mini-SDLC stage 5: re-anchor periodic execution slot in worker chains (child of #513)

[kyle-sexton]

This was generated by AI during triage (decomposition intake, rule 8f).

Part of #513.

Scope (from the control-tower plan comment on #513)

Give the re-anchor skill set (do-your-research, point-dont-copy, follow-our-standards, reason-dont-recite, recheck-against-upstream, mind-your-maxims, tighten-your-output) a defined periodic slot in lane/worker execution — no longer human-invoked-only.

Files: Stage 1 SSOT (re-anchor slot placement + cadence); brief templates.

Acceptance criteria

• SSOT names which re-anchor skills fire at which lane boundary and at what cadence.
• Depth-scaled like every other lane (per Stage 2's depth-tier semantics) — never omitted, only shallower at lower depth.

Prerequisite

Depends on Stage 1 (mini-SDLC stage 1: durable pipeline-shape SSOT + work-skill wire-in, child of #513, tracked as #613) — do not start until that lands. No unresolved OPERATOR DECISION is named against this stage specifically in the #513 plan.

STOPGAP NOTE: plain child issue pending the #552 native sub-issues convention — convert when #552 lands.

[kyle-sexton]

This was generated by AI (decide lane — decision-brief pass).

Problem

Stage 5 must give the re-anchor corrector set a defined slot inside an orchestrated worker chain so re-orientation (research/standards/task-contract posture) is no longer human-invoked-only. The open question: at what point(s) in a chain does a re-anchor fire, and how is that cadence specified and enforced — where "periodic" here means boundary-triggered recurrence, not a fixed step- or time-interval (the AC's own wording is "at which lane boundary and at what cadence").

Today

No mechanism fires re-anchor periodically or in-chain. It is user-invoked / posture-set only.

• All 13 correctors are slash-command skills under plugins/re-anchor/skills/*/SKILL.md; the shared method (re-anchor → audit → correct → report) is plugins/re-anchor/context/re-anchor-audit-correct.md. Invocation model is explicit: "Firing a corrector is a re-anchor... before the work, or just to set posture" (plugins/re-anchor/README.md:15-17).
• Repo-wide sweep (git grep -in "re-anchor:" on origin/main): every occurrence is inter-skill cross-references or user-typed eval prompts. No hook, automation, or orchestration composition invokes any re-anchor skill.
• work-items:work (plugins/work-items/skills/work/SKILL.md) — the eventual host — contains zero re-anchor invocation and no mini-SDLC lane composition; the "cadence" references in that plugin are recurring-maintenance scheduling, unrelated.
• session-flow:reanchor (plugins/session-flow/skills/reanchor/SKILL.md) is a different concern (re-anchors factual premises, not standing rules) and is itself manually invoked, not periodic (see its own :89-91).
• Stage 1 SSOT (mini-SDLC stage 1: durable pipeline-shape SSOT + work-skill wire-in (child of #513) #613) is OPEN — the pipeline-shape SSOT that would name lane boundaries does not exist yet.

Options

1. RECOMMENDED — Lane-boundary slot: relevance-routed + depth-scaled; lock the principle now, wire the vocabulary when mini-SDLC stage 1: durable pipeline-shape SSOT + work-skill wire-in (child of #513) #613/mini-SDLC stage 2: frontier-tier complexity-dispatcher — depth-scaled lanes + upward-summarization contract (child of #513) #715 land. Re-anchor fires at defined lane boundaries, the corrector chosen by the lane it precedes (relevance-routed), and depth-tier (from mini-SDLC stage 2: frontier-tier complexity-dispatcher — depth-scaled lanes + upward-summarization contract (child of #513) #715) controls how deep — inline lightweight check at low depth, dispatched fresh-context corrector at high depth — never omitted, only shallower. Straw-man routing (illustrative, pending mini-SDLC stage 1: durable pipeline-shape SSOT + work-skill wire-in (child of #513) #613 boundary names): do-your-research before research/plan; follow-our-standards + reason-dont-recite before review; tighten-your-output before PR; use-your-skills at dispatch. Reversible: cadence lives as data in the Stage 1 SSOT (single edit point), not hardcoded. This is the placement mini-SDLC stage 5: re-anchor periodic execution slot in worker chains (child of #513) #718's own AC already implies ("which re-anchor skills fire at which lane boundary and at what cadence" + "depth-scaled... never omitted, only shallower"), made concrete.

2. Fixed-cadence every-N-steps slot. Fire every N chain steps regardless of lane. Foreclosed by mini-SDLC stage 5: re-anchor periodic execution slot in worker chains (child of #513) #718 AC (which says lane boundary, not step interval): "N steps" is undefined until mini-SDLC stage 2: frontier-tier complexity-dispatcher — depth-scaled lanes + upward-summarization contract (child of #513) #715 names a chain step, it fires context-blind about which corrector is relevant, and risks mid-lane waste. Reversible but off-frame.

3. Event-triggered on a drift signal (additive, not standalone). Fire only on a detected drift signal (verifier finding, scope-fence breach, review-finding class). Cannot stand alone under mini-SDLC stage 5: re-anchor periodic execution slot in worker chains (child of #513) #718 AC — "never omitted" forbids an omittable reactive-only trigger, and no drift-signal producer exists yet. Survives only as a deepen-on-drift layer on top of the option-1 baseline. Reversible.

4. Defer — decide nothing now. mini-SDLC stage 5: re-anchor periodic execution slot in worker chains (child of #513) #718's AC is written entirely in Stage 1 lane-boundary and Stage 2 depth-tier vocabulary; mini-SDLC stage 2: frontier-tier complexity-dispatcher — depth-scaled lanes + upward-summarization contract (child of #513) #715 additionally has an unresolved OPERATOR DECISION on the depth-tier taxonomy. Choosing wiring now would hardcode against an undefined frame. But this decides nothing — the placement principle is derivable from the parent AC today; only the concrete boundary/depth-tier names are blocked.

Recommendation

Option 1. The real fork is 1-vs-4: option 1 locks the placement principle now (lane-boundary + relevance-routed + depth-scaled, with option-3 drift-deepening as an additive layer) and defers only the concrete boundary/depth-tier vocabulary to when #613 and #715 land; option 4 decides nothing. Prefer 1 because the design is fully derivable from #718's own acceptance criteria even though the wiring is blocked — so the brief captures the principle and parks the SSOT edit behind its prerequisites. Depth-scaling is not incidental: a corrector that self-audits the same context that produced the work carries the exact same-context bias #304 flags, so "inline at low depth → dispatched fresh-context corrector at high depth" is what makes the re-anchor slot honest and keeps it consistent with #513's implementer ≠ reviewer invariant.

Cross-links

• work-items(work): per-issue execution should compose installed skills as a role-separated mini-SDLC (implementer ≠ reviewer ≠ verifier, ephemeral fresh-context agents) #513 parent umbrella (mini-SDLC; re-anchor named as a periodic lane).
• mini-SDLC stage 1: durable pipeline-shape SSOT + work-skill wire-in (child of #513) #613 Stage 1 SSOT (hard prerequisite — defines the lane boundaries this slot attaches to; OPEN).
• mini-SDLC stage 2: frontier-tier complexity-dispatcher — depth-scaled lanes + upward-summarization contract (child of #513) #715 Stage 2 depth-tier framework (blocks the depth-scaling spec; unresolved OPERATOR DECISION on the taxonomy).
• mini-SDLC stage 3: separated reviewer lane (implementer ≠ reviewer) before PR (child of #513) #716 / mini-SDLC stage 4: immutable test-spec manifest guardrail (child of #513) #717 sibling stages (separated reviewer lane / immutable test-spec manifest).
• program: fresh-eyes checkpoint audit — tag skill actions for same-context bias and retrofit delegation #304 fresh-eyes / same-context bias — the rationale for dispatching the corrector fresh-context at higher depth.
• re-anchor: add a check-the-lanes corrector (org-wide WIP survey before driving cross-repo work) #350 proposed check-the-lanes re-anchor corrector — a new corrector that would also need a boundary slot (survey-before-drive); worth co-designing the slot taxonomy so it has a home.