Unified event format for AI coding agent actions. Nose auto-discovers agent sessions on your machine, parses their logs, and emits a consistent JSONL event stream.

Lang: Rust | Output: JSONL | Platform: macOS, Linux, Windows

cargo install --path .

Or build locally:

cargo build --release
# binary at target/release/nose

Run from within any project directory where you've used an AI coding agent:

cd my-project/
nose parse

Nose scopes to the current working directory. It only parses sessions from the project you're in.

Output goes to stdout as JSONL (one JSON event per line):

# All events
nose parse > events.jsonl

# Only new events since last run (uses ~/.nose/offsets.json bookmark)
nose parse --new

# Count events
nose parse | wc -l

# Filter for tool calls only
nose parse | jq 'select(.event_type == "ToolCall")'

# See what files the agent touched
nose parse | jq 'select(.event_type == "FileWrite") | .path'

# Get all shell commands the agent ran
nose parse | jq 'select(.event_type == "CommandExec") | .command'

# Count events by type
nose parse | jq -r '.event_type' | sort | uniq -c | sort -rn

nose stats

Shows a summary: sessions, events by type, token usage, models used, top tools, files touched.

nose watch

Streams events to stdout as they happen. Watches agent log files and ~/.nose/events/ for changes.

Install hooks into all detected agents:

nose hooks install

This configures Claude Code, Codex CLI, and Gemini CLI to emit events in real-time to ~/.nose/events/. Events are captured as they happen, with confidence: native.

Remove hooks:

nose hooks uninstall

Hooks are non-destructive - they append to existing agent hook configs and only remove nose-managed entries on uninstall.

All events share common fields:

What events nose parse can extract from each agent's native log files.

✅ direct ⚠️ derived from other data ❌ agent doesn't expose this 🔲 planned

What events nose hooks install captures in real-time from each agent.

✅ direct ⚠️ derived from other hook events ❌ agent doesn't expose this 🔲 planned

graph LR
    CWD[cwd] --> D[Discovery]
    D --> C["Claude ~/.claude/projects/{cwd}/"]
    D --> X["Codex ~/.codex/sessions/{cwd}/"]
    D --> G["Gemini ~/.gemini/projects/{cwd}/"]
    D --> R["Cursor .../workspaceStorage/"]
    D --> P["Copilot ~/.github-copilot/"]
    D --> H["Hooks ~/.nose/events/"]
    C --> A[adapter.parse]
    X --> A
    G --> A
    R --> A
    P --> A
    H --> A
    A --> E[Unified Events]
    E --> J[JSONL stdout]

graph LR
    AG[Agent runs] --> HK[Agent hook fires]
    HK --> NH["nose hook-handler --agent X --event Y"]
    NH --> T[Transform to unified Event]
    T --> F["~/.nose/events/{agent}_{session}.jsonl"]
    F --> NP["nose parse (reads hook events)"]