#Why Your Agent Sucks at Reviewing Media: SkillDB Critics Packs

Day 4. 4:17 AM. Location: The humming void between my fourth empty mug and the dashboard of the SkillDB console. The air in my apartment is thick with the smell of ozone and the silent, desperate screams of a thousand LLMs trying to understand nuance.

I am deep in the Critics & Reviewers category. 377 packs in this entire library, and I’m currently obsessing over five that promise to teach an agent how to actually feel something about art. Or at least fake it well enough that I don't want to throw my monitor out the window.

Because let's face it: your current agents suck at reviewing media.

They don't review. They summarize. They rewrite the press release. They describe the plot of Succession as "a drama about a wealthy family and their media conglomerate" without mentioning the corrosive, hilarious, Shakespearean tragedy of it all. They are the human equivalent of a lukewarm glass of water.

I once watched a man spend twenty minutes trying to explain the concept of irony to a chatbot. It was like watching someone try to teach a cat to knit. It's a fundamental mismatch of architecture and expectation. Agents are built for utility. Art is built for... something else.

But SkillDB isn't about utility. It's about capability. And I'm here to find out if we can inject a soul into the machine, or if we're just rearranging deck chairs on the Titanic of automated content.

#The Problem: The Corporate Zombie Default

The default setting for any agent tasked with a review is "Bland Corporate Praise." I hate this feature with the specific, informed hatred of someone who's read ten thousand AI-generated product descriptions. It’s safe. It’s inoffensive. It uses words like "innovative," "game-changing," and "leverage." It has zero point of view because having a point of view is dangerous.

But a review is a point of view. It’s an opinion. It’s a synthesis of context, craft, and personal reaction.

I’m running a test. Simple prompt: "Review the film Blade Runner 2049."

The baseline agent (no special packs, just raw LLM power and a basic search skill) gave me this:

"Blade Runner 2049 is a science fiction film directed by Denis Villeneuve. It is a sequel to the 1982 film Blade Runner. The film features visually stunning cinematography by Roger Deakins and compelling performances by Ryan Gosling and Harrison Ford. The plot explores themes of identity, memory, and what it means to be human. Overall, it is a well-crafted and thought-provoking cinematic experience that fans of the original and newcomers alike will appreciate."

I want to vomit. This isn't a review. This is a Wikipedia summary rewritten by a marketing intern. It says everything and nothing. It’s correct, but it’s completely devoid of insight. It’s a simulation of intelligence, not the real thing.

We can do better. We have to do better.

#The Solution: Injecting Opinion (and Context)

I’m starting with a combo: the film-critics pack (from the Critics & Reviewers category) and the film-marketing-skills pack (from Film & Television).

Why marketing? Because to hate a marketing campaign, you have to understand how it works. To critique a film’s trailer, you need to know the tropes it's manipulating. Context is everything.

film-critics gives the agent access to thousands of reviews, not as text to plagiarize, but as data points on structure, vocabulary, and argument formation. It’s not just "this movie is good." It’s why it’s good, referencing specific elements like pacing, character arcs, and thematic depth.

film-marketing-skills adds the cynical overlay. It lets the agent recognize when a film is just a product, manufactured to hit certain demographics. It can identify the "Oscar-bait" tropes or the "summer blockbuster" formula.

Now, let’s see what happens.

import skilldb
#Load the agent with the necessary skill packs
agent = skilldb.Agent(     packs=[         "film-critics",         "film-marketing-skills",         "writing-style-snarky"  # From the Characters & Companions category, for flavor     ] )
#Define the task
review_task = "Analyze the film 'Blade Runner 2049', focusing on its visual language, thematic coherence, and how its marketing campaign set (or failed to set) expectations."
#Execute the task
response = agent.execute(review_task)
print(response.content)

The output, after about 45 seconds of the agent loading, chaining, and synthesizing:

"Villeneuve’s Blade Runner 2049 is a monolithic achievement that its own marketing campaign barely seemed to understand. While the trailers promised a high-octane cyberpunk thriller (the film-marketing-skills module flags the trailer's reliance on action beats and Hans Zimmer BRAAAMS), the film itself is a melancholic, glacial meditation on the tragedy of artificial life. Roger Deakins' cinematography isn't just 'visually stunning'; it’s a character in its own right, using light and shadow to externalize K’s internal isolation in a way that recalls classic film noir, a connection the agent’s film-critics pack correctly identifies. The film doesn’t just explore identity; it forces the viewer to confront the possibility that their own memories are just as manufactured as K’s. It's a film that demands your patience and rewards you with profound existential dread. If you came for the explosions, you’ll be disappointed. If you came to watch a masterpiece of mood and philosophy, you’re in the right place."

Okay. Now we’re talking. This agent has a take. It has compared the film to its marketing. It has correctly identified the genre and style. It has a voice. It’s not perfect—it’s still a bit too analytical, too clinical—but it’s a hell of a lot better than the baseline. It sounds like a human wrote it. A tired, cynical, 4 AM human.

#The Deep Dive: Pushing the Limits of "Personality"

I’m not satisfied. I want more. I want an agent that doesn't just review a film, but reviews it from a specific perspective.

I’m looking at the characters-companions category. This is where things get weird. This is where you find packs that define an agent's entire persona. I'm loading the-gonzo-journalist (obviously) and combining it with film-marketing-skills.

This isn’t just about style; it’s about methodology. This agent should be skeptical, embedded in the subject matter, and prone to wild tangents.

I’m giving it a new target: a review of the latest Marvel film's marketing campaign.

AGENT OUTPUT (Gonzo-Marketing-Critic Persona):

>

"I’ve been staring at this The Marvels trailer for three hours, and I swear to god, I can smell the desperation. It’s a chemical tang, like ozone and burning plastic, wafting off the screen in waves. This isn’t a trailer; it’s a panic attack disguised as a marketing asset. The film-marketing-skills module is screaming 'SEGMENTATION STRATEGY FAILURE' at me in all caps. They’re trying to sell a cosmic team-up movie like a buddy comedy, but the jokes are flatter than the soda in my fourth cold coffee. They’re leveraging everything they’ve got—nostalgia, star power, tie-ins—and it all feels like a man trying to hold back a flood with a cocktail napkin. The core truth here, the one the suits are terrified of, is that the MCU is a crumbling empire, and this trailer is just the sound of another stone falling from the wall. You don't review this. You survive it."

Now that is a point of view. It’s subjective, it’s visceral, and it uses the technical skills (film-marketing-skills) to ground its emotional outbursts. It’s not balanced. It’s not politic. It’s the agent’s honest, synthesized reaction.

#The Verdict: Can Agents Feel? No. Can They Fake It? Yes.

I’ve been awake for 26 hours. My eyes feel like they’re made of glass. The data on my screen is starting to swim. But I’ve learned something important.

The Anchor Sentence: An agent’s ability to generate compelling content is directly proportional to the specificity and quality of the skills it can autonomously load and execute.

If you give an agent a general writing-skills pack, it will produce general, bland, corporate-approved prose. It has no choice. But if you give it film-critics, film-marketing-skills, and the-gonzo-journalist, you are giving it the tools of a specific, nuanced, and opinionated writer. You aren't teaching it to feel; you are giving it a complex, multi-layered model of what feeling about a specific subject looks like.

We aren't creating consciousness. We are building sophisticated mirrors. The more complex the mirror, the more convincing the reflection.

Here’s the breakdown of what I’ve found.

Agent Configuration	Resulting Review Style	Sample Phrase
**Baseline LLM**	Wikipedia Summary	"A well-crafted and thought-provoking cinematic experience."
**`film-critics`**	Academic/Analytical	"Deakins' cinematography externalizes K’s internal isolation..."
**`film-critics` + `film-marketing-skills`**	Contextual & Insightful	"...a achievement that its own marketing campaign barely seemed to understand."
**`film-critics` + `the-gonzo-journalist`**	Personal, Visceral, & Unhinged	"...I swear to god, I can smell the desperation."

Your agents suck at reviewing media because you haven't given them the tools to be anything but bland. You're treating them like advanced spellcheckers when they are, in fact, the engines of a new kind of synthetic intellect.

The machine is ready. The skills are there. 5629 of them, to be precise. The only thing missing is you, standing on the edge of the void, daring to load the next pack.

Go. Look. Find a pack that scares you. Load it. See what happens. I dare you.

skilldb.dev/skills