feat: statusline prime-drift indicator (deferred)

[kyle-sexton]

Proposal (Fable, opus-5-prompting workstream; deferred to tracker at plan approval 2026-07-26)

Add a statusline cue that fires when the session's primed model-adaptation chapter has drifted
from the running model/session state — e.g. the session armed
plugins/playbooks/skills/fable-5/context/model-adaptation/opus-5.md (the primed artifact) but
the effective model is no longer Opus 5 (mid-session model switch, --model override, or a
spawned context inheriting the primed text without the model it was written for).

Trigger

Manual priming proves forgettable in practice (recorded in the approved opus-5-prompting PLAN's
deferred questions, 2026-07-26): the fable-5 playbook's meta-rule 3 makes the
model-adaptation chapter mandatory at arm time, but nothing re-checks the chapter/model pairing
when session state changes afterward. A statusline indicator makes the drift visible without
adding instruction payload.

Notes

• Primed artifact to key on: context/model-adaptation/<model-version>.md under the playbooks
  fable-5 skill; version routing is owned by that skill's meta-rule 3.
• Cheap and optional; statusline implementation approach is open.

[kyle-sexton]

This was generated by AI (triage lane, gate-audit pass — cycle 2 retry, instance melo-lap-001-triage-20260727T125559Z).

Verdict: real, actionable, but not agent-ready — the design is open by the issue's own admission.

The drift condition (primed context/model-adaptation/<model-version>.md vs. effective model) is well-motivated, but the body states plainly that the "statusline implementation approach is open," and the issue is self-labelled deferred and optional. That is an open design choice, so it stays human-gated rather than being handed to a worker. Blast radius is bounded to a statusline indicator, hence work-class: scoped rather than structural.

Cross-reference: #1697 — both were deferred out of the same opus-5-prompting plan approval (2026-07-26) and share that workstream.

Labels: priority: low, work-class: scoped, needs-human, status: needs-decision (removed priority: needs-triage).

[kyle-sexton]

This was generated by AI in the decide-and-implement lane (operator authorization: human gates become decision work for frontier consensus). Decision: closed, won't build — superseded at the source, and the statusline surface cannot observe what the indicator needs. Two independent grounds. First, the re-check gap that motivated this (nothing re-checked the chapter/model pairing after arm time) has since been closed inside the fable-5 playbook itself: meta-rule 3 now states that the identity it resolves 'is not guaranteed to hold', treats any in-context evidence of fallback or model switch as a re-resolution trigger against the model now answering, and the full arm mode covers long autonomous runs where trigger-time reads are unreliable — that is the drift handling this indicator was meant to approximate, landed where the routing is owned. Second, the indicator is not buildable as the 'cheap and optional' item this was deferred as: a statusline process would need to know which adaptation chapter this session primed, and no artifact persists that fact anywhere outside the session's own context — building one means a new persisted priming-state seam with staleness and cleanup contracts, i.e. structural work, and meta-rule 3's own sourced finding (Fable 5 system card §1.5/§8.3) records that a classifier fallback none of the surface signals relay 'is undetectable from inside the session', so the session could not reliably write the very state the statusline would read. If a priming-state seam ever exists for other reasons, this can be re-proposed against it; sibling deferral #1697 is untouched by this closure. Capability-tier consequences of the same model generation landed separately via #1293.