learn

A command that drafts a Semgrep rule from a description of a bug or security problem. Semgrep is a code-scanning tool that matches patterns in source files.

In plain words
What is it for?
Use it to describe what failed, generate a proposed rule with test examples, verify its matches, and promote approved rules into the active collection.
Why use it?
It turns a known incident into a repeatable check while keeping the draft separate until a person reviews and approves it.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/milindgaharwar/fettle/learn
Clone the repo
git clone --depth 1 https://github.com/MilindGaharwar/fettle
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 300 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin 95% copy Near-identical to another mod in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.00300
Opus 5 $0.00000 $0.00150
Sonnet 5 $0.00000 $0.00060
Haiku 4.5 $0.00000 $0.00030

Measured 2d ago against content hash 1f84b19209fb, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

learn scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

Origin

This is a copy

95% identical to fettle-learn — 5 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.

commands/learn.md · 48 lines

What it actually says

/fettle:learn

Generate a semgrep rule from an incident description.

Usage

When the user invokes /fettle:learn, ask for the incident details:

  1. What failed? (the bug or vulnerability)
  2. What code pattern caused it?
  3. What should it look like instead?

Then run:

fettle learn --incident "INCIDENT_TEXT" --auto-save

The drafted rule lands in the rules/proposed/ quarantine (status: proposed) with:

  • Semgrep pattern
  • Citation (incident reference)
  • Violating fixture (tests/fixtures/learned/)
  • Clean fixture (tests/fixtures/learned/)

Proposals are NEVER loaded by gates. Promotion to rules/learned/ is an explicit human step.

Verification

After generating, verify the rule works against its fixtures:

semgrep --config rules/proposed/<rule-id>.yml tests/fixtures/learned/<rule-id>_violation.py

Should match the violation fixture and NOT match the clean fixture.

Approval

Review the quarantined proposals and promote the good ones:

fettle rules list
fettle rules promote <rule-id>

Promotion moves the rule to rules/learned/ (status: learned), where it can be loaded via .fettle.toml. fettle rules promote refuses proposals with an empty pattern.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 48 lines · 0 tokens per session scan A 1f84b19209fb

Subscribe to this mod's changes

learn is a command published in the GitHub repository MilindGaharwar/fettle (2 stars, last pushed 3d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 300 tokens. A static security scan graded it A with 0 findings. It is 95% identical to fettle-learn, differing in 5 lines, and is treated as a copy.