agentleFS
erikunhaerikunha/vault-of-loci/.agents/skills/calibration-coach/SKILL.md

calibration-coach

Skill0 starsChanged 5 days agoMITView on GitHub

Read the learning ledger and show the learner how well their confidence matches the result, then ask for a probability from 0 to 100 before the next reveal and record it; use it when the learner asks whether their confidence is telling them anything, and it writes to the ledger, so run it only when asked.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run the coach and read the report.

    line 14

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Tutor mode: the learner attempts first, and the agent never writes the learner's notes or answers. Hints come one step at a time and only after an attempt: hint, strong hint, partial example, answer. Confidence is asked before the reveal. The agent never grades an attempt the learner did not confirm.

The question this skill answers: is the confidence telling the learner anything? The coach measures the gap between how sure the learner was and what happened. It does not measure what was learned.

  1. Run the coach and read the report. It shows the Brier score with an explicit probability, the Brier score with the 1 to 5 scale, the reliability table and the sharpest cases.
python3 -I 90_System/Scripts/calibration_coach.py
  1. For one objective, or as JSON to read the numbers closely:
python3 -I 90_System/Scripts/calibration_coach.py --objective <id>
python3 -I 90_System/Scripts/calibration_coach.py --json
  1. Say in one line what the gap means for what to study next. Being sure and wrong is a different gap from being unsure and right:
    • Sure and wrong: the material is not where the learner thinks it is. Reread the concept note of that objective and test it again.
    • Unsure and right: the knowledge is there and the confidence is behind it. Keep the same material, use it again later, and let the confidence come up on its own.
    • Many bins with few attempts: the ledger is too small to say anything. Keep going and read it again later.
  2. From the next reveal on, ask for a probability from 0 to 100 instead of only the 1 to 5 scale, and keep asking for the 1 to 5 scale as well. Say that a number in percent is what makes the gap visible. Do not explain the concept while asking.
  3. Record each attempt with the probability the learner gave. The learner confirms the result first. A percentage from 0 to 100 becomes a number from 0 to 1, so 70 becomes 0.7.
python3 -I 90_System/Scripts/record_attempt.py --objective <id> --item <id> --mode <mode> \
  --correct --confidence <1-5> --probability <0-1>
  1. Show the learner the line that was written. Never edit or delete a line of the ledger, and never write the learner's answers, guesses or notes.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.