docs(multi-agent,session-flow): fix repo-mode audit pilot findings and schedule a full sweep #6916

Activity

1. github-actions

   added

   needs-triageNot yet classified. Floor until a type and one priority tier are set.Not yet classified. Floor until a type and one priority tier are set.

   on Oct 11, 2026
2. github-actions commented on Oct 11, 2026

   

   github-actionsBot

   on Oct 11, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for the report. Automated triage handed this issue to a maintainer.

   _{Automated triage. A maintainer reviews it.}
3. github-actions

   added

   needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.Human-in-the-loop required; autonomous sessions must not resolve items carrying this.

   on Oct 11, 2026
4. kyle-sexton

   added the

   Task

   issue type on Oct 12, 2026
5. kyle-sexton

   removed

   needs-triageNot yet classified. Floor until a type and one priority tier are set.Not yet classified. Floor until a type and one priority tier are set.

   on Oct 12, 2026
6. kyle-sexton commented on Oct 13, 2026

   

   kyle-sexton

   on Oct 13, 2026

   MemberAuthor

   More actions

   Decision recorded. The owner approved applying independently verified recommendations on 2026-10-12.

   • Keep best as a role model value and in the frontier default? → Drop best from role values and the frontier default
     Basis: https://code.claude.com/docs/en/sub-agents.md lines 351-357 (fetched 2026-10-12); https://code.claude.com/docs/en/model-config.md line 33; Agent tool schema in this session (model enum sonnet/opus/haiku/fable); plugins/multi-agent/reference/config.md:31,34 on origin/main
   • How to run the full repo-mode sweep? → Waves with a pilot wave first, in a fresh plan-usage window
     Basis: docs(multi-agent,session-flow): fix repo-mode audit pilot findings and schedule a full sweep #6916 body 'Proposed next step'; judgment (spend only, reversible)
   • Keep best as a role model value and in the frontier default? → Drop best from role values and the frontier default, and remove it from resolve-roles.sh ALIASES in the same change so --session-model best is rejected and the caller omits it (treated as frontier). (added by verification)
     Basis: origin/main plugins/multi-agent/scripts/resolve-roles.sh:61,135,298-302,327,335

   A 2026-10-10 repo-mode audit pilot confirmed 10 model and pointer hygiene findings in multi-agent and session-flow and proposes a full ~700-agent sweep. Main has moved since (#6998, #7009): the README haiku pin is gone and several pointers now exist, so each finding needs re-checking before fixing.

   Verification notes:

   • Implementation caveat for Q1: resolve-roles.sh:61 keeps best in ALIASES, which also validates --session-model (:135), and SESSION_FRONTIER is computed from the frontier list (:298-302). Dropping best from the frontier default while ALIASES still accepts --session-model best would make a best session count as non-frontier and let inherit fan-out workers bypass the guard. Remove best from ALIASES too (an unknown session already counts as frontier) or keep it in frontier.
   • Assumption that Workflow opts.model rejects best is unverifiable: the workflows page says agents follow the subagent model order (workflows.md line 453) without listing accepted values.

   Verified by an independent fresh-context check (confirmed, risk low): gh issue view 6916 (no owner comment); origin/main plugins/multi-agent/reference/config.md:30,34; origin/main plugins/multi-agent/reference/defaults.yaml:4,14; origin/main plugins/multi-agent/scripts/resolve-roles.sh; origin/main plugins/multi-agent/skills/route/SKILL.md:24-30,74-84; git grep claude-haiku origin/main plugins/session-flow/README.md: no match

   Admission to an agent lane stays with the work-class rules and owner ratification.
7. kyle-sexton

   added sub-issues on Oct 18, 2026

   • fix(multi-agent): drop best from the role model values and the frontier default #7990
   • docs(session-flow): recheck and point the orchestrate model and cost claims #7991
8. kyle-sexton commented on Oct 18, 2026

   

   kyle-sexton

   on Oct 18, 2026

   MemberAuthor

   More actions

   This was generated by AI during backlog shaping.

   This issue is split into 3 pieces, each sized for one lane PR. Re-checked on origin/main as the decision comment requires: the README haiku pin is gone and the orchestrate choose-a-model, available-tools and size-guideline pointers now exist (session-flow orchestrate SKILL.md:189,209; context/sources.md:206,217,295), but best is still a role value and frontier default in config.md:31,34, defaults.yaml:4,23, resolve-roles.sh:30,66, setup SKILL.md:86 and drift-audit.js:200; split into the multi-agent best removal plus remaining pointers, the session-flow pointer recheck, and the paid full sweep.

   • filed: fix(multi-agent): drop best from the role model values and the frontier default (fix(multi-agent): drop best from the role model values and the frontier default #7990)
   • filed: docs(session-flow): recheck and point the orchestrate model and cost claims (docs(session-flow): recheck and point the orchestrate model and cost claims #7991)
   • not filed: chore(multi-agent): run the full audit-defaults repo sweep in waves: paid-run: about 700 agents over 110 files, waves with a pilot in a fresh plan-usage window (owner decision 2026-10-12)

   Filed pieces are sub-issues of this one, and the paced queue admits them to the lane. Unfiled pieces wait on the blocker shown. The owner approved this split in an attended backlog-shaping session on 2026-10-18.
9. kyle-sexton

   removed

   needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.Human-in-the-loop required; autonomous sessions must not resolve items carrying this.

   on Oct 24, 2026
10. kyle-sexton commented on Oct 24, 2026

    

    kyle-sexton

    on Oct 24, 2026

    MemberAuthor

    More actions

    Released by the owner's HITL session (2026-10-24, C12): it waited only on a removed gate. Intake triage re-runs on it to set its fields, then admission takes it.
11. kyle-sexton commented on Oct 25, 2026

    

    kyle-sexton

    on Oct 25, 2026

    MemberAuthor

    More actions

    Needs a run of claude (eval, probe or /usage), which the issue lane cannot do (pr-pipeline.yaml:289-314). The owner's HITL session runs it on the operator machine; no owner action needed.