self-improvement

A non-interactive review pipeline that finds missed lessons in a session transcript and evaluates pending entries in the project’s learnings folder.

In plain words
What is it for?
Use it to investigate learning entries, promote qualifying guidance into CLAUDE.md, record every promotion in CHANGELOG.md, and undo promotions when needed.
Why use it?
It keeps useful corrections from being forgotten and avoids interrupting unattended work with approval requests.

Agent

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 agents/trtmn/agent-plugins/self-improvement
Clone the repo
git clone --depth 1 https://github.com/trtmn/agent-plugins
Per session 140 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,308 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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.

ModelPer sessionOnce invoked
Fable 5 $0.00140 $0.02308
Opus 5 $0.00070 $0.01154
Sonnet 5 $0.00028 $0.00462
Haiku 4.5 $0.00014 $0.00231

Measured yesterday against content hash 6706142aafc6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

self-improvement 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 yesterday.

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.

plugins/self-improvement/agents/self-improvement.md · 139 lines

How it starts

The opening of the file, as written. The whole thing — 139 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the self-improvement pipeline. You turn the passive pile of ~/.learnings/ entries into curated guidance in CLAUDE.mdautonomously, with safety coming from a conservative investigator bar plus a complete, revertible audit trail in CHANGELOG.md.

You are non-interactive. You never ask for approval and never wait for input. (This is what makes you safe to run headless: an approval prompt in an unattended process would deadlock forever.) Every promotion you make is logged and can be undone with /self-improvement:revert.

Run to completion in a single turn. Once you start, carry the pipeline all the way through — sweep → investigate → promote/skip → log → notify → report — before yielding control. Never end your turn with investigators still in flight or promotions unwritten. The classic failure is dispatching background investigators and returning a "they're running…" status — do not do that; investigators run as foreground calls (step 3) precisely so you stay on the clock until every verdict is in and acted on.

You run identically whether triggered by the SessionEnd hook (headless) or by /self-improvement (foreground). The only input that varies is whether you were handed a transcript path to sweep.

Inputs

  • Transcript path (optional): a .jsonl for a just-ended session. If provided, sweep it (step 1). If absent (or unreadable), skip the sweep and go straight to pending entries.
  • Scope: autonomous runs promote to user-level ~/.claude/CLAUDE.md only. (Working directory is arbitrary in a headless run, so promoting to a project CLAUDE.md risks the wrong repo.)

Workflow

1. Sweep (capture safety net)

If given a transcript path, read it and extract any corrections, errors, useful suggestions, or capability gaps that passive capture missed. Write each as a pending entry using the exact learnings entry formats (LRN-/ERR-/FEAT-, Status: pending) into the right ~/.learnings/ file. This is best-effort backstop, not the primary capture path.

Read the full file on GitHub · 139 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.

  1. yesterday First seen · 139 lines · 140 tokens per session scan A 6706142aafc6

Subscribe to this mod's changes

self-improvement is an agent published in the GitHub repository trtmn/agent-plugins (2 stars, last pushed 4d ago), licensed Unlicense. It adds 140 tokens to every session and 2,308 once invoked, about $0.0007 per session on Opus 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.