Repository navigation
perf(context-budget): measure the fleet listing fraction on Opus 5.5 and Fable 5.1 and set skillListingBudgetFraction #12044
Description
Refs: #6458
Blocked by: #11954
Outcome
#6458's decision comment of 2026-10-27, criterion 2, says that after the Q185 pilot merges, a comment states the chosen fraction with its shed count per model (Opus 5.5 and Fable 5.1), and .claude/settings.json skillListingBudgetFraction equals it. #11954 runs the pilot and puts setting the fleet budget out of its own scope ("Setting the fleet listing budget (#7588, #6458)"). This slice is that remaining step. .claude/settings.json:61 still reads "skillListingBudgetFraction": 0.07, the interim value Q187 keeps until the pilot ships.
Done when
- perf(session-flow): run the description trim pilot and measure skill visibility #11954 is closed with its
Probe result:comment posted. - A comment on feat(context-budget): keep plugin skill descriptions visible when the skill listing overflows #6458 gives, for Opus 5.5 and Fable 5.1, the number of skill descriptions shed at the current fraction and at each candidate fraction tried, measured in real sessions (for example through
/harness-ops:audit-skill-visibilityor/skill-quality:check listing-budget), and names the chosen fraction with itsBasis:. - A PR sets
.claude/settings.jsonskillListingBudgetFractionto the chosen fraction, and links that comment.
Out of scope
- The session-flow description trim pilot itself (perf(session-flow): run the description trim pilot and measure skill visibility #11954).
- Re-grading invocation modes (closed under docs(invocation-mode): grade sync and tidy-work KEEP per the #6458 decision #9176).
Activity
github-actions commentedon Nov 1, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Small
Complexity: Low
Uncertainty: Knowable
Flags: none
Blocked by: #11954
Understood: after the #11954 pilot closes with its
Probe result:comment, measure skill descriptions shed per model (Opus 5.5, Fable 5.1) at the current and candidate fractions, and post the results and the chosen fraction on #6458. Then open a PR that setsskillListingBudgetFractionin.claude/settings.jsonto that fraction. The issue is blocked by #11954, which is still open. The measurement in real sessions needs a host run, so it is routed there. No likely duplicates found among open issues.Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 1, 2026
Not admitted to the issue lane: a criterion names .claude/settings.json, which the issue lane cannot push (patch-protected-path); a builder session takes it; it carries host-run. Admission takes the issue once each of those is met, when intake triages it again (
.github/actions/resolve-config/admission-rules.json).melodic-automation-lanes commentedon Nov 1, 2026
The latest triage verdict holds nothing: admission moved the issue to
status: confirmed; its comment says why.kyle-sexton commentedon Nov 2, 2026
Claimed. Backlog HITL host-run worker, 2026-11-01. Measuring skill-listing shed counts in fresh sessions on Opus 5.5 and Fable 5.1 at the current and candidate
skillListingBudgetFractionvalues, then preparing the settings change on branchhost-run/12044-listing-fraction. #11954 is closed with its probe result.kyle-sexton commentedon Nov 2, 2026
Done. Each criterion, checked:
Probe result:comment..claude/settings.json:61on main already holds"skillListingBudgetFraction": 0.07, the chosen value.Gap, stated: Fable 5.1 got no live turn, because
-pbills usage credits on Fable without asking. Its counts come from listing captures that sent no model request. Opus live sessions matched the captures at 0.07, 0.03 and 0.02.This was generated by AI.