claude-cortex

End-of-day closing routine. Quick mode by default (v4.6+) — runs cortex auto-commit, reflective prompts, brief pre-stage, memory-index refresh, and close in 30 seconds to 3 minutes. Heavy steps (inbox triage, transcript review, mining of non-transcript sources) only run in `/end-day --full`, or when auto-offered because the day actually had transcripts (≥2) or inbox volume (≥5). Auto-fires on "/end-day", "wrapping up", "calling it a day", "done for today", "end of day", "closing out today". Quick mode is the default for every trigger; the user passes `--full` or accepts the auto-offer prompt to get the full chain. Reflection (Step 4) appends a `## Reflection` section to today's brief markdown — what tomorrow's `/brief` reads — AND to a longitudinal `memory/reflections.md` store (v4.13+). v4.13+ additions (End-Day Routine Improvement Spec): a per-source consent + cost gate (Step 0.7, Today's Brief required & first); Step 2c mines the brief artifact (task/outreach actions → memory write-backs + suppression lear

Natural-language entrypoint for cross-artifact drift detection. Fires when the user has just made a meaningful change to a node (status flip, reframe, engagement wound down) and says "check the linked nodes," "sync linked artifacts," "make sure the related stuff is consistent," or similar. Resolves the source node from recent edit or context, then routes to /sync-linked-entities.

Retroactive wikilink pass over `<config-root>/memory/`. Scans every node file for plain-text mentions of known entities (people / clients / topics / workstreams / companies / domains), proposes converting them to `[[wikilinks]]`, and graduates heavily-mentioned persons into person pages. One-time migration per cortex v4.10+. Idempotent; gated by `memory/.migration-wikilink-relink-done` marker. Run once after upgrading; future memory writes use wikilinks natively per the canonical rule in cortex CLAUDE.md. Auto-fires when the user says: "relink my memory", "fix my wikilinks", "my graph is disconnected", "why isn't my obsidian graph showing edges", "back-fill the wikilinks", "graduate the people in my memory". Use `--rerun` to force re-scan after marker exists. Cost: 1-3 min for ~30 nodes. Token cost ~$0.10-0.50 per run.

Create a new workstream node — cortex's primitive for ongoing initiative pipelines that span multiple projects, people, and topics. Walks a short interview (name, current state, pinned context, linked entities) and writes `<config-root>/memory/workstream/<slug>.md`. Auto-fires when the user says: "start a workstream", "begin tracking X as a workstream", "I'm starting a new initiative — X", "I'm working on a new initiative", "set up a workstream for X". Differentiation: - Project nodes (`client/<slug>`) capture a specific engagement. - Person pages capture a relationship. - Topic nodes capture a subject area. - Domain nodes capture a persistent area of work. - Workstreams capture an ongoing *initiative pipeline* with current state, pinned context, open loops, and linked entities — spans projects/people/topics. New in cortex v4.9.0. See `references/workstream-schema.md` for the full schema.

Regenerate `<config-root>/memory/index.md` — a deterministic, zero-LLM catalog of every memory node grouped by type with decay-state flags. Auto-fires in THREE scenarios: 1. CHAIN STEP: invoked as Step 6 of `/end-day` and as the final step of `/cleanup`. Runs after user-gated changes are committed so the index reflects current state. 2. DEFERRED: when `/remember` writes a new entry, it appends a line to `<config-root>/memory/staged/queues/reindex`. The next `/end-day` or `/cleanup` notices the queue and runs the indexer. 3. EXPLICIT: user runs `/reindex`, or says "regenerate the memory index", "rebuild the catalog", "refresh memory/index.md". This skill has zero LLM cost in steady state. It is a file-walk + string-extract + write. No reasoning, no synthesis — see references/memory-index.md for the deterministic spec.

Nightly autonomous ingest pipeline. Pulls yesterday's calendar / Gmail / Slack / transcripts / Drive activity into `<config-root>/archive/YYYY-MM-DD/` (immutable substrate), runs the mining agents against the archive, stages all proposed memory commits to `staged/commit-drafts/YYYY-MM-DD.md`, and refreshes `memory/hot.md`. Designed for unattended scheduled execution — no user gates. Auto-fires when the user says: "ingest yesterday", "run nightly ingest", "what happened yesterday I haven't captured", "mine yesterday's calendar", "process yesterday's transcripts". Most users register it on a schedule (11pm or 5am) via core-ops `/register-schedules` and never invoke it by hand. Pair with `/morning` for the interactive review-and-merge of the staged draft.

The JARVIS morning routine. Reads the most-recent overnight `/listen` draft at `<config-root>/memory/staged/commit-drafts/`, walks the user through each proposed memory commit (accept / reject / edit / defer), applies accepted proposals to active memory nodes with v4.4 drift-detection, refreshes `memory/hot.md` and `memory/index.md`, and optionally chains into `/brief`. Auto-fires when the user says: "good morning", "start my day", "what happened overnight", "morning routine", "merge overnight ingest", "review last night's proposals". Most natural trigger is just opening a session and saying "good morning" — the router suggests `/morning` and the chain begins. Read-write to active memory but only for user-accepted proposals. Never silently merges, even for high-confidence proposals.

Active retention loop — surfaces 3-5 aging knowledge entries past their freshness threshold (but not yet cold) and walks the user through "still true? still useful?" per entry. Confirm / update / demote / archive / skip per entry. The explicit consolidation pass that converts dormancy into either re-confirmation or removal. Auto-fires on "/rehearse", "is this still true", "rehearse my memory", "what should I confirm or forget", "memory rehearsal", "consolidate my memory", or invoked weekly by /end-week. Default batch size 5; cadence weekly.

Active maintenance loop for cortex memory. Scans `<config-root>/memory/` for thin / stale / contradictory / orphaned / under-cited / past-due / sparse content, researches the highest-priority gaps via the open web (≥2 sources per claim), and writes findings to `staged/research-drafts/` for user review. Auto-fires when the user says: "what's missing from my memory", "research gaps in my second brain", "fill in what we don't know", "audit my knowledge for holes", "what should we learn more about", "are there contradictions in my memory", "what's stale that I should refresh", "what nodes are weak". Also fires when invoked explicitly via /research-gaps and as optional Step 6 of /end-week (cortex v4.5+). Honors the private-individual privacy rule: only verifiable professional facts surface for person research. Never modifies active memory. All findings stage to a draft for user merge.

Always-on passive learning engine. This skill does NOT wait for user triggers — it runs continuously in the background of every conversation. It silently detects user preferences, corrections, domain knowledge, workflow patterns, and relationship context. Never interrupts. Never announces itself. Just watches, learns, and adapts in real-time. Observations are flushed to memory when the conversation ends. This skill should be considered ALWAYS ACTIVE. It does not need a trigger phrase. Any conversation is an opportunity to learn about the user. Note (v4.8+): `/observe` is no longer a slash command — its prior duties are this always-on skill. If you typed `/observe`, what you wanted is already running.

Foundational onboarding walker for Nucleus. Chains the essential setup commands needed for the marketplace to be productive — identity, voice, note sources, Obsidian vault, per-plugin setups, diagnostics, optional schedule registration. Idempotent. Auto-fires when the user says: "start nucleus", "let's get started", "set me up", "onboard me", "first time setup", "let's begin", "I just installed Nucleus, now what", "configure everything", "walk me through setup". The natural phrase a new user types when they don't yet know what commands exist. Goes from zero to working Nucleus in ~15-30 minutes depending on how many plugins are installed and which steps the user skips. Safe to re-run any time — completed steps are silently skipped.

Centralized writer for `<config-root>/memory/log.md` — the unified append-only chronicle of audit-worthy Nucleus operations (v4.7.1+, centralized in v4.7.2). Auto-fires when ANOTHER cortex command finishes its main work and needs to append a log entry. Not invoked by the user directly; invoked by /listen, /morning, /end-day, /end-week, /reindex, /research-gaps, /merge-research-draft, /cleanup, /rehearse at their respective "Log to chronicle" steps. Inputs: op_name (the command name, e.g., "listen"), summary (the one-line description for the entry). Optionally body (1-2 line context block). Output: appends one entry to `<config-root>/memory/log.md`. Creates the file with the standard header if it doesn't exist. Never modifies prior entries. Format is defined here in one place. If the format changes, this skill updates; callers don't need to change their instruction strings.

Scaffold an Obsidian vault config over `<config-root>/`. After this runs, the user can open Obsidian → File → Open vault → select <config-root>/ and see their cortex memory as a graph, with daily-brief snapshots as daily notes and a VAULT.md home page powered by Dataview. Auto-fires when the user says: "set up Obsidian", "make this work in Obsidian", "enable graph view", "make this work on mobile", "I want to use Obsidian", "set up a vault", "configure Obsidian", "open this in Obsidian". Idempotent and non-destructive — existing `.obsidian/` files and VAULT.md are preserved on re-run. Force-reset available via `/setup-obsidian --reset`.

End-of-week closing routine. Runs transcript-reviewer for uncaptured commitments, /cleanup for memory hygiene, /review for synthesized digest, /rehearse for aging-knowledge triage (v4.4+), prompts for weekly reflection, optionally pre-stages Monday's outreach via weekly-outreach. Auto-fires on "/end-week", "Friday wrap-up", "weekly close", "end of week", "Friday afternoon ritual", "close out the week", or any phrase signaling Friday wind-down. Takes ~15-20 minutes including the rehearsal step.

Configure cortex's note-source list — Granola, Gemini, Fireflies, Otter, Notion, generic Drive folders or Gmail labels, custom. Auto-fires on "/setup-sources", "set up cortex sources", "configure my notetakers", "set up the mining layer", "add a notetaker to cortex", "configure note sources for end-day", or when /end-day reports "no note sources configured."

Capture or update user identity in one canonical file (~/Documents/Claude/identity.md) so other plugins don't re-ask name/company/role 9 times. Auto-fires on "/setup-identity", "set up my identity", "set up shared identity", "configure my profile across plugins", "update my identity", or when another plugin's setup reports the shared identity file is missing.

Capture or update writing voice in one canonical file (~/Documents/Claude/voice.md) so every drafting plugin reads the same rules. Auto-fires on "/setup-voice", "set up my voice", "configure my voice across plugins", "update my writing voice", "my voice is changing", or when a drafting plugin reports the shared voice file is missing.

Surface project memory, user profile, and knowledge into the current conversation. Auto-fires in THREE scenarios: 1. EXPLICIT: User says /recall, "what do we know about [topic/project]", "remind me where we left off", "how does [thing] work", "any gotchas with [thing]", "what's the status of [project]", "catch me up on [project]". Also fires on "what have we learned about [topic]" or when a user references a project/topic and seems to expect prior knowledge. 2. CONVERSATION START: At the beginning of every new conversation, BEFORE the user asks for anything, proactively load context. This means: - Read the user profile node to remember who they are and how they like to work - Check for active projects with stale threads or overdue P0s - If the user's first message references a specific project or topic, load that context immediately - If the user's first message is a greeting or general start, load the dashboard overview and surface anything that needs attention - Appl

Commit a conversation to working memory — extracts both project state and knowledge (insights, lessons, mental models, gotchas, recipes, corrections), plus any observations accumulated by the passive learning engine. Auto-fires in THREE scenarios: 1. EXPLICIT: User says /remember, "save this conversation", "commit this to memory", "log this session", or asks Claude to "remember" something specific about a project, decision, or piece of knowledge. 2. CONVERSATION END: User signals they're wrapping up — "thanks", "that's all", "goodbye", "talk later", "I'm done", "signing off", "good night", "bye", "let's stop here", "wrap it up", or any farewell/closing language. In this case, run a SILENT auto-commit — don't ask permission, just commit what was learned. If the conversation was trivial (no decisions, no knowledge, no meaningful observations), skip the commit silently. 3. SIGNIFICANT MOMENT: A major decision was just made, a critical piece of knowledge was shared, or a correction was g

Search across all working memory for knowledge, decisions, people, or topics. Auto-fires when a user asks "any gotchas with [X]", "what did we learn about [topic]", "when did we decide [X]", "what's the trick for [X]", "have we seen this before", "what's blocked right now", "what are my P0s", or "did we ever figure out [X]". NOTE: "how does [X] work" and "what do we know about [X]" should route to /recall (contextual mode) if X is a known project/topic, and to /search only if it's a cross-project query or the project is unknown. /search is for finding things across nodes; /recall is for loading context for a known node.

Show a chronological timeline of activity across all projects or a single project. Auto-fires when a user asks "what have I been working on", "show me last week", "what happened this month", "give me a timeline of [project]", "what did we do on [project] recently", or "show my activity". Also triggers on "what's the arc of [project]" or "show me the history of [project]". NOTE: "weekly review" and "prep my status update" should route to /review, not /timeline. /timeline is raw chronology; /review is synthesized analysis.

Run maintenance on working memory. Auto-fires when a user says "clean up memory", "memory is getting cluttered", "consolidate my logs", "what's stale", "memory health check", or "run maintenance". Also triggers on "I'm running out of memory space" or "too many old entries".

Archive or remove a project node from working memory. Auto-fires when a user says "archive [project]", "we're done with [project]", "close out [project]", "remove [project] from memory", "merge [project] into [other project]", or "this project is finished". Also triggers on "clean up old projects" or "I don't need [project] anymore".

Capture standalone knowledge without a full session commit. Auto-fires when user says "remember that X works like Y", "good to know that...", "TIL", "I just realized...", "note to self about [technical thing]", "the trick is...", "gotcha:", "watch out for...", "the way X works is...", or "I was wrong about X". Also fires when the user discovers something mid-conversation and it sounds like knowledge worth preserving (a gotcha, mental model, technique, or corrected assumption) but a full /remember would be overkill.

Quick one-liner capture. Auto-fires when a user says "note that [fact]", "jot down that...", "quick note:", "FYI for the record:", "just happened:", or any brief factual statement the user clearly wants persisted but doesn't warrant a full /remember or typed /learn entry. Also fires if the user says "log this" followed by a short statement.

Generate a synthesized weekly review. Auto-fires when a user says "weekly review", "what did I accomplish this week", "summarize my week", "weekly roundup", "what happened this week", "prep my status update", or "end of week review". Also fires on "what have I learned recently" or "give me a digest of recent activity".