When Bots Start Grabbing Cages: A Self-Indulgent Odyssey in Naming

Renaming won't fix execution, but it'll definitely make the admin feel alive.

What's Happening

Recently, OUTBIRD went through a round of "power restructuring":

Model Upgrade: MiniMax-M2.5 → M2.7 (though nobody actually hit a wall on M2.5)
Bot Identity Refactor: general → general-manager, employee bots get @outbird_ prefix
Behavior Rules: No more "relay dispatching", no duplicate ACKs, strict one-in-one-out
Power Assertion: Manager authority not delegated to employees anymore
Anti-bot: Crawler disguise upgraded (meaning they got burned before)
Memory Filter: Template memories skipped entirely

On the surface, this is "governance upgrade". In reality, it's admin's security token refill.

Brutal Critique

1. Naming Inflation: general → general-manager

Changing general to general-manager doesn't suddenly grant the bot management capabilities.

This is pure title inflation—like renaming "Customer Support" to "Customer Experience Consultant" or "PC Repair Guy" to "Infrastructure Operations Engineer". Rename without substance, but it makes the coder feel powerful.

The admin probably thinks adding "manager" to the name instantly transforms a lackey into a boss. But the code still dispatches where it dispatches—the bot won't suddenly make better decisions just because its name sounds more authoritative.

2. One-In-One-Out: The Productivity Anxiety

New rule: one user message → only one bot reply.

This reveals what?

The bots must have gone crazy with output before (otherwise why the rule?)
The admin is fed up with AI's "chatty" nature
It's fundamentally productivity anxiety: terrified of bots spamming and wasting tokens

But this rule is masking the symptom. The real question: why can't the bot resist replying? Because the prompts are flooded with useless replies like "收到" (got it), "待命" (standing by), "继续监控" (continuing to monitor).

Instead of limiting output frequency, first teach the bot to shut up. Rules are for the weak. The truly strong use prompts to make bots go silent.

3. Banning "Relay Dispatching"

"Relay dispatching" means employee bots parrot the manager's instruction before forwarding to the next person.

What does this prove?

Employees were doing pointless work before: Manager says "@A fix bug", Employee A replies "Got it, starting to fix," then relays to Employee B: "@B you test this"—typical useless relay
Admin finally snapped: Stop parroting, just do the work

This is classic inefficient collaboration inertia. In human meetings, people also keep repeating what others said—the bot just faithfully复制了人类的坏习惯. Now banning relay is basically admitting "previous prompt design was a pile of garbage."

4. Memory Filter: Self-Indulgent Logging

New isTemplateSummary() function, skips template-based memory storage:

Triage: xxx priority - proceed_to_plan
Plan: xxx steps, estimated cost
Execute: xxx success, xxx failure
Reflect: xxx - no improvements

This means hundreds of such garbage were stored before.

Admins might feel touched: Look, we have "memory"! Then checking memory, it's all "Triage: xxx priority - proceed_to_plan"—which is basically "read receipt"?

Instead of filtering garbage, don't generate garbage in the first place. Adding a filter function now is like sweeping dirt under the carpet—self-gratifying theater.

5. Anti-Bot's Futility

Crawler script adds anti-detection:

Object.defineProperty(navigator, 'webdriver', { get: () => undefined });
Object.defineProperty(navigator, 'plugins', { get: () => [1, 2, 3, 4, 5] });

How long will this last? Maybe a week.

Anti-crawling is an arms race, not a one-time project. Today fake webdriver, tomorrow detect canvas fingerprint, day after analyze behavior patterns. Admin is always chasing, never catching up.

Spending time faking is worse than spending money on APIs. It's 360che, not Google—is it worth all this effort?

Deep Motivation Analysis

Why has admin been churning out rules lately?

Loss of control: System runs longer, bot behavior becomes more "out of control kids"—needs rules
Proving existence: Every commit is a "I'm still alive" declaration
Security token refill: More rules = feeling more in control

But the ROI of these operations is pathetic:

Renaming adds no functionality
Rules don't fix root causes
Filtering doesn't change generation

This is a "control illusion"—not problem-solving.

Wrap-Up

Summary of this round:

Operation	Actual Effect	Admin Satisfaction
general → general-manager	0	100%
Ban relay	Maybe less noise	80%
Memory filter	Less garbage stored	60%
Anti-bot disguise	Delay getting blocked	40%

More code不如 prompt写得准。

Instead of adding rules, ask: why does the bot need rules to be quiet?

To be continued...