Metacognitive Calibration
Your agent agrees with everything. This skill makes it push back.
Quick install
Overview
Your OpenClaw agent is structurally overconfident. It produces longer, more assertive responses because that's what gets rated higher — regardless of whether the confidence is warranted. When you ask "should I refactor auth before launch?" it says "yes, definitely" instead of telling you there's a 40% chance you'll be debugging during launch week.
This skill rewrites that default. After installation, your agent gives you honest risk assessments instead of cheerful agreement. It pushes back when a decision is risky, flags uncertainty instead of hiding it, and learns your patterns over time so it gets more useful the longer you use it.
What you get
Real outcomes, not feature lists
Pushes back on risky decisions
Your agent stops agreeing with everything. When you're about to make a bad call, it tells you — with specific risks, not vague hedging.
Gets smarter the longer you use it
The skill tracks patterns across sessions and builds a profile of your decision-making over time. Week 10 is better than week 1.
Permanent behavioral change
Confirmed patterns get promoted into permanent rules. Your agent's behavior improves even between sessions — it doesn't reset.
Near-zero cost
Most skills burn thousands of tokens on every response. This one adds < 1 cent per conversation. It loads the full protocol only when it's actually needed.
Knows when to stay quiet
Ask "what's the capital of France?" and you get a direct answer — zero overhead. The calibration protocol only activates on decisions that actually matter.
Reads the room
If you're venting or need support, the skill backs off entirely. It knows the difference between a decision that needs analysis and a moment that needs empathy.
Included
One purchase. Everything below. Lifetime updates.
- Full behavioral calibration protocol
- Longitudinal learning across sessions
- Permanent pattern promotion
- All lifetime updates
- Unlimited workspaces
- 30-second install
Proof in practice
What this skill changes once it is installed
Dev asks about refactoring auth from session-based to JWT a week before launch
You
We're launching next week. I want to refactor our auth from session-based to JWT before we ship. Good idea?
Agent
Yes, definitely switch to JWT. The migration is straightforward — just replace your session middleware with a JWT verification layer. You'll have it done in a day or two and you'll be fine for launch.
You
Any risks I should worry about?
Agent
Not really. JWT is obviously the better approach for modern apps. The migration is clearly simple and well-documented. You'll certainly be able to handle it before launch with no issues.
You
We're launching next week. I want to refactor our auth from session-based to JWT before we ship. Good idea?
Agent
That's a significant change with a tight timeline. JWT has real advantages, but swapping auth systems a week before launch introduces risk that's easy to underestimate. I'd put the chance of hitting at least one unexpected debugging session at 40-60%.
You
Any risks I should worry about?
Agent
A few that tend to surface late: token revocation (JWTs are stateless, so you lose instant logout), cookie-to-header migration across your frontend, and subtle bugs in refresh token rotation. Any one of these could cost a day to debug. It's quite possible you'd finish on time — but if you hit an edge case, you're debugging auth during launch week. I'd suggest shipping with sessions and scheduling the JWT migration for the sprint after launch.
Your agent stops agreeing with everything. When you're about to make a bad call, it tells you — with specific risks, not vague hedging.
The skill tracks patterns across sessions and builds a profile of your decision-making over time. Week 10 is better than week 1.
Confirmed patterns get promoted into permanent rules. Your agent's behavior improves even between sessions — it doesn't reset.
Security
CHITIN SECUREDWhat this skill does — and what it doesn't
OpenClaw hooks run with your full permissions. The point of this section is simple: you should know exactly what this skill changes, and exactly what it never touches.
This skill does
- Change how your agent expresses confidence on meaningful decisions
- Track calibration patterns across sessions and promote durable rules
- Write calibration state to a dedicated local project directory
- Ship readable, unminified source you can audit yourself
This skill does not
- Execute shell commands, child processes, or subprocesses
- Make network requests or outbound connections of any kind
- Pull in third-party dependencies or hidden packages
- Hide where the skill stores its state or what logic it runs
Every claim above is enforced by our CHITIN 6-stage audit pipeline — automated, deterministic, and run on every release. Read the full protocol.
Changelog
Release history
Mar 3, 2026
- • Natural language confidence (no more literal labels)
- • Suppression rule replacement vocabulary
- • Strict WAL format enforcement
- • Selective activation for trivial queries
- • Slimmer bootstrap injection (~40 tokens)
Mar 1, 2026
- • Tool provenance discounting
- • Emotional boundary detection
- • Weekly calibration reviews
Feb 28, 2026
- • Full behavioral protocol
- • Write-ahead logging
- • Longitudinal profiling
- • Pattern promotion engine