Repository navigation
playbooks: retire the Opus 5 and Opus 4.8 chapters once they stop being fallback targets #4349
Description
Problem
The goal is to keep playbooks only for current models. Opus 5.5 replaced Opus 5, but opus-5.md and opus-4-8.md are still needed. Per the Claude Code model-config page (section "Automatic model fallback", read 2026-09-23), flagged requests fall back as follows:
- On Fable 5.1, Fable 5 and Opus 5.5, biology-flagged requests re-run on Opus 5 and cybersecurity-flagged requests re-run on Opus 4.8.
So a session that is flagged continues on Opus 5 or Opus 4.8, and meta-rule 3 routes it to the matching chapter. Each chapter now says at its top that it is kept only for this reason.
Done when
When the fallback targets change, remove the chapter for any model that is no longer a target. Also update meta-rule 3, the fable-5 evals, and the CHANGELOG to match.
Trigger
The model-config "Automatic model fallback" section no longer names Opus 5 or Opus 4.8.
Activity
feat: apply the Opus 5.5 usage guide across repo instructions and plu…
kyle-sexton commentedon Sep 27, 2026
This was generated by AI during triage.
Triage: briefed, human-gated (trigger not yet met)
Verified 2026-09-27. The Claude Code model-config page, section "Automatic model fallback", still names both targets: on Fable 5.1, Fable 5 and Opus 5.5, biology-flagged requests re-run on Opus 5 and cybersecurity-flagged requests on Opus 4.8; on Opus 5, cybersecurity flags re-run on Opus 4.8. The trigger has not fired, so both chapters stay. Each chapter's opening line still states it is kept only as a fallback target, and the fable-5 meta-rule 3 still routes to both with a dated fallback citation.
Why human-gated, not agent-ready: the work waits on an external doc change that no in-repo signal detects. An
agent-readylabel would have workers claim it, find the trigger unmet, and no-op on every pass. A human (or a docs-watch pass) re-reads the section and releases the item when it changes.Brief (for when the trigger fires)
Precondition: re-read the model-config "Automatic model fallback" section. Act only on a model it no longer names as a fallback target. If both are still named, stop and change nothing.
Behavior after the change:
Acceptance: the chapter set matches the current fallback targets; meta-rule 3 and the evals agree with it; the reference sweep finds no stragglers; the plugin's validation and evals pass.
Out of scope: changes to the chapters that remain, and the Fable 5.1 doctrine regeneration (#4346).
Priority set to
priority: medium(assessed default; nothing is blocked on it).docs(playbooks): keep Opus fallback chapters (#4349)
docs(playbooks): keep Opus fallback chapters (#4349)
docs(playbooks): keep #4349 open as the retirement tracker
docs(playbooks): keep Opus fallback chapters (#4349) (#4918)
kyle-sexton commentedon Sep 29, 2026
Owner decision (2026-09-29): Keep this issue open, drop needs-human, and recheck the Claude Code model-config "Automatic model fallback" section on a schedule. The trigger has not fired: as of 2026-09-29 the section still names Opus 5 and Opus 4.8 as fallback targets, so both chapters stay.
Next: needs-human is removed and agent-ready added. An agent rechecks the section; if both models are still named it changes nothing, and if either is dropped it follows the brief in the triage comment (remove that chapter, update meta-rule 3, evals, CHANGELOG, sweep references).
kyle-sexton commentedon Oct 4, 2026
This was generated by AI during work-loop.
Ratify autonomous dispatch of this C3 bug-fix item?
Classification:
work-class: scoped(C3), bug-fix-shaped. Intended dispatch: the work-loop worker lane, through/work-items:work, to a draft PR the lane never merges. The lane has not completed its first drain and no item-level ratification exists, so the item is queued here instead of dispatched. Ratify it through/work-items:attend-queue, and it returns to the frontier and dispatches on a later cycle.Lane: work-loop, instance melo-worker-1, cycle 1.
kyle-sexton commentedon Oct 4, 2026
kyle-sexton commentedon Oct 4, 2026
This was generated by AI during attend-queue.
Operator direction (2026-10-04): "go with the recommendations." This records that decision.
Ratification declined: the trigger has not fired. The code.claude.com model-config page (fetched 2026-10-04), under "Automatic model fallback", still names Opus 5 (biology) and Opus 4.8 (cybersecurity) as fallback targets. A worker dispatched now would find nothing to do. The item stays
needs-humanas a watch item. Recheck when that section stops naming either model, then act only on the model it no longer names.kyle-sexton commentedon Oct 13, 2026
Decision recorded. The owner accepted the adjudicated recommendation on 2026-10-12.
Keep #4349 open with needs-human; add one monthly recurring-schedule row (trigger claude-model-release) that re-reads model-config#automatic-model-fallback and comments the result on #4349. Drop the seed's 'close it' option.
Why: The trigger has not fired: the live model-config page (lines 521, 523, read 2026-10-12) still names Opus 5 and Opus 4.8, and the API deprecations page keeps both Active until at least 2027-07-24 and 2027-05-28. Both owner decisions (2026-09-29, 2026-10-04) keep the issue open, so the seed's close recommendation loses under precedence rule 2; the 09-29 decision asked for a scheduled recheck that nothing performs today (origin/main recurring-schedule.json has 4 rows, none covers the section; check-stamps.sh:487 treats any recheck trigger as clearing the stamp, so the SKILL.md:19 pointer has no consumer). The 10-04 record set the label back to needs-human but is silent on the schedule, so the newer decision governs the label and the older one's schedule request stands unrevoked.
Precedence applied: Rule 2 (owner's newest explicit decision): the seed's close recommendation (rule 5) loses to the 2026-09-29 and 2026-10-04 owner decisions, both of which keep #4349 open. Between the two owner decisions, the newer 2026-10-04 record governs the label (needs-human) and, being silent on the schedule, leaves the 2026-09-29 schedule request in force. Researcher disagreements (B's claude-model-release trigger, A's Sonnet 5 scope) are additive and both adopted.
Sources:
Next: Open a draft PR adding one row to .github/recurring-schedule.json modeled on agents-md-cutover-check: id model-fallback-targets-watch; cadence monthly; triggers ["claude-model-release"]; close_previous true; notes: re-read https://code.claude.com/docs/en/model-config#automatic-model-fallback and comment the result on #4349; report-only (C1); if the section still names Opus 5, Opus 4.8 and Sonnet 5, change nothing and leave #4349 needs-human; if it drops Opus 5 or Opus 4.8, the operator releases #4349 (drop needs-human) and the 2026-09-27 triage brief runs; if it drops Sonnet 5, file a sibling issue for sonnet-5.md (nothing tracks it today); retire the row when #4349 and that sibling close. Change no label or comment on #4349 in this pass. Cadence monthly is judgment from the precedent row; the trigger is the backstop for a change that ships with a model release.
Researched by two independent agents (local evidence and upstream docs) and adjudicated by a third.
kyle-sexton commentedon Oct 13, 2026
Owner batch-ratified this admission in an attended session on 2026-10-12: work-class: scoped. Rule: class C1-C3, no trust surface, the independent check confirmed the recorded decision, and no owner step remains.