Repository navigation
docs(rfc): work ledger caps a goal's open items; its item total is advisory - #19352
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Before merge, a maintainer should confirm iamwhatever's acceptance. The RFC lands as [DESIGN-REVIEWED] e4f330c Description read[DESCRIPTION-READ] efb77a7b66eabe3c0ef1323b0a4e95466453e504a6dc21e55143665145a4d4a0 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
…visory Records the decision #19338 implements: a goal board refuses `create` on its open-item count (`concurrency_cap`, default 6, 1..32, `goal_concurrency_cap_reached`), and `item_cap` becomes an advisory line the read reports as `cap_exceeded`. The user's ruling: cap parallel work, not lifetime totals. Lands as `accepted` with an index row: the maintainer accepted the design in the conductor chat on 2026-10-12 (~06:12 UTC). #19338 then rebases onto the base that carries it.
9b9f2ca to
e4f330c
Compare
|
Self-review (crew-pr-reviewer, read-only): RECOMMEND APPROVE This docs-only PR records the decision that a goal's board is bounded by open items ( |
buluoray
left a comment
There was a problem hiding this comment.
LGTM. I checked this at head e4f330c.
What it records: a goal is capped on items open at once (default 6) rather than on items created over its lifetime, and the lifetime total becomes advisory. Today a goal whose items finish quickly is refused at 20 even with nothing running, and the conductor then asks about "spend".
Checked:
- The lifetime cap being moved is your own earlier work (#16987, #17076, #18111), and nothing in
docs/decisions/covers work-ledger caps, so no other maintainer's decision is overridden. rfc-conductor-work-ledger.mdis not edited; the new RFC amends itsgoal_item_cap_reachedentry in its own document.- Cited symbols are real on main in
work_ledger.py:_create_item,MAX_ITEMS_PER_CONDUCTOR = 32,DEFAULT_GOAL_ITEM_CAP = 20,MAX_GOAL_ITEM_CAP,CODE_GOAL_ITEM_CAP_REACHED,CODE_ITEM_CAP_EXCEEDED,ConductorRecord(noconcurrency_capyet), and the "spend is bounded" text in_CONDUCTOR_SYSTEM_PROMPT. - The planned check order (stored total, store open bound, then concurrency cap) means a
concurrency_capat its ceiling still yieldsitem_cap_exceededfirst, consistent with main's codes. - The README row matches.
scripts/docs-lint.shpasses. CI at this head: 88 success, 13 skipped, 0 failed.
Problem / Motivation
Goal: Record the decision that the work ledger bounds a goal by how many items it has open at once, and reports its lifetime total instead of refusing on it.
Today
_create_iteminsrc/kiro_crew/work_ledger.pyrefuses a goal board'screatewithgoal_item_cap_reachedonce the goal has createditem_capitems (20 by default), closed ones and every re-plan counted. A goal whose items land and close quickly is refused with nothing running, and the conductor asks the user about "spend" when nothing is being spent. The user's ruling on the dispatch for #19338: cap parallel work, not lifetime totals; keep the total visible.The implementation is open as #19338. Its First Principles lane blocks: a changed default for a first-class agent needs a decision recorded on the base branch. This document is that record.
Why it matters
The bound that protects the host and the budget is how many workers run at once: each open item is a live session. A lifetime total stalls a healthy goal on its own history. Without a per-goal bound on parallel work, the only one is the store's 32-open ceiling.
Not a goal
MAX_ITEMS_PER_CONDUCTOR(32) or the stored bound (256); the latter isrfc-work-fold-forget-closed.md's.What changed (motivation → approach → change)
docs/request-for-change/rfc-work-ledger-concurrency-cap.md, statusaccepted: the problem as the code stands at0a47021090, what bounds spend, the decision (a newconcurrency_cap, default 6, range 1..32, refusingcreateasgoal_concurrency_cap_reached;item_capkept as an advisory line withcap_exceededon the read;goal_item_cap_reachedremoved), what is not decided, the alternatives considered, and how acceptance is recorded. It amendsrfc-conductor-work-ledger.mdin its own document, so that one is not edited here.docs/request-for-change/README.md: one index row,accepted, verified against0a47021090, pointing at #19338.The status is
acceptedon landing because the maintainer (iamwhatever) already accepted this design in the conductor chat on 2026-10-12 at about 06:12 UTC: a hard cap on open items, the total cap advisory only. The RFC's status line, its acceptance section and the README row cite that. Merging this PR puts the record on the base; #19338 then rebases onto it.Backwards compatibility
Compatible: documentation only.
Removes nothing: two documentation files gain text; no code, key or route changes.
Tests
N/A — documentation only.
scripts/docs_lint.pyandscripts/docs-lint.shpass locally on this tree.Manual verification
N/A — the Docs Lint gate reads both files; there is nothing to run.
Related Issues
Records the decision for #19338. Follows #18111 (the refusal this decision moves).
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)