memory-compounding

A process for keeping an agent's saved memory accurate, useful, and free of repeated or outdated lessons. It reviews memory files and stored entries and assigns each one an action, such as keep, update, combine, replace, or delete.

In plain words
What is it for?
Use it during monthly memory cleanup, after a significant debugging session, when a lesson appears several times, or when saved advice no longer matches the project.
Why use it?
Unreviewed memory can grow stale, repeat the same advice, or conflict with the current code. Regular cleanup helps future work rely on clearer information.

Skill for Claude CodeCodex

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 skills/romiluz13/auto-pi/memory-compounding
Any agent
npx skills add romiluz13/auto-pi --skill memory-compounding
Clone the repo
git clone --depth 1 https://github.com/romiluz13/auto-pi

Made for: Claude Code, Codex.

Per session 41 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 671 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 $0.00041 $0.00671
Opus 5 $0.00020 $0.00336
Sonnet 5 $0.00008 $0.00134
Haiku 4.5 $0.00004 $0.00067

Measured 2d ago against content hash 389fb4896a83, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory-compounding 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.

skills/memory-compounding/SKILL.md · 66 lines

How it starts

The opening of the file, as written. The whole thing — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Memory Compounding

Memory is an index, not a transcript. Without review it accumulates stale entries that mislead future work. This skill is the discipline for keeping memory sharp, not just large.

When to use

  • Monthly memory hygiene (AGENTS.md: "review both, prune stale entries").
  • You notice the same gotcha recorded 3+ times.
  • A memory entry contradicts the current code (trust the code).
  • After a non-trivial debug/build cycle that produced a reusable learning.

The 5-outcome review rubric

For each entry in ~/.pi/agent/pi-hermes-memory/MEMORY.md, USER.md, failures.md, and the SQLite memories table (via memory_search), apply exactly one outcome:

Outcome When Action
Keep Accurate + useful Leave as-is
Update Correct but incomplete Add the missing detail
Consolidate Same lesson appears multiple times Merge into one entry, remove duplicates
Replace Outdated or superseded by current code Replace with the current truth
Delete No longer applies (framework changed, code removed) Remove

The promotion rule

When the same gotcha appears 3+ times in failures.md or across memory entries, promote it: write a dedicated reference file or a new skill section. A lesson that keeps recurring is a pattern, not a one-off — it earns a permanent home.

Solution docs (durable cross-project learnings)

After any non-trivial debug/build cycle, evaluate whether to write a solution doc to docs/solutions/:

  • Write if: the problem took 3+ hypotheses to solve, OR the bug pattern appears in 3+ files, OR the solution contradicts a common assumption.
  • Skip if: the fix was mechanical (typo, import error, one-line change).

Solution doc format:

# [Problem Title]
Category: debugging | architecture | testing | workflow | conventions
Tags: [comma-separated]
Date: YYYY-MM-DD
## Problem
[symptoms, not just the error message]
## What Didn't Work
[failed hypotheses and why]
## Solution
[what worked — with code example]
## Why
[the underlying principle]
## Prevention
[how to prevent this class of problem]

Read the full file on GitHub · 66 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. 2d ago First seen · 66 lines · 41 tokens per session scan A 389fb4896a83

Subscribe to this mod's changes

memory-compounding is a skill published in the GitHub repository romiluz13/auto-pi (11 stars, last pushed 17d ago), licensed MIT. It adds 41 tokens to every session and 671 once invoked, about $0.0002 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 2026-08-30.