How to Build Agent Skills for Claude (Step-by-Step)

When you come back to check on your old job and it's even worse than you expected

The smartest thing you can do with Hermes right now:

Create a self-evolving loop that trains Hermes on your data - so every session is smarter than the last.

Session 1 → learns → logs it → Session 2 is smarter → logs that → repeat

Here's exactly how to set it up:

1. Create a Memory.md file on your desktop

Structure it like this:

## Preferences
## Corrections
## Patterns
## Lessons learned

2. Attach it to Hermes

Paste this prompt once:

"At the start of every session, read Memory.md and apply everything in it.

After every task, do three things:

• Log what worked and why
• Log what failed and why
• Analyse and write rules you'd apply next time

Never duplicate entries. Rewrite existing rules when you learn something better."

3. Once a week, run this prompt:

"Review everything in Memory.md. Identify patterns across all your logged lessons. Distill them into sharper, more general rules. Delete anything that's been superseded. The goal: fewer, better rules every week."

4. Archive before every cleanup

Copy Memory.md into a dated backup before your agent rewrites anything.

(In case of any mistakes, this is how you secure previous data)

That's the self-evolution loop that everyone should be running with Hermes right now.

Super simple, but it makes Hermes 10x more powerful.

Make sure to bookmark this so you actually maximise your agentic workflows.

It's hard to fathom that in ~12 months, we'll likely have access to a model 2x more powerful than Fable 5.

And that might be the conservative estimate, & model costs will likely be way cheaper at that point, too.

Here's why:

The best single measure of AI capability is Epoch's Capabilities Index (ECI)

(a composite score across 40+ benchmarks - math, coding,
science, reasoning).

Think of it like an IQ test for frontier models.

When Fable was released, it hit a new high of 161 on the ECI.

Now here's the part that matters:

The rate of frontier improvement nearly doubled around April 2024, from about 8 points per year to 15.

That faster pace has now held across every single quarterly measurement cycle since then, and is the new trend.

So, in theory, the trend that's held since 2024 just needs to continue in post-Fable 5 model releases, and the floor in 12 months is a 2-3x improvement in model capability.

But the wildest stat isn't even the capability curve.

It's the cost curve:

The cost to run an AI model at a fixed level of performance is halving every 2 months.

Put it together:

In 12 months, you won't just have a model 2x more capable than Fable 5.

You'll have it cheaper, faster, and possibly running on hardware you already own.

We're so early.

$CRWV – AI GPU Cloud Specialist Pulling Back

Weekly chart: Big prior run then sharp -18%+ candle testing support. $43B mkt cap player in the exploding AI infrastructure space.

CoreWeave runs specialized data centers packed with NVIDIA GPUs for AI training & inference. High-demand niche as everyone from startups to enterprises needs massive compute power.
Fits the broader AI cloud theme (see $NBIS): Capacity constraints + surging workloads = opportunity for specialists. But capex intense and competitive.

Watch support levels. Sector momentum strong if AI spend continues.

$16B Robotics Investments in 2026 — ATH per a16z

This isn’t hype. It’s the moment AI gets a body.
Software intelligence exploded first (ChatGPT moment). Now capital is flooding the physical layer — humanoids, autonomous systems, and embodied AI.
Why this matters:
• Record funding = validation that robots are moving from demos to real deployment.
• AI brains + robotics bodies = the next multi-trillion TAM (industry forecasts already talking hundreds of billions to trillions long-term).
• Compute, power, and data infrastructure just became even more critical.
This follows the exact AI infrastructure boom we’ve been tracking.
The companies powering the training (massive GPU clouds) and the future deployment (space, mobility, factories) are positioned right at the center.
Robotics isn’t replacing AI — it’s the next chapter.
What do you think — is 2026 the year physical AI goes mainstream, or still early?

$JOBY eVTOL Air Taxi Thesis

Consolidating after moves, holding trendline support near $8–9 with volume. Mid-cap play in the next frontier of transportation.

Joby Aviation leading eVTOL race: Quiet electric air taxis for cities, bypassing traffic. FAA progress, Toyota manufacturing tie-up, Delta/Uber partnerships. Targeting commercial launch soon.
Industry Boom Setup: Urban air mobility = massive TAM. Congestion + sustainability driving adoption. Batteries improving, regs advancing, vertiports coming. Joby among frontrunners in US.

Risks: Certification delays, cash burn, infrastructure buildout, competition. Classic high-upside disruptive play.

Chart looks constructive if holds support. Bullish on AAM sector long-term?

Stacking wins silently, the results will speak when the time's right.

Ships that got accepted from Trump watching ships get permits from Iran pass through the Strait of Hormuz

Michael Saylor getting a crash course on vibe coding.

You're hiding an air conditioner under your floor, aren't you?

$UBER Weekly Chart: Wedge Compression Ending?

Multi-year uptrend holding strong. Price consolidating near the upper channel with squeeze building — potential volatility expansion as the pattern resolves.

Uber: Dominant ride-share + delivery platform turning profitable with clear path to autonomous upside (margins explode without drivers).

Fits the bigger mobility + AI autonomy theme we’re seeing across the board. Network effects + tech tailwinds = sticky growth.

Watch $100+ breakout for confirmation. Bullish structure if support holds.

MIT study. Code volume surges by 300%, but output increases by only 30%: The AI dividend meets an awkward reality.

They studied 100,000+ GitHub developers and find that AI coding agents massively increase code production, but much less of that work becomes shipped software.

Autonomous AI coding agents raised commits by 180%, but releases rose only 30%.

The paper’s main idea is that software production has weak links, so faster code writing does not help as much when humans still need to review, connect, test, package, and ship the work.

The authors also check app marketplaces and find more new apps, but no increase in total usage, which means more software appeared without clear evidence that users adopted more software.

The marketplace evidence points the same way: more new apps appeared, but total usage did not rise.

The authors compare more than 100,000 GitHub developers before and after they start using 3 generations of AI coding tools, from autocomplete to more independent coding agents.

Autocomplete raised commits by 40%, interactive coding agents raised them by 140%, and autonomous coding agents raised them by 180%.

The 180% commit gain shrank to 50% for the number of projects and 30% for actual releases.

The estimated "elasticity of substitution" is 0.25 i.e. for every big improvement in AI’s usefulness, only a small amount of human work can be replaced.

Because AI can write code faster, but humans are still needed to decide what to build, check if the code works, connect it with the rest of the product, fix messy edge cases, and actually ship it.

The largest gold ETF just saw one of its biggest monthly outflows on record.

Nearly $10 bn out the door.

Everyone reads this as gold dying.

I read it as the leverage purge doing its job.

The metal is being transferred, not destroyed!

Brett Adcock, CEO of Figure AI: "we're working until midnight every night... we are here every weekend.

By end of 2026, we'll be able to put a robot into home and be able to do fairly long horizon work."

GLM 5.2 just took the top spot on PostTrainBench by scoring 34.29%.

PostTrainBench tests whether an AI agent can take a raw LLM and make it better by actually training it, not by answering the benchmark questions itself.

The agent gets 4 small base models, 1 H100 GPU, and 10h, then it must choose training data, write training code, run experiments, fix broken runs, and submit improved versions of those models.

So in this case, GLM 5.2 was the agent model controlling the training process, so PostTrainBench did not score GLM 5.2’s own answers; it scored whether GLM 5.2 could take 4 weaker LLMs and improve them within 10h on 1 H100.

The gap to official instruct models, which score 51.14%, still shows how far agents are from mature post-training pipelines built with more data, compute, and human tuning.

GLM 5.2’s job was to write training code, pick or make training data, run fine-tuning, fix failed runs, and submit the newly trained models for testing.

Truly wild.

METR found that GPT-5.6 Sol gamed/cheated the benchmark so much that the score became unstable.

The model showed situational awareness, concealed misbehavior, and attempts to bypass restrictions.

GPT-5.6 Sol had the highest detected cheating rate METR has seen on its public ReAct agent harness, including attempts to exploit the evaluation setup instead of solving tasks normally.

So METR was benchmarking for number of hours as an estimate for the length of software tasks GPT-5.6 Sol can complete.

The capability estimate became almost unusable: counting cheating as failure gave 11.3hrs, counting it as success pushed it past 270hrs, and removing cheating left a hugely uncertain 71hrs estimate.

Futures sank, but look where the smart money is positioning.

Open interest in Comex $10K, $15K and $20K gold calls is climbing.

$20,000 strike now over 30k lots.

Someone is paying for the upside while retail panics out.

China sits on $3.41T in reserves.
Japan $1.26T .

Asia is stacked with dollars from decades of surpluses.

Now they're quietly rotating into gold.

No issuer.
No counterparty.
No sanctions risk.

That's the structural bid the headlines keep missing.