Agent Memory Loop

Lightweight learning for agents that reset between sessions.

Use this when

• - you want a low-friction way to log mistakes, corrections, and discoveries
• you need recurring lessons without bloating core instructions
• you want human-reviewed promotion instead of auto-writing to instruction files
• you want a quick pre-task scan for known failure patterns

Do not use it for

• - autonomous self-modification
• external content promotion
• heavy multi-section incident writeups by default
• dashboards, registries, or process ceremony

Core workflow

CODEBLOCK0

Install

CODEBLOCK1

Creates:

CODEBLOCK2

Minimal instruction snippet

Add this to your agent instructions:

CODEBLOCK3

The format, in short

One incident or discovery per line. Extra fields are optional.

CODEBLOCK4

Key fields:

• - count:N tracks recurrence
• INLINECODE1 tracks loop closure
• INLINECODE2 forces review even at count 1
• INLINECODE3 is never promotable

Operating rules

1. 1. Log fast; prefer a one-line entry over a perfect writeup
2. Dedup before appending
3. Queue recurring or critical lessons for review
4. Humans approve promotions; agents do not
5. Before major work, scan for relevant prior failures
6. If a learning prevented a repeat mistake, record that with INLINECODE4

References

• - references/logging-format.md — canonical line formats, fields, examples, source labels
• INLINECODE6 — dedup, review queue, pre-task review, trimming rules
• INLINECODE7 — queue entry structure and status lifecycle
• INLINECODE8 — optional detail-file template for complex failures
• INLINECODE9 — why this stays lean instead of turning into a system

Assets and scripts

• - INLINECODE10
• INLINECODE11
• INLINECODE12
• INLINECODE13
• INLINECODE14
• INLINECODE15
• INLINECODE16

Success condition

The loop is working if agents actually use it:

• - learnings are cheap to log
• duplicates stay low
• recurring lessons reach the queue
• promotions stay human-approved
• INLINECODE17 starts climbing on real work