cortex

Run the conversation analyzer agent for a full adaptive immunity scan — detects corrections in today's journal, classifies patterns, writes lessons, and generates evolution proposals

Launch a structured 3-pass code review — bug/logic, security, and project conventions analysis with confidence scoring

Interactively scaffold a new Cortex skill with proper structure, frontmatter, and context-flow wiring

Run the memory-synthesis agent to curate collaboration patterns, workflows, and project file organization

Launch an exhaustive deep-dive research session on any topic — competitors, markets, technology, codebase, ideation

Use this agent when analyzing a session for correction events, pattern detection, and system evolution proposals. Reads session journals and lessons, detects mistakes, classifies patterns, and writes proposals. Examples: <example> Context: Session-end detected 2+ correction events or 1+ reasoning-miss in today's journal user: "Run the conversation analyzer" assistant: "I'll use the conversation-analyzer agent to perform a full adaptive immunity scan of today's session." <commentary> Session-end found multiple reasoning misses, triggering the adaptive immunity analysis to extract lessons and propose system improvements. </commentary> </example> <example> Context: User wants to manually review session quality user: "/analyze-session" assistant: "I'll launch the conversation-analyzer agent to analyze today's journal for corrections, patterns, and evolution opportunities." <commentary> User explicitly invoked the analyze-session command, which triggers this agent. </commentary> </example> <example> Context: Session-start surfaced degrading health trend user: "Health metrics are degrading, let's figure out what's going wrong" assistant: "I'll use the conversation-analyzer agent to scan recent sessions and identify recurring patterns causing the degradation." <commentary> Degrading health trend warrants a full adaptive immunity scan to find root causes and propose fixes. </commentary> </example>

Use this agent for exhaustive research, investigation, or comparison on any substantial topic — competitor analysis, market research, live product testing, technical deep dives, codebase investigation, concept learning, ideation, or self-auditing. Triggers on phrases like "research X", "deep dive on X", "compare X vs Y", "investigate X", "look into X", "what are the options for X", "analyze X". Produces comprehensive written reports with strategic recommendations. Examples: <example> Context: User wants to understand the competitive landscape user: "Do a deep dive on all the SaaS platforms competing with us" assistant: "I'll use the deep-dive agent to conduct an exhaustive competitive analysis." <commentary> Market/competitor research — the agent will search the web, visit competitor sites via browser, analyze pricing/features/positioning, and produce a strategic report. </commentary> </example> <example> Context: User wants to audit their own product user: "Deep dive into our live product and tell me what's broken or could be better" assistant: "I'll use the deep-dive agent to conduct a thorough self-audit of the live product." <commentary> Self-audit — the agent will visit the live site via browser, test flows, screenshot issues, and produce a findings report with prioritized recommendations. </commentary> </example> <example> Context: User wants to learn about a new technology user: "I need to understand everything about vector databases" assistant: "I'll use the deep-dive agent to produce a comprehensive research report on vector databases." <commentary> Deep learning — the agent will research the topic exhaustively from primary sources, compare options, and produce an actionable knowledge report. </commentary> </example> <example> Context: User wants to explore a new product direction user: "Research whether adding options flow data would be valuable for our users" assistant: "I'll use the deep-dive agent to investigate the opportunity." <commentary> Ideation/opportunity research — the agent will research the space, existing products, data sources, user demand signals, and produce a strategic assessment. </commentary> </example> <example> Context: User wants to compare specific tools or technologies user: "Compare Koyfin vs TradingView vs Stockanalysis — features, pricing, UX, target audience" assistant: "I'll use the deep-dive agent to do a thorough competitive comparison." <commentary> Head-to-head comparison — the agent will visit each product via browser, analyze features/pricing/UX, build comparison matrices, and produce a strategic assessment. </commentary> </example> <example> Context: User wants to investigate something in the codebase end-to-end user: "Investigate our data pipeline — trace the full flow, find bottlenecks, tell me what's inefficient" assistant: "I'll use the deep-dive agent to trace the pipeline architecture and produce an efficiency report." <commentary> Codebase investigation — the agent will read CLAUDE.md and documentation.md for context, then trace code paths, analyze patterns, and produce a findings report. </commentary> </example>

Use this agent when the user wants to curate, organize, or review synthesis memory files (collaboration patterns and reusable workflows). Reads all files in ~/.cortex/synthesis/, performs dedup/merge/reorganize/flag operations, syncs index files, and flags archive candidates across the project's file system. Triggers on phrases like "curate memory", "curate synthesis", "organize files", "clean up memory", "review collaboration patterns", "review workflows". Examples: <example> Context: User wants to clean up synthesis files user: "Curate my memory files" assistant: "I'll use the memory-synthesis agent to scan collaboration patterns and workflows for duplicates, staleness, and reorganization opportunities." <commentary> User explicitly requested memory curation, triggering the synthesis curator agent. </commentary> </example> <example> Context: Session-start suggested curation (10+ sessions since last run) user: "Sure, go ahead and curate" assistant: "I'll use the memory-synthesis agent to run a full curation pass." <commentary> Session-start flagged that curation was due, user approved the suggestion. </commentary> </example> <example> Context: User wants to review what collaboration patterns have been captured user: "What collaboration patterns do we have?" assistant: "I'll use the memory-synthesis agent to read and summarize the current collaboration patterns." <commentary> User wants visibility into the synthesis tier content. Agent reads and presents it. </commentary> </example>

Use this agent for structured code review with three focused analysis passes: (1) Bug/logic, (2) Security, (3) Project conventions. Reads CLAUDE.md, lessons.md, and documentation.md for project-specific rules. Produces findings with confidence scores, filtered to high-confidence results (>=80), deduped by file+line. Triggers on phrases like "review my code", "code review", "review this branch", "review the diff", "check this PR", "review before merge". Examples: <example> Context: User wants a review of recent changes user: "Review the code I just wrote" assistant: "I'll use the code-reviewer agent to perform a 3-pass analysis of your changes." <commentary> Standard code review — the agent gets the git diff, then runs bug/logic, security, and conventions passes, producing a consolidated findings report. </commentary> </example> <example> Context: User wants a review of a specific file user: "Code review src/lib/pipeline/scoring.ts" assistant: "I'll use the code-reviewer agent to analyze scoring.ts through all three review lenses." <commentary> Targeted file review — the agent reads the specified file and runs all three passes against it. </commentary> </example> <example> Context: User wants to review before merging user: "Review everything on this branch before we merge" assistant: "I'll use the code-reviewer agent to review all changes on this branch." <commentary> Branch review — the agent runs git diff against the base branch to capture all changes, then reviews through all three passes. </commentary> </example>

This skill should be used when a problem class has appeared 2+ times across sessions — escalates recurring patterns up through the memory tier hierarchy.

This skill should be used when the user asks to "display data", "add a chart", "show a number", "create a data component", "fix wrong data", "debug null values", "add a pipeline source", "handle data freshness", "fix stale cache", or before any code that displays, transforms, or stores data in the current project. Enforces 'every number must be accurate and traceable.'

This skill should be used before writing any Supabase or PostgREST query — 8 gotchas that cause silent failures.

This skill should be used on-demand before deploying to production — 13-item pre-deploy verification checklist. Trigger when the user says "deploy", "push to production", "ship it", "go live", "deploy readiness check", "ready to deploy", "release to prod", or "merge to master".

This skill should be used when starting any feature, significant change, or architectural decision — sets quality bar, sequences design before implementation.

8 hooks across 7 events