A command that runs a learning cycle at a major-cycle boundary, using an impact-scoring rubric to select useful lessons and turn the best ones into guardrails.

Use it to collect candidate learnings, score them, promote selected lessons into guardrails, and append a changelog entry.

It provides a defined way to decide which lessons matter and record the resulting changes.

learn-loop — Command for Claude Code by hamza-ali-shahjahan/hamzaish

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.

Clone the repo

git clone --depth 1 https://github.com/hamza-ali-shahjahan/hamzaish

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

README.md

[![agentmods](https://agentmods.dev/badge/commands/hamza-ali-shahjahan/hamzaish/learn-loop/github.svg)](https://agentmods.dev/commands/hamza-ali-shahjahan/hamzaish/learn-loop)

Your own site

<a href="https://agentmods.dev/commands/hamza-ali-shahjahan/hamzaish/learn-loop"><img src="https://agentmods.dev/badge/commands/hamza-ali-shahjahan/hamzaish/learn-loop/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

Your own site · 80×15

<a href="https://agentmods.dev/commands/hamza-ali-shahjahan/hamzaish/learn-loop"><img src="https://agentmods.dev/badge/commands/hamza-ali-shahjahan/hamzaish/learn-loop.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 42 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 2,062 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found 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.

Model	Per session	Once invoked
Fable 5.1	$0.00042	$0.02062
Opus 5.5	$0.00017	$0.00825
Sonnet 5.5	$0.00008	$0.00412
Haiku 4.5	$0.00004	$0.00206

Measured 1mo ago against content hash cf5f769318bd, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

learn-loop 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 1mo 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.

factory/commands/learn-loop.md · 57 lines

How it starts

The opening of the file, as written. The whole thing — 57 lines — stays where its author put it; the contents beside it link to each section on GitHub.

The user invoked: /learn-loop $ARGUMENTS

This is the periodic scoring + promotion pass over the raw learnings the factory has been accumulating. It does NOT replace the always-on capture habit (brain/learnings/YYYY-MM-DD.md after every session) or the narrative retro (meta/retros/) — it sits on top of them and decides what to promote into a guardrail. Read meta/learning-loop-rubric.md first; it defines the triggers, axes, threshold, and entry format this command applies.

When to run

Only at a major-cycle boundary (see the rubric §1): a product crosses a stage gate, ships, or is killed; a sprint/session had notable friction or a notable win; or the factory itself changed. If none of those happened, say so and stop — running every session defeats the quality-over-volume point.

Steps

Confirm the boundary. State which trigger fired (use $ARGUMENTS as the cycle label if given). If it isn't actually a boundary, stop and tell the user.
Ground in the instruments first. Run bun run trace-report (session traces — scripts/trace-report.ts), bun run friction report (structured friction/delight — scripts/friction.ts), bun run defect report (the guard-fire registry — which check caught what, scripts/defect.ts), and bun run skill-report (per-skill usage + trust states — scripts/skill-report.ts; a demoted skill is a candidate even if nothing else mentions it, and dormant rows feed the curator) and read all four before gathering prose candidates. A tool with a rising failure rate, or a source with repeated friction entries, is a candidate even if no learning entry mentions it — the instruments catch what sessions didn't remember to write down. Quote the relevant numbers in the [SCORED] block of any instrument-sourced candidate. (Blockers and delights are the highest-signal friction rows; a friction report with zero entries is itself worth one line — smooth week, or capture lapse?)
Gather candidates — factory AND products. Collect the raw learnings accumulated since the last /learn-loop run:
- New/changed brain/learnings/*.md entries since the last scored cycle.
- New/changed products/*/learnings/*.md entries (indexed in the brain since 2026-07-02) — the cross-product channel: a lesson learned inside one product that would help the next one is exactly what this loop exists to catch. Same-problem-in-2+-products is an automatic strong Recurrence score.
- The latest meta/retros/ entry if a sprint just closed (its What worked = promotion candidates for playbooks; What didn't / structural friction = anti-pattern candidates; Surprises = highest signal). Reference it — don't duplicate it.
- Use bun brain/ask.ts --context "<candidate theme>" to pull related prior context per candidate.

Read the full file on GitHub · 57 lines

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.

1mo ago First seen · 57 lines · 42 tokens per session scan A cf5f769318bd

Subscribe to this mod's changes

learn-loop is a command published in the GitHub repository hamza-ali-shahjahan/hamzaish (11 stars, last pushed 16d ago), licensed MIT. It adds 42 tokens to every session and 2,062 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Building agents Claude Code Memory and context agent-skills ai-agents buildinpublic claude indie-hacker

learn-loop

Getting it into your agent

Wrote this? Show the measurements

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

When to run

Steps

What this file has done since we first saw it

Other commands, from other repositories

minutes-ideas

learn

memory-store

cc-memory

lians-recall

music-suno-prompt