision-paragraph split from item 222 to 224, which item-222's chunk had also used
Merged
#366 Hold each heavy command for one of a heavy slot per 4 CPUs, and each worker until a quarter of memory is available
Merged
#368 Count, per finished chunk, the findings that arose from misreading code that had already shipped
Merged
#367 Give "record", "outcome" and "places" one meaning each across bin/, and rename larv_review/git.py by what it holds
Merged
#365 Keep item 218 blocked on item 170 whether or not 170 changes the routing steps
Merged
#364 Keep the routing steps as they are, and re-tune them once 50 chunks have shipped under them
Merged
#363 Say in larv-review's skill that the auto-reviewer's round commit takes no round title
Merged
#362 Sort the decision log by title, and rewrite its six amendment-log entries as the current state
Merged
#361 Hold the escalation mark in one module, and carry a ruling's step-up through the cycle's states
Merged
#360 Have rule-hook refuse a round commit past round 3 unless an override quotes the human answer
Merged
#359 Split review_launches into model_routing's choice of each review's rung and whole prompt sentences that render it
Merged
#358 State each routing rule once in the docstring of the code that enforces it, cut loop-claude --help's routing paragraph, and name the project's rung project in bin/loop_claude
Merged
#357 Archive the route as a JSON record loop-claude reads back field by field, and name the stepping functions and the judge's taken-in levels by what they do
Merged
#356 Compute the next release, and whether work has landed towards it, in one function that chunk_pick and larv-review's scoring trigger share
Merged
#355 Bundle every unblocked item of a release that fires the same review passes, whatever its purpose, and ship each item above R 5 alone
Merged
#354 List larv-review's passes in one place that round_titles and the ladder derive from, and check the pass table against it
Merged
#353 Move the 9 from the Shape items to the items that lift Locality, and split the floor item by place
Merged
#352 Read the status line's branch through git_repo.git, and state the comprehension-floor rule as it holds
Merged
#351 Move the 0.5.0 items scoring 14 or more to 0.4.0
Merged
#350 Add the principle of about seven things per level, and have the architect read its limits from it
Merged
#349 Park a session's files outside its checkout in the chunk's review folder, since a headless session has no scratchpad
Merged
#348 Keep a command's words together across a redirect in shell_command's segments
Merged
#347 Move the vote counting out of larv-review's votes, so reading a votes file decides nothing
Merged
#346 Have larv-review's scoring-run trigger count unreleased changes as chunk-preflight does
Merged
#345 Spell out the judge's prompt in judge.py and call the worker the worker in it
Merged
#344 Describe the archive layout where it lives, and make help, decisions and the README easier to read
Merged
#343 State when work has landed towards a release once, in work_landed
Merged
#342 Give todo_table.items one way to say a plan has no items table
Merged
#341 Cut release 0.3.0
Merged
#340 Move item 170 to release 0.4.0, where 10 chunks will have shipped under today's steps
Merged
#339 Step a task's worker one rung down for L 1, as for L 2
Merged
#338 Record the ruling on the routing calls: L1 steps one rung down
Merged
#337 Start loop-claude's worker with ScheduleWakeup disallowed
Merged
#336 Make the contributor a standing member of every project's audience
Merged
#327 Measure each chunk by its merge time
Merged
#335 Give settings.json's model one writer per machine, and sort its keys
Merged
#334 Make larv-review's round cap and re-entry easier to read
Merged
#333 Run every tool's git through git_repo.git, with rule-hook's guards as an option
Merged
#332 Retitle the comprehension-floor decision to say the floor doesn't block anything, define the comprehension terms and explain OVERRIDE
Merged
#331 Merge item 182 into item 197, which names the file both questions read through
Merged
#330 Plan 0.3.0 as the day-to-day items scoring over 10 plus every item scoring 30 or more, and move the rest to 0.4.0 and 0.5.0
Merged
#329 File item 201: save tmux setups and restore them after a reboot, once the human has said what they need
Merged
#325 File item 199: start loop-claude's worker with ScheduleWakeup disallowed
Merged
#328 Cut release 0.2.0
Merged
#324 State code health as every project's first achieve goal, weighed four times while the comprehension overall is under 7.0, unless a README voids it
Merged
#323 File item 197: edit an answer at loop-claude's terminal questions with the arrow keys
Merged
#322 Run each review pass on its own rung, lowered only where a replay of its rounds shows a lower rung finds what the project's rung finds
Merged
#321 Accept an escalation quote only from transcript entries after the worker's last step-up, so a later trigger steps the worker again
Merged
#320 Reinstate the 0.97 goal step, which lilleman wants after all, and record why the sum-reciprocal weights stay dropped
Merged
#319 Restore the sum-reciprocal goal weights, which the research found fit best, and record dropping the flat step
Merged
#318 Drop the question on defects' lead, which the 2026-10-16 decision answers
Merged
#317 Weigh each achieve goal 0.97 times the one above it, so a goal far down a project's list counts almost as much as its first
Merged
#316 Move the next chunk's pick into one module, so git_repo only runs git and todo_table only parses the plan
Merged
#315 Plan item 180 as the step mark that keeps an escalation after a step-up
Merged
#312 Mark a renamed release in todo.md, so a release whose items moved unmarked reads as finished
Merged
#314 Have item 189 file, in each project with its own readability goal, an item to drop it
Merged
#313 File items 188 and 189: flatter goal weights, and one code-health goal every project shares
Merged
#310 Have each reviewer write its full report to a file and return the worker only its findings' one-line summaries
Merged
#311 Move items 185 and 177 into release 0.2.0
Merged
#309 Read a release whose last items all moved to other releases as renamed, never finished
Merged
#308 Answer an interrupt on loop-claude's terminal and resume the worker headless with the answer
Merged
#307 Step the worker up as the judge quotes an answer when the run also cites an escalation trigger
Merged
#306 Write negations as people say them: doesn't stop anything, not stops nothing
Merged
#305 Step the worker up as the judge interrupts when the run also cites an escalation trigger
Merged
#304 Plan stepping the worker up after an interrupt that also asked for a higher rung
Merged
#303 Count seat agreement in paired rulings against scoring runs
Merged
#302 Run a preflight before each fresh worker and put its facts in the worker's prompt
Merged
#301 Route the judge and the reviewers on the highest E + I in their chunk, an item the worker takes in included
Merged
#300 Archive each judge reply the loop rejects, with why and the reply
Merged
#299 Grant the bench's runs only the tools a review needs, in place of skipping every permission
Merged
#298 Anchor each level of L, E and I with shipped items, and have score-outcomes list them for a scorer to compare against
Merged
#296 Score risk as L + E + I, and route a worker on its item's L and its reviewers on its E + I
Merged
#297 Plan answering an interrupt at loop-claude's own prompt and resuming the worker headless
Merged
#295 Refuse a review launch below the rung loop-claude gives its worker
Merged
#294 Plan routing a worker on L and its reviewers on E + I, I's reach across every session, and a size recount after the worker's first read
Merged
#293 Re-score R, S and C of 39 items against finished chunks, so R + C routes hard items above Haiku
Merged
#291 Hold each test to earning its keep, and take a test's expected values from its item
Merged
#292 Drop the principles auditor's restated round scope, which review-findings already states
Merged
#289 Run the principles audit on every chunk that changes code, config or tests
Merged
#290 Plan a hook check that keeps reviewers, seats and the security pass at or above the project's rung
Merged
#288 Plan a preflight script for each fresh worker, and an audit each release for steps a script can take over
Merged
#287 Point stability-reviewer at the technical principles in place of its own copies
Merged
#286 Record that 10 finished chunks is the threshold for score-outcomes and the reading-cost trend
Merged
#284 Split the routing decision into one entry per rule
Merged
#285 Plan running the prose pass on the lowest rung a replay qualifies
Merged
#283 Rank todo.md items by value per cost, with a Moves column naming how far each item moves each goal
Merged
#282 Mark each README goal maintain or achieve, give each achieve goal a signal, a counter-signal and a reason, and let a maintain goal veto an item
Merged
#281 Cut release 0.1.0
Merged
#280 Rank each todo.md blocker above the items it blocks, allow no other order in todo.md, and route loop-claude on the first item that does not wait
Merged
#278 Resume a worker one rung up next cycle when it ends its run asking for a higher rung and the judge finds the escalation trigger it cites
Merged
#279 Plan a researched scoring method, and keep its evidence in docs/scoring-and-goals-research.md
Merged
#276 Replay archived stability rounds on the rungs below the project's, and keep the stability reviewer on the project's rung, as no cheaper rung found every finding
Merged
#277 Keep an item the human says to keep, marked owner
Merged
#275 Let a reviewer or seat that rules the work below standard rerun one rung up in its round, and make that ruling an escalation trigger
Merged
#272 Move the outcome measures into release 0.2.0 alone and the readability items and later releases up one minor
Merged
#274 Let a worker hand a too-hard task to a subagent one rung up per escalation trigger, at most two above its item's rung
Merged
#273 Score and reshape the plan only on the project's rung or above
Merged
#271 Answer PR #270's prose round in item 145
Merged
#270 Plan the stability reviewer on the worker's rung once a replay backs it
Merged
#269 Drop item 90 and its citations in the routing decision
Merged
#268 Keep a delegated task's R + C open to any rung in item 141
Merged
#267 Plan scoring and planning on the project's rung or above, and escalation to a stronger rung
Merged
#266 Run the judge on the rung its chunk's item picks, never below the project's rung
Merged
#265 Measure routing figures every 20 cycles and raise each call for a routing change once
Merged
#264 Route each subagent of a chunk by its rung
Merged
#263 Run a chunk's worker on the rung its item picks, below the project's too, with its reviewers kept on the project's rung
Merged
#262 Route each chunk's worker by its item's R + C on one ladder of model and effort rungs
Merged
#261 Move items 89, 90 and 93 into 0.1.0
Merged
#260 Move the readability items into 0.2.0, and item 132 into 0.3.0
Merged
#259 Move items 88, 91, 92 and 94 into 0.1.0
Merged
#258 Plan the places release 0.1.0's scoring run found hardest into release 0.1.0
Merged
#257 Read the plan and the decisions only as far as the chunk needs
Merged
#256 Narrow each review pass to the findings worth a fix, so a pass converges in its second round
Merged
#255 Drop the comprehension panel as a merge gate, and feed the places its seats found hard to the architect, the stability reviewer and item filing
Merged
#254 Score each item's complexity, and calibrate R, S and C against what finished chunks measured
Merged
#253 Mutation-test the lines each chunk changes in the stability pass
Merged
#252 Move item 109 back to 0.1.0
Merged
#251 Move item 109 to 0.2.0
Merged
#250 Put the README's Goals and Audience into every fresh worker prompt
Merged
#249 File item 108: mutation-test the lines each chunk changes, and make each mutant no test catches a stability finding
Merged
#248 Say what lilleman widened the worker's test-first rule to
Merged
#247 Write a failing test first wherever a test can check the change
Merged
#245 Have the worker write the simplest change, test first, reading a principle only where named, and move the rules only reviewers apply to larv-review's standards
Merged
#246 File item 107: loop-claude puts the README's Goals and Audience into every fresh worker prompt, and the item sets the scope
Merged
#244 File item 104: the worker implements the simplest change and loads no technical principle; the reviewers refactor it to the bar
Merged
#242 Make CLAUDE.md a short router, and move each topic's rules to a file under rules/ that loads when a session touches a matching file
Merged
#243 Add deep modules as a technical principle, and put functional core, imperative shell second
Merged
#241 Enforce the checkable rules in a hook, with one override marker that larv-review's overrides pass rules on, and drop them from CLAUDE.md
Merged
#240 Plan 0.1.0 as the CLAUDE.md router, the hook-enforced rules, the narrower reviewers and the measured scoring, and move every other item one release later
Merged
#239 Ship task routing in 0.1.0, and move every other item one release later
Merged
#238 Plan a token and routing review every 20th cycle across all projects
Merged
#237 Plan task routing: a complexity score per item, one model-and-effort ladder, and routes for the judge, reviewers, worker and subagents
Merged
#236 Let a project decision opt out of the architect, the product-owner pass or the comprehension check
Merged
#235 Goals state aim and intent; how a goal is met goes in a method section below them
Merged
#234 Group bin/ by tool, and lift its comprehension scores towards the floor
Merged
#232 Rule each dimension in the paired ruling, and count a ruling as worse when a dimension that falls scores under the floor
Merged
#233 Spell the reading-cost item's title on one read, and keep the stability reviewer's deletions to what it created
Merged
#231 Move larv-review's mechanical rules into a tested bin/larv-review, so the skill says only when to run it and what each answer means
Merged
#230 Bench questions a script answers from the code against the panel's hardest places, and keep only the result
Merged
#229 Measure what would lower review's token use: bench static metrics against the panel's hardest places, and rescore item 73 by its reports' measured cost
Merged
#228 Decide the ratchet by a paired ruling, and check the floor on each release's last item
Merged
#227 Plan item 73: measure first, then choose between write access and a wrapper
Merged
#226 Spend larv-review's rounds only where a fix can change the outcome
Merged
#225 Measure each chunk's reading cost from loop-claude's archive and keep it in its own record
Merged
#224 Keep every scoring run in an append-only comprehension history
Merged
#223 Plan cutting review cost and measuring comprehension cheaply from real work
Merged
#222 Put quality with the fewest interrupts first in the goals
Merged
#220 Rerun larv-review's scoring run before merge when a fix landed after it, and make a score under the floor or the ratchet a 🔴
Merged
#221 Make DRY a guide: break it where the code reads, reasons or maintains better, and let a surface people write accept a second spelling
Merged
#218 Give README an Update section that moves a machine's checkout to the latest tag
Merged
#219 Let a project decision suspend larv-review's scoring run for the plan items it names
Merged
#216 Check Claude Code's settings.json changes for a credential before committing them, and give Install the checkout commands
Merged
#217 Add sane defaults as a technical principle, at the edge beside explicit over implicit
Merged
#215 Audit only the principles that apply, and rerun only the architecture reviewers a fix answers
Merged
#214 Resume an unfinished chunk with the judge's quote of the work remaining
Merged
#213 Make larv-review's pass names and comprehension floor readable on their own
Merged
#212 Bench a new syntax with its users, and never let comprehension fall
Merged
#211 Keep every answer on record, so loop-claude does not ask it twice
Merged
#210 Archive each verdict's evidence, and stamp each outcome with its UTC time
Merged
#209 Record that the judge is to read CLAUDE.md whole
Merged
#208 Fix the places the 2026-10-03 scoring runs named under Shape and Self-sufficiency
Merged
#207 Make larv-review §3 readable on its own, name its rules, and record each pass's rounds in the branch's log
Merged
#206 Make the judge's prompt readable on its own, ask once before the full scan, and sort imports by their first binding
Merged
#205 Hand the security pass back to the run, and accept a judge's elided quote
Merged
#202 Give the judge what the project records and ten times more of the chunk's transcript
Merged
#204 File larv-review's security pass ending the run's turn as a defect
Merged
#203 Add principle, a fifth exempt reason, for a finding of the principles audit
Merged
#201 Install every plugin settings.json enables, and prove the bench's private files stay off the remote
Merged
#200 Keep each project's model per machine, and merge a settings.json change through its own PR
Merged
#198 Wait out a usage limit and retry, without counting it as a failed run
Merged
#199 Keep settings.json in the key order Claude Code writes
Merged
#197 Test that .gitignore keeps private files out, fence the reviewers to reading, and enable the full security scan
Merged
#196 Score and rank todo.md in one table by the formula and goal weights it states, held to its bar
Merged
#195 Make comprehension a technical principle under KISS
Merged
#194 Archive every cycle's transcript and its session folder, however the cycle ends
Merged
#193 Keep the scoring run's seat prompts in the skill and each project's values in its AGENTS.md
Merged
#192 Keep a choice confined to one place in a comment there
Merged
#191 Rewrite the status line in Python, and test it
Merged
#190 Order the 0.1.0 plan by goal rank, then by value within each goal
Merged
#187 Name loop_claude's modules and words by what each holds, and tie --help's exit 1 to the code
Merged
#189 State each decision whole in its title, and delete the ones any reader would reach anyway
Merged
#188 Keep handoff material in the repo
Merged
#186 Give larv-review's finding rules one home
Merged
#185 Plan waiting out a usage limit in loop-claude
Merged
#184 Stop the loop when a finished chunk's session cannot be deleted, and give the session one owner
Merged
#183 Name each entry shape written_by_human checks, and annotate every parameter in loop_claude
Merged
#182 Name the loop's states with one transition function, and inject its side effects
Merged
#181 Merge rebuild into main
Merged
#180 Port main's subagent rule and Opus 5.5 model into rebuild
Merged
#179 Title each decision by what holds now, and make the small fixes filed against bin/, the README, .gitignore and stability-reviewer
Merged
#178 Write larv-review §3's record, round and re-entry rules as numbered rules, and exempt a chunk into rebuild from the scoring run there
Merged
#177 File the places the panel on rebuild at 8585257 named under the floor
Merged
#176 Ground the judge's quote with markdown emphasis and whitespace ignored
Merged
#175 Order the plan by the goals, group it into chunks, and let tiny items bundle
Merged
#174 Plan chunk statistics and a per-release workflow review
Merged
#173 Give goal 4 its sub-goals, and define quality
Merged
#172 Title what Comments and documentation adds to Prose, and cut what Prose already says
Merged
#171 Order CLAUDE.md by what a session needs first, and say when order is worth a fix
Merged
#170 Rank prose priorities clarity first, and rephrase the working loop by them
Merged
#169 Fold one spelling per goal into the technical principles' DRY, read as knowledge
Merged
#168 Move compose-not-entangle into the technical principles, down to functions
Merged
#167 Name the sandbox loop-claude runs in, and git as its seam to production
Merged
#166 Record the human's answers to the rebuild's principles audit
Merged
#164 Audit every principle and goal over the whole project in the architecture pass
Merged
#165 Let a decision name any technical principle under principles/
Merged
#163 Add functional core, imperative shell to the technical principles, and let a decision link one
Merged
#162 Add functional core, imperative shell to the technical principles, and let a decision link one
Merged
#161 File keeping larv-review's prose fixes to what the chunk touched
Merged
#159 Let rebuild's larv-review review a chunk into rebuild
Merged
#160 Ship a rebuild chunk with its worktree's larv-review
Merged
#158 Drop the finished bench item: a blind judge ruled rebuild's larv-review at least as good as main's on all four cases
Merged
#157 Source a list or threshold a consumer relies on
Merged
#156 Plan carrying main's judge-quote fix into the rebuild
Merged
#155 Ground the judge's quote with markdown emphasis and whitespace ignored
Merged
#154 Decline a nit in the summary alone, and define answered
Merged
#153 Run the comprehension floor when the architecture pass is skipped by record
Merged
#152 Give every ask in larv-review a no-human branch
Merged
#151 Let only a record older than the chunk skip a pass
Merged
#150 Give the prose pass its own reviewer agent, prose-reviewer
Merged
#149 Re-enter a nit's fix like any fix, so prose fixes are reviewed too
Merged
#148 Re-enter a nit's fix only where it changes behaviour, and give a capped pass the cap's ask
Merged
#147 State goals as one-line aims read like slogans, with no vagueness bar
Merged
#146 Ask for goals short enough to read as slogans
Merged
#145 File goals as one-line aims first, and bench larv-review against not degrading
Merged
#144 File the places the bench found in rebuild's larv-review
Merged
#143 Name the pass table's column Trigger, define a pass, and plan benching the rebuilt larv-review
Merged
#142 Check loop_claude's type annotations with mypy in the tests
Merged
#141 Annotate the return type of every function in loop_claude
Merged
#140 Define a round, its prompt and when its pass is done under larv-review's Round cap alone
Merged
#139 File the places the panel on rebuild at 2c74bbc named under the floor
Merged
#138 Say in the README why the tests run on the host and need uv
Merged
#137 Say why a leading < marks Claude Code's entry and why a trust failure only warns
Merged
#136 Group a cycle's modules under cycle/ and the loop's under loop/
Merged
#135 Keep formats.py to Claude Code's shapes, with paths and file reads at their callers
Merged
#134 Tell a user entry the human wrote apart with one predicate
Merged
#133 Route a fix through larv-review's Re-entry in one read
Merged
#132 Build premises.sections without filling found through an alias
Merged
#131 Hand run_cycle and judge_through_interrupts one cycle context
Merged
#130 Declare LoopState beside decode_state, as the state a reload carries
Merged
#129 Retry the worker and the judge through one counter, in retry.py
Merged
#128 Give a subagent the least its job needs, stated in its prompt
Merged
#127 File the places the panel on rebuild at 2f9ab2f named under the floor
Merged
#126 Name human_needed.py for the exit-2 stops it holds
Merged
#125 Rename Paths.projects and Session.projects to claude_code_transcripts
Merged
#124 Define a round once, in larv-review's Round cap
Merged
#123 State in outcome.py what the loop does with each outcome
Merged
#122 Resolve decode_state's field types from string annotations too
Merged
#121 Return the updated ~/.claude.json from mark_trusted
Merged
#120 Return a typed verdict from the judge
Merged
#119 File the places the panel on rebuild at 7f9de5c named under the floor
Merged
#118 Show the judge a reply the human types while claude works
Merged
#117 Group loop_claude's modules and move the findings document beside larv-review
Merged
#116 Give failure one spelling
Merged
#114 Name the minute's sleep and the exec at their call sites
Merged
#113 Return the verdict from run_cycle's judge–interrupt loop
Merged
#112 Point stability-reviewer's escalations at larv-review's pass table
Merged
#111 State what a pass's round is and that the round cap counts every one
Merged
#110 Record why the rebuild's merge warning stands in three places
Merged
#108 File the places the panel on rebuild named under the floor
Merged
#106 Make the reviewer agents pure reporters
Merged
#104 Hold every reviewer to one findings document
Merged
#102 Cover the transcript archive in goal 2 and plan its backup
Merged
#101 Archive the raw transcript by content hash with its verdict, and stop on a failed archive
Merged
#100 Move the Claude Code internals from --help beside the modules relying on them
Merged
#99 Move the judge, the interrupt, trust and model resolution into their own modules
Merged
#98 Move the transcript, the session and the worker into their own modules
Merged
#97 Record the human's approval of skipping per-chunk scoring during the rebuild
Merged
#96 Lay out loop-claude as a launcher over the loop_claude package
Merged
#95 Keep main's --reloaded state across the rebuild
Merged
#94 Pin loop-claude's behaviour in black-box tests
Merged
#93 Point the rebuild's remaining items at rebuild's todo.md
Merged
#92 Plan the rebuild of loop-claude and the review process on the rebuild branch
Merged
#91 File the 2026-09-28 panel's places
Merged
#90 Send an external auto-reviewer's behavior-changing fix back through §5
Merged
#88 Define chunk in loop-claude --help
Merged
#86 Leave prose copies to "Comments and documentation"
Merged
#85 Gate the comprehension floor on every dimension and the overall
Merged
#84 Keep the fits-in-a-head thresholds in one list
Merged
#83 Split larv-review's re-entry rules into the passes they re-enter
Merged
#82 Open every todo item with a bold one-sentence tagline
Merged
#81 Move the review protocol into principles/, beside the rules the reviewers already read
Merged
#80 Rename granted, model and project for what they hold and return
Merged
#79 Name a CHANGELOG line among the prose one spelling per goal covers
Merged
#78 Reword the unreleased changelog so no line restates loop-claude --help
Merged
#76 Say at judge what its flags buy and at archive why the stamp is the mtime
Merged
#75 Record why every worktree of the checkout is trusted before an interrupt
Merged
#74 Name each clause of typed for the entry kind it drops
Merged
#73 Say in run_cycle that the interrupt appends to the transcript and archive skips a written stamp
Merged
#71 Return every failure to main and exit only there
Merged
#70 Build and validate the reload state in one unit
Merged
#67 Record Python 3.11 as the floor bin/ runs on
Merged
#68 Run loop-claude on a model saved per project, asking for it on the first run
Merged
#66 Say could not read when the trust file is not an object whose projects maps paths to objects
Merged
#65 Say one spelling in two places is the same case, its copy rejected by naming the place that keeps it
Merged
#64 Exit with a message when README.md is not UTF-8
Merged
#63 Group bin/loop-claude by the Claude Code internals it rests on, main and run_cycle first
Merged
#62 Name the promotion event and the size prompts in systems-architect, leaving trigger to Vocabulary
Merged
#61 Give the scoring run its own heading in comprehension-panel, cited bare from larv-review
Merged
#60 Define looping reviewer and round cap under Vocabulary
Merged
#58 Drop the model setting: Claude Code clears it whenever the organization default is newer
Merged
#57 Record that a reload into an older loop-claude is unsupported while only the developer runs it
Merged
#56 Carry the reload state as one JSON argument and exit naming a reload without a task
Merged
#55 Render the human's typed turns for the judge, so an answered question is not asked again
Merged
#54 Name limitations in the lead of the negation rule
Merged
#53 Say why the session name starts with A, what the judge slice bounds, and what opus[1m] selects
Merged
#52 Bound goal 3 at the remote and drop the forge's visibility from the audience
Merged
#51 Plan the reload-state fix for loop-claude
Merged
#50 Keep in CHANGELOG.md only what a machine taking the release must know
Merged
#49 Delete the prose four places restate the code
Merged
#47 Reject empty evidence in grounded by name
Merged
#46 Pass the checkout root to project, read_premises and checkouts
Merged
#45 Say why the worker failed: the exit code, or exit 0 without a result event
Merged
#44 State the fits-in-a-head thresholds once, cite steps by name, fold the Quality Bar
Merged
#43 Place larv-review's Re-entry section after the passes it cites
Merged
#42 Define a human turn at typed, the entry it tests
Merged
#41 Carry the reload counters on argv, off the LOOP_CLAUDE_ names --help documents
Merged
#40 Say why trust cannot read ~/.claude.json and look up trust without inserting entries
Merged
#39 Record the 2026-09-25 four-seat score and file the places that cap it
Merged
#38 Cut the smaller-codebase forecast from the one-spelling rule
Merged
#37 Run loop-claude from the checkout root wherever inside it is started
Merged
#36 Spell the false-claim rule once, covering what a program prints, and cite it
Merged
#35 Name what the premises lack when loop-claude exits without them
Merged
#34 Say once what a reviewer records, the decision log's home and shape staying in CLAUDE.md
Merged
#33 Give the comprehension floor its own heading and log its calibration
Merged
#32 Scope the negation rule to prose describing the code as it stands
Merged
#31 Bound a definition at the code to one line
Merged
#30 State once what a comment keeps, scoped to prose
Merged
#29 Name the set -u empty-array idiom and the @sh eval in statusline.sh
Merged
#28 Define chunk, plan and release in the judge rules
Merged
#27 Resolve every cited CLAUDE.md heading in a test
Merged
#26 Name the session, the archive folder and its env var loop-claude
Merged
#25 Merge project_path into project: one path relative to ~/code, or to / outside it
Merged
#24 Keep ### sub-headings inside their ## section when reading the premises
Merged
#22 Name the session invariant run_cycle resumes on, and say which outcome ended an interrupt
Merged
#21 State larv-review's re-entry rules once
Merged
#20 Name the reload handoff where main reads it
Merged
#19 Retry the judge under one rule: an unquoted verdict fails like an incomplete one
Merged
#18 Record the 2026-09-24 four-seat score and file the places that cap it
Merged
#17 Name every top-level entry in the README, the principles' audience, and the subject of the comprehension-floor decision
Merged
#16 One word per concept: premises, trigger, gate, reviewer, cycle, interrupt, chunk, memory file
Merged
#15 Name no home path in settings or agents, and show reset times in the host time zone
Merged
#14 Fix the round loop's judge retry, reload and archive modes in loop-claude
Merged
#13 Name the Claude Code internals loop-claude rests on in its --help
Merged
#12 Spell the review pipeline once, in the larv-review skill
Merged
#11 Define "a pointer is not a definition" once, in CLAUDE.md
Merged
#10 Publish by tag, with a CHANGELOG backfilled for 0.1.0
Merged
#9 Let the project's settings pick loop-claude's models
Merged
#8 Cite CLAUDE.md by heading and drop the comment-bar copies
Merged
#7 Forbid --autostash when updating the main checkout
Merged
#6 Run this repo on Fable 5.1
Merged
#5 Point the working loop at the skill's prose-only path and scope its goals exception to the first write
Merged
#4 Move the reviewer protocol into agents/review-protocol.md
Merged
#3 Delete agent-shared
Merged
#2 Plan the comprehension work from the nine-seat panel
Merged
#1 Track loop-claude; continue an unfinished chunk in its session and reload the script each round
2 Pull requests proposed by 1 user
Proposed
#652 File item 466: review once more each prose fix that rewrites a checkable claim
Proposed
#690 File items 513 to 517: prose negations, git in tests, shell mutation, secrets in Docker builds, security pass in worktrees