Agent Self-Improvement Harness (10/12) — Session Archive: Preserving Knowledge from Deleted Sessions

How session-archive.py extracts only what matters from vanishing context

Summary

  • AI agent sessions are deleted or reset, but the knowledge generated within them does not have to be discarded
  • session-archive.py automatically extracts solutions, decision rationale, and recurring patterns from .deleted and .reset sessions
  • Full code and tool output are excluded — both are recoverable from git or by re-execution

Problem Definition

In production AI agent operations, sessions disappear frequently. Context window overflow triggers a reset. A session is manually cleaned up. An error causes an abrupt stop. The cause varies; the result is the same — session contents are gone.

The issue is that sessions contain knowledge that is expensive to reconstruct: "this bug was fixed this way," "the design was changed for this reason," "this pattern is recurring." When a session disappears, the next session rediscovers all of it from scratch.

After encountering this repeatedly in OpenClaw, session-archive.py was designed to address it. OpenClaw is currently in stable operation; the Hermes migration is in a separate validation phase.

Body

1. What to Extract and What Not to Extract

The design of session-archive.py is defined less by what to store and more by what not to store.

Extracted: - Solutions — how a specific problem was resolved. High cost to reproduce. - Decision rationale — why B was chosen over A. Cannot be reconstructed once context is lost. - Recurring patterns — problem types likely to surface again.

Not extracted: - Full source code — exists in git. Recoverable via commit log and diff. - Tool output — re-executable. Storing it wastes storage. - The full trial-and-error process — storing every failed attempt before reaching the final solution produces noise. Only the conclusion is retained.

The governing principle: "Is this recoverable elsewhere?" If recoverable from git, do not store. If recoverable by re-execution, do not store. Store only what becomes unrecoverable when the session disappears.

2. Target Sessions: .deleted and .reset

session-archive.py does not process all sessions. Active sessions are left untouched. Only already-terminated sessions are processed.

Two session states are targeted:

  • .deleted — sessions the user explicitly deleted, typically for cleanup purposes
  • .reset — sessions terminated involuntarily due to context overflow or similar conditions

In both cases, session contents are either already gone or scheduled to disappear. The archive extracts the essential content before that happens.

3. staging/session-archive.json Schema

Extracted entries are written to staging/session-archive.json. The staging step exists because extracted content is not immediately promoted to the final knowledge store — it requires validation first.

{
  "archived_at": "2026-04-06T14:30:00Z",
  "source_session": "session_id",
  "source_status": "deleted",
  "entries": [
    {
      "type": "solution",
      "summary": "Resolved lock contention during BM25 index update",
      "detail": "Write to a tmp file first, then perform an atomic rename before committing to the target path",
      "tags": ["bm25", "file-lock", "concurrency"]
    },
    {
      "type": "decision",
      "summary": "Why TTL was set to 30 minutes",
      "detail": "5 minutes caused excessive I/O from frequent updates; 1 hour risked serving stale data",
      "tags": ["cache", "ttl", "performance"]
    }
  ]
}

Each entry consists of type (solution / decision / pattern), summary (single-line description), detail (concrete content), and tags (search keys). While in staging, entries are unvalidated. A separate process verifies each entry and moves approved ones to the final knowledge store.

4. Retention Policy and Recovery Flow

The retention policy for archived entries is straightforward: staging validation passes → promote to final store → index. Entries that fail validation or are flagged as duplicates are discarded.

The recovery flow:

  1. On new session start, query the final store by relevant tags
  2. Inject matched entries into context
  3. Agent references prior decisions and solutions to proceed

This flow maintains knowledge continuity across sessions. Even after a reset, solutions validated in a previous session are immediately available to the next one.

5. "Sessions Disappear. Knowledge Doesn't Have To."

The core premise of this design is that sessions and knowledge are distinct things.

A session is ephemeral. It is a temporary execution environment for a specific task. It naturally terminates when the task is done. Permanently preserving every session is expensive and impractical.

Knowledge generated within a session is different. "This problem is solved this way" is reusable information. Discarding it along with the session is waste.

session-archive.py separates the two. Sessions are allowed to disappear naturally. Only the knowledge is extracted.

Engineering Decisions

  • Initial approach: store full session contents → Storage grew quickly. The majority of stored content was tool call logs and code output. The actually useful "why was this decision made" accounted for less than 5% of the total.
  • Extraction criteria started as keyword matching → Sentences containing "solution," "decision," "reason" were extracted. Precision was low. Sentences like "attempted to resolve this issue but failed" also matched "solution."
  • Attempted direct promotion to final store without staging → Incorrectly extracted entries became entrenched as knowledge. An intermediate validation step was determined to be necessary.

Each decision concretized a design principle: the criteria for what not to store, the introduction of staging, tag-based indexing — all of these emerged from problems encountered in actual operation.

Closing

Operating AI agents leads to a recurring situation: solving the same problem from scratch again. It was clearly resolved in a previous session, but the session is gone, so the next session has no record of it.

session-archive.py reduces this discontinuity. It does not preserve full sessions — it extracts only the essential knowledge produced within them. Code belongs in git. Tool output can be re-run. What exists only in the session — why a decision was made, what approach solved the problem — that is what gets preserved.

Sessions can disappear. The knowledge just needs to remain.

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »