Recheck Against Upstream Deep

Recheck against upstream — deep

The fan-out tier of the sibling /discipline:recheck-against-upstream. Same upstream-conformance discipline; heavier execution. Where the base skill rechecks the one surface in play inline, this one fans fresh-context subagents out over a whole subsystem, framework, or repo — doc-by-doc — the execution tier the base skill's context cannot cover from within itself.

The shared method — re-anchor, audit, correct forward, report, and the tone that firing this is not an accusation — lives in [${CLAUDE_PLUGIN_ROOT}/context/re-anchor-audit-correct.md](../../context/re-anchor-audit-correct.md). The discipline this re-anchors, its portable baseline, and the three divergence categories (gap / deliberate / undocumented) live in the sibling [recheck-against-upstream](../recheck-against-upstream/SKILL.md). Read both; this file adds only the fan-out delta. There is no separate copy of the discipline or the taxonomy here — update the sibling and this tier follows.

When this tier, not the inline recheck

Reserve the fan-out for when the upstream surface area is broad enough to justify the subagent cost: a whole framework's configuration, an entire integration's API usage, a subsystem that leans on many upstream contracts at once. For a single surface or a short session, the inline recheck in the sibling is the right tool; this tier is overkill.

The fan-out

Run this in place of the sibling's inline audit and correct-forward steps:

1. Enumerate the surfaces. List every upstream-dependent surface in the

subsystem/repo under review — each config block, API call site, infra definition, and documented contract the work rests on. Do not spot-check one.

1. Fan out, throttled, doc-by-doc. Dispatch fresh-context subagents

(blind to the assumptions that produced each surface) to fetch the CURRENT official upstream docs for that surface and classify its divergence per the sibling's three categories. Throttle the dispatch in bounded waves rather than launching one agent per surface at once — a sustained wide fan-out trips server-side burst overload (529s) and loses agents mid-run. Cap concurrency to a modest wave (roughly a dozen or fewer); process the surfaces wave by wave.

1. Retry the failed subset only. If an agent errors or times out, retry

that surface once; on a second failure mark it unverifiable — an honest skip, never a false pass. Never blind-re-run the whole fan-out to recover a few stragglers.

1. Checkpoint the partial ledger mid-run, if a durable slice exists. So

a crash mid-fan-out does not lose completed waves, checkpoint the partial ledger to the session's durable topic-memory slice when one is available; where the session has no such durable store, proceed without it rather than asserting a persistence surface. This is the only persistence this tier performs — nothing is mandatory beyond it.

1. Merge and report an inline divergence ledger. One list keyed by

surface: its category (gap / deliberate / undocumented) and the current upstream source that resolved it. Correct gaps toward upstream this turn; re-check that deliberate divergences still hold and flag any the docs have overtaken; surface undocumented ones for the human with both options.

Routing findings onward

• **Category 1 (gap) and category 3 (undocumented) → OFFER work-items

routing.** When a work-item / issue-tracker capability is installed, offer to route the actionable gaps and the undocumented divergences awaiting a human decision into tracked work items. Degrade to a prose offer (a listed set of would-be items) when no such capability is present — never assume a tracker.

• Category 2 (deliberate) is report-only. A recorded rationale that

still holds is not an action item; raise it only when the current docs have obsoleted it, at which point it becomes a gap and routes with the others.

What this skill does NOT do

• Not a lighter inline pass. For a single-surface recheck without the

subagent cost, use the sibling /discipline:recheck-against-upstream.

• Does not force persistence. The mid-run checkpoint is best-effort crash

safety on a durable slice when one exists; it never mandates a store or invents one.

• Does not fabricate conformance or a finding. An honest per-surface

"matches current docs" or "unverifiable" is the right output when true.

Gotchas

• Throttle is not optional at scale. The failure mode is a

surface-heavy subsystem firing one agent per surface simultaneously; the wave cap and failed-subset retry are what keep the fan-out reliable.

• Blind subagents, or it is not fresh context. An agent handed the

assumption that produced a surface re-confirms the same drift. Verify against the current upstream doc, not the reasoning for the state.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: melodic-software
• Source: melodic-software/claude-code-plugins
• License: MIT

Install and usage instructions live in the source repository linked above.