Upgrade

1. Grounded in current state. No recommendation without a Prior Status tag (🆕/🔶/💬/🚫) backed by file:line evidence gathered this run. Already-implemented items go to Skipped Content with evidence — that's the proof the prior-state check ran. Rejected ideas (MEMORY/KNOWLEDGE/REJECTED/) only resurface with a named reason the context changed. This standard binds internal synthesis inference exactly as it binds external findings: before any absence-claim earns a 🆕/CRITICAL tag, the thing claimed missing must be positively probed this run (grep/read for it), never inferred. A reported absence you did not check is a fabricated finding — the 2026-08-06 scan shipped a false CRITICAL across seven skills by grepping context: fork while never grepping background:. Internal inference is exempt from nothing.
2. A technique, not a pointer. Quote or code-block the actual content; name the exact LifeOS file or component it improves; include What It Is and How It Helps LifeOS (≤2 concrete sentences each). The test: if "show me the technique" has no answer, it doesn't ship. Content with nothing extractable goes to Skipped with a reason — skip boldly rather than dilute.
3. Won't break what exists. Check backward compatibility against current skills, hooks, and workflows before recommending adoption.
4. Formatted per the contract. References/OutputFormat.md is the single source of truth for section order, Prior Status legend, table columns, and hard rules.

Source labels in output: GitHub: claude-code vX.Y.Z · YouTube: Creator @ MM:SS · Docs: Section · Blog: Title.

• Hard deadline, fail-open — never block on a straggler. Set a synthesis deadline (~4 min) at dispatch; report with whatever is back when it hits. A missing source is listed as ⏳ timed out in Sources Processed; it degrades coverage, never delays the report. (2026-07-18: one hung GitHub-trending agent stalled a run ~1 hour.)
• Right-size the fan-out (~8 agents). Small-file reads collapse into one agent; network sources get short budgets and are first to drop. Over-fan-out is the failure the reflection corpus flags most.
• Budget the claude-code-guide freshness spawn per-spawn — it dies on a broad ask. Asking one spawn to cover the whole Claude Code surface (hook events, settings.json, slash commands, SKILL/subagent frontmatter, MCP, SDK, API) returns nothing: it fans out a batch of doc lookups on its first turn and the combined results blow its context window, surfacing as Prompt is too long ~10s after spawn. The prompt length is not the cause — a trivial prompt to the same agent type in a larger parent conversation completes fine; how much the ask makes it FETCH is the variable. Cap at ~3 areas per spawn, tell it to look things up rather than pull whole pages, and split the surface across parallel spawns so losing one costs a slice instead of the whole freshness check. Diagnostic tell: sibling agents in the same batch all succeed while this one dies — that points at the spawn's context budget, not the batch. Fallback that needs no agent at all: read Claude Code CHANGELOG from sources.json directly and diff the version range. (public PR #1659, @elhoim.)
• Idle teammate ≠ delivered result. Spawned agents sometimes go idle without sending output — a one-line SendMessage nudge recovers them. Nudge once; don't re-spawn.
• GitHub search 422s on bare OR between qualifiers with no free-text term. Give every query a free-text term; don't retry a 422 inside the budget.
• Absence from settings.json ≠ a dead hook. It's GENERATED by MergeSettings; hooks also fire via dispatchers (PreToolGuard) and Pulse HTTP. Flag "verify before concluding dark," never assert dead from absence.
• Check sources in parallel, not sequentially — the run is network-bound.