Development

Upgrade

Generate prioritized LifeOS upgrade recommendations via 4 parallel threads (prior-work audit, user context, external source collection, internal reflections), output as a ranked Discoveries table plus tiered CRITICAL/HIGH/MEDIUM/LOW recommendations with Prior Status tags and before/after code

06
Workflows
01
Tool
00
References
13
Triggers

The Problem

Knowing what to work on next is harder than doing the work. Generic AI can summarize what's new in an ecosystem, but it has no idea where you actually are, what you're trying to achieve, or whether a shiny new technique closes any real gap in your setup. The result is a list of interesting things with no signal about what matters — or worse, recommendations for things you already built. Most upgrade advice is source-driven: here's what came out this week. But what came out this week and what you should build next are almost never the same question.

How This Skill Approaches It

The skill runs four parallel threads that converge on ranked, personalized recommendations. Thread 0 audits prior work — Algorithm runs, ISAs, KNOWLEDGE, hooks, skills — so nothing already shipped gets re-recommended. Thread 1 reads your TELOS goals, IDEAL_STATE dimensions, and CURRENT_STATE/INFRASTRUCTURE.md to compute gap tuples (metric, current, target, direction) per goal — this is the load-bearing input. Thread 2 pulls from 30+ external sources: Anthropic blog, changelogs, GitHub trending, YouTube channels. Thread 3 mines algorithm-reflections.jsonl for recurring friction patterns. Recommendations are ranked by impact × ease × confidence, where impact comes from how many active TELOS goals a change touches and how far it moves their gap. Every recommendation includes a gap_dimension, gap_distance, telos_links, before/after implementation, and a Prior Status from Thread 0. The Recommend workflow leads with gaps; the Upgrade workflow leads with sources. TwitterBookmarks scans X bookmarks through the same four-thread analysis.

In Action

What you say to your DA, and what the Upgrade skill actually does.

  • You say "what should i work on next"
    Runs Recommend: reads TELOS goals and CURRENT_STATE/INFRASTRUCTURE.md in parallel, computes per-goal gap tuples, cross-references Thread 0 (prior work audit) to skip already-shipped items, and returns tiered recommendations (CRITICAL/HIGH/MEDIUM/LOW) each tagged with gap_dimension, gap_distance, telos_links, and concrete before/after implementation steps.
  • You say "check for upgrades, anything new from anthropic or in the ecosystem"
    Runs Upgrade: hits 30+ sources in parallel via Tools/Anthropic.ts (blogs, changelogs, GitHub releases, YouTube channels), extracts specific techniques with code examples and timestamps, maps each to a LifeOS component or INFRASTRUCTURE.md row, and filters through Thread 0 to skip already-implemented items before ranking.
  • You say "mine my reflections for recurring friction patterns"
    Runs MineReflections: reads algorithm-reflections.jsonl, clusters recurring friction by pattern type, surfaces the top candidates as Thread 3 Reflection-tagged recommendations with specific file paths and proposed fixes.

Inside the Skill

The thinking, frameworks, and architecture that distinguish this skill from a generic version of the same task.

Extraction Rules

Extract, don't summarize. Techniques, not recommendations. Verify prior state before recommending.

  1. Every output item is a TECHNIQUE — a specific pattern, code snippet, configuration, or approach.
  2. Quote or code-block the actual content — show exactly what was said/written.
  3. Map to LifeOS components — every technique connects to a specific file, skill, workflow, or system component.
  4. Verify Prior State (Thread 0 gate) — Before emitting ANY recommendation, confirm against Thread 0 inventory: is it already in Algorithm / PATTERNS.yaml / hooks / SKILL files / KNOWLEDGE / prior ISAs? Assign a Prior Status emoji and cite evidence. Items that are ✅ DONE go to Skipped, not Recommendations.
  5. Two mandatory description fields, ≤2 sentences each, concrete and specific:
  • What It Is: the technique itself — what it does, how it works, what capability it provides
  • How It Helps LifeOS: the specific benefit — which component improves, what gap it fills
  1. Provide implementation — show before/after code or specific steps.
  2. Skip, don't dilute — if content has no extractable technique, put it in Skipped Content with reason.

Source Type Labels:

Label Meaning
GitHub: claude-code vX.Y.Z Specific version release notes
YouTube: Creator @ MM:SS Video with timestamp
Docs: Section Name Documentation section
Blog: Post Title Blog post

Configuration

Skill Files:

  • sources.json — Anthropic sources config (30+ sources)
  • youtube-channels.json — Base YouTube channels (empty by default)
  • State/last-check.json — Anthropic state
  • State/youtube-videos.json — YouTube state
  • State/github-trending.json — GitHub trending state (seen repos)
  • State/twitter-bookmarks-seen.json — Previously processed bookmark URLs

User Customizations (~/.claude/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Upgrade/):

  • EXTEND.yaml — Extension manifest
  • youtube-channels.json — User's personal YouTube channels
  • user-sources.json — Additional source definitions (e.g., github_trending block)

Tool Reference

Tool Purpose
Tools/Anthropic.ts Check 30+ Anthropic sources (blogs, GitHub repos, changelogs, docs) for updates

Key Principles

  1. Extract, Don't Summarize — pull specific techniques, never just link to sources.
  2. Quote the Source — actual code, doc quotes, or transcript excerpts.
  3. LifeOS-Contextualized — every technique maps to a specific LifeOS file, skill, or component.
  4. Explain "Why You" — "this helps because your [X] currently [Y]".
  5. TELOS-Connected — reference user's goals and challenges when explaining relevance.
  6. Skip Boldly — if content has no extractable technique, skip it entirely.
  7. Implementation-Ready — provide actual code changes, not vague recommendations.
  8. Claude Code Freshness via claude-code-guide — when discoveries involve Claude Code internals (hooks, settings, slash commands, MCP, agent types, keybindings, Agent SDK, Claude API), spawn Agent(subagent_type="claude-code-guide") to verify LifeOS's current references match the latest API surface.

Anti-Patterns (What NOT to Output)

These output patterns are FAILURES:

❌ Bad Output Why It's Wrong ✅ Correct Output
"Check out R Amjad's video on Claude Code" Points to content instead of extracting it "@ 5:42, R Amjad shows this technique: [quote]"
"v2.1.16 has task management improvements" Vague summary, no technique "v2.1.16 adds addBlockedBy parameter: [code example]"
"Consider looking into MCP updates" Recommendation without extraction "MCP now supports [specific feature]: [docs quote]"
"This could be useful for your workflows" Vague relevance "This improves your Interceptor skill because [specific gap it fills]"
"Several videos covered AI agents" Count without content "[N] videos skipped — no extractable techniques"
"This helps because it improves things" Vague benefit "How It Helps LifeOS: SecurityValidator currently only blocks commands. additionalContext enables reasoning context before tool execution, making decisions more nuanced."
"A new hook feature" No description of what it IS "What It Is: PreToolUse hooks can return additionalContext that gets injected into the model's context before execution, enabling reasoning-based decisions rather than binary blocks."
"Top 3 Actions" or flat recommendation list No priority tiers Recommendations section with 🔴/🟠/🟡/🟢 tiers, each with LifeOS Relevance column
Recommendations at the bottom Actionable items buried after technique dump 🔥 Recommendations section appears SECOND, technique details third
Recommending something already implemented Wastes user trust Move to Skipped with file:line evidence
Re-surfacing rejected ideas without new context Drift from prior decisions Only re-recommend if reason has changed; say what changed
Missing Prior Status column Bypasses Thread 0 gate Every recommendation row cites evidence from Thread 0's inventory

The test: if you can say "show me the technique" and there's nothing to show, you've failed.

Workflows

  • Upgrade.md — primary workflow; full four-thread analysis with prioritized recommendations
  • MineReflections.md — Thread 3 standalone (deep reflection mining)
  • AlgorithmUpgrade.md — Algorithm-version-bump-focused upgrade flow
  • ResearchUpgrade.md — deep dive on a specific upgrade opportunity
  • FindSources.md — discover and evaluate new sources to monitor

Gotchas

  • Check ALL sources in parallel — Anthropic blog, changelog, YouTube channels, GitHub releases. Don't check sequentially.
  • Upgrades must not break existing skills or workflows. Verify backward compatibility before applying.
  • Full upgrade check can take 5-7 minutes. Use run_in_background: true for the outer agent.

Workflows & Routing · 6

Each workflow is one job the skill runs. The trigger phrases route your request to the right one — this is the skill's routing table.

  1. 01
    Upgrade Workflows/Upgrade.md

    check for upgrades, check sources, any updates, check Anthropic, check YouTube, upgrade, pai upgrade

  2. 02
    MineReflections Workflows/MineReflections.md

    mine reflections, check reflections, what have we learned, internal improvements, reflection insights

  3. 03
    AlgorithmUpgrade Workflows/AlgorithmUpgrade.md

    algorithm upgrade, upgrade algorithm, improve the algorithm, algorithm improvements, fix the algorithm

  4. 04
    ResearchUpgrade Workflows/ResearchUpgrade.md

    research this upgrade, deep dive on [feature], further research

  5. 05
    FindSources Workflows/FindSources.md

    find upgrade sources, find new sources, discover channels

  6. 06
    TwitterBookmarks Workflows/TwitterBookmarks.md

    check bookmarks, scan bookmarks, twitter bookmarks, X bookmarks, bookmarks for upgrades, what have I bookmarked

Tools · 1

Deterministic executables the workflows call — the code that does the real work, not prompt scaffolding.

  • Anthropic.ts

How to Invoke

Say any of these to your DA and LifeOS activates the Upgrade skill automatically:

  • "upgrade"
  • "system upgrade"
  • "check Anthropic"
  • "new Claude features"
  • "algorithm upgrade"
  • "LifeOS upgrade"
  • "check bookmarks"
  • "scan bookmarks"
  • "twitter bookmarks"
  • "X bookmarks"
  • "bookmarks for upgrades"
  • "what have I bookmarked"
  • "mine reflections"

Or invoke explicitly:

Skill("Upgrade")

Want LifeOS to do this for you?

Install LifeOS on your machine — your DA gets the Upgrade skill plus 46 others, all hooked into one Life OS.