agent-wal
bowen31337 作
Notes & PKM概要
Write-Ahead Log protocol for agent state persistence.
説明
このスキルでできること
Write-Ahead Log protocol for persisting agent state to disk before responding. Prevents corrections, decisions, and context from being lost when conversations are compacted. Entries are stored as append-only JSONL and can be replayed at session start to restore lost context.
Unlike relying on in-memory context alone, WAL entries persist to disk and survive compaction, so corrections made early in a session are not silently discarded.
使いどころ
- Preserving a user correction across a session restart
- Logging an architectural decision before continuing a long conversation
- Flushing buffered context before anticipated compaction
- Replaying missed corrections at the start of a new agent session
- Batching state changes during a complex multi-step task
ワークフロー例
User says: 'No, use Podman not Docker for all tooling'
- 1
Runs
wal.py append agent1 correction 'Use Podman not Docker for all tooling'before responding - 2
Confirms entry is written to the JSONL WAL file at ~/clawd/memory/wal/agent1.wal.jsonl
- 3
Responds to the user with the corrected behavior applied
- 4
On next session start, runs
wal.py replay agent1to surface unapplied entries - 5
Marks the entry as applied after incorporating it into context
Correction persisted to disk and restored at the next session start, with no manual re-statement needed from the user
インストール
clawhub install bowen31337/agent-walCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。