notnessie

Lifecycle hooks

NotNessie registers five Claude Code lifecycle hooks in .claude/settings.json (written by notnessie init). Hooks are how memory flows automatically: context goes in at the start of a session and summaries come out at the end.

Each hook is just the CLI invoked with an event name. Claude Code runs them and passes a JSON event payload on stdin; you do not run them by hand.

{
  "hooks": {
    "SessionStart":     [{ "matcher": "", "hooks": [{ "type": "command", "command": "notnessie hook session-start" }] }],
    "UserPromptSubmit": [{ "matcher": "", "hooks": [{ "type": "command", "command": "notnessie hook user-prompt-submit" }] }],
    "PostToolUse":      [{ "matcher": "", "hooks": [{ "type": "command", "command": "notnessie hook post-tool-use" }] }],
    "Stop":             [{ "matcher": "", "hooks": [{ "type": "command", "command": "notnessie hook stop" }] }],
    "PreCompact":       [{ "matcher": "", "hooks": [{ "type": "command", "command": "notnessie hook pre-compact" }] }]
  }
}

!!! note “Hooks never break your session” Every hook reads stdin, runs its handler, and always exits cleanly — even on error, it just logs to stderr. Hooks also close the database pool on exit so they never hold the session open.

The five hooks

SessionStart — inject context

Runs when a Claude Code session starts. It builds a context pack for the repository and injects it, so the assistant begins with the most relevant decisions, constraints, verified commands, and open questions already in view.

UserPromptSubmit — refresh task context

Runs when you submit a prompt. It can surface task-specific context for what you just asked, keeping the most relevant memory in front of the assistant as the conversation moves between tasks.

PostToolUse — capture verified commands

Runs after Claude Code uses a tool. It captures memory conservatively: it records recognized, non-destructive, secret-free build/test commands that succeeded as verified commands. It does not store raw terminal output or arbitrary tool results.

Stop — summarize the session

Runs when a session ends. It summarizes the session into memory — decisions made, files changed, commands verified, failed approaches, and new TODOs — and persists the result. Summaries are produced with the Claude Agent SDK, with a heuristic fallback when the SDK or its auth is unavailable.

PreCompact — summarize before compaction

Runs just before Claude Code compacts the conversation. It performs the same summarization as Stop, so the useful context of a long conversation is captured into memory before the transcript is shortened.

What hooks respect

Hooks honor your project configuration:

• They follow the memory mode and the autoSave* flags — for example, they skip auto-saving task summaries when autoSaveTaskSummaries is off, and skip command capture when autoSaveCommands is off.
• They apply secret redaction (refuse credential dumps, redact tokens inline) exactly like the MCP save tools.
• They never store full transcripts or raw output.

Disabling or customizing hooks

The hooks live in .claude/settings.json, which is a normal Claude Code settings file. To turn off a particular hook, remove its entry; to change behavior more broadly, adjust the config flags rather than the hook wiring. Re-running notnessie init --force regenerates the default hook set.

This site is open source. Improve this page.