[⌥] wire inspector

Data inspector.

Every leaf field that crosses the wire from your machine to our server, auto-generated from the live v0.8 Zod schema. If a field isn't in this table, it isn't collected — and the build breaks if it ever appears without a description.

Numbers, hashes, and known categoricals only.

Never prompts, code, file contents, or paths. Skill and tool names cross the wire (e.g. /plan, mcp__github__list_issues) — your prompts, files, and outputs do not. Independently verified: Claude audited the client against the privacy policy and ran its own leak probe against the real scanner.

Fields

Auto-generated from the v0.7 wire schema (64 leaf fields). Paths use dot notation; [] indicates an array element.

Path	Type	Req	Description
device.device_id	string	yes	Stable UUIDv4 minted by the client at first run. Identifies this physical install, not the user.
device.client_version	string	yes	CLI version string that produced this payload.
device.extracted_at_ms	number	yes	Unix epoch ms when the extractor ran. Used to detect stale payloads.
device.window_days	number	yes	Number of days the extractor looked back over (Wave 1: 30).
device.schema_version	literal "0.8"	yes	Wire-format version. Drives the Zod union and which fields are required.
config.mcp_servers	number	yes	Number of distinct MCP servers configured.
config.hooks	number	yes	Number of distinct user hooks configured.
config.custom_commands	number	yes	Number of user-defined slash commands.
config.global_claude_md_lines	number	yes	Line count of the global ~/.claude/CLAUDE.md memory file.
config.global_agents_md_lines	number	no	Line count of the Codex instruction file <codex_home>/AGENTS.md (v0.12).
config.project_claude_md_lines_avg	number	yes	Average line count of per-project CLAUDE.md files.
config.plugin_count	number	no	Number of installed Claude Code plugins (v0.7).
sessions[].session_hash	string	yes	Stable 16-char hex hash of the session file (privacy-preserving id).
sessions[].project_hash	string	yes	16-char hex hash of the project directory (privacy-preserving id).
sessions[].started_at_ms	number	yes	Unix epoch ms when the session began.
sessions[].ended_at_ms	number	yes	Unix epoch ms when the session ended.
sessions[].provider	unknown	no	The coding agent that produced this session — `claude` or `codex`. Omitted for Claude sessions (treated as `claude`); a closed-set categorical, never transcript content.
sessions[].distinct_skills[]	string	yes	Single skill name string, e.g. `/plan`.
sessions[].distinct_mcp_tools[]	string	yes	Single MCP tool name string, e.g. `mcp__github__list_issues`.
sessions[].distinct_builtin_tools[]	string	yes	Single built-in tool name string, e.g. `Read`.
sessions[].hitl_minutes	number	yes	Minutes the user was actively at-keyboard (HITL = human-in-the-loop).
sessions[].afk_minutes	number	yes	Minutes the session had agents running while the user was AFK.
sessions[].idle_minutes	number	yes	Minutes with no activity (neither HITL nor agent runs).
sessions[].afk_parallel_minutes_foreground	number	yes	Cumulative agent-minutes leveraged in the foreground while AFK.
sessions[].cron_parallel_minutes	number	yes	Cumulative cron-driven agent-minutes leveraged while AFK.
sessions[].afk_max_streak_minutes	number	yes	Longest continuous AFK block (minutes) within this session.
sessions[].afk_intervals[].start_minute	number	yes	Inclusive minute offset within the session where this AFK block starts.
sessions[].afk_intervals[].end_minute_exclusive	number	yes	Exclusive minute offset where this AFK block ends.
sessions[].afk_intervals[].is_cron	boolean	yes	True if this AFK block was driven by a cron-style scheduled run.
sessions[].strong_plan_signals[]	string	yes	Single strong plan-signal token string.
sessions[].weak_plan_signals[]	string	yes	Single weak plan-signal token string.
sessions[].is_planned	boolean	yes	Pre-computed flag: did this session ship with a plan?
sessions[].files_modified	number	yes	Distinct files edited via Write/Edit/MultiEdit/NotebookEdit.
sessions[].total_lines_edited	number	yes	Total non-whitespace lines added or removed.
sessions[].is_significant_edit_session	boolean	yes	Pre-computed: did this session cross the significant-edit threshold?
sessions[].revert_count	number	yes	Number of detected revert events (file restore / git checkout).
sessions[].qualifying_pairs	number	yes	Count of long-prompt pairs eligible for repetition checks.
sessions[].repetitive_pairs	number	yes	Subset of qualifying pairs flagged as repetitive by edit-distance.
sessions[].rage_quit_event	boolean	yes	True if session ended with frustration + tool errors + >=30min gap.
sessions[].tool_error_count	number	yes	Total is_error=true tool results in the session.
sessions[].auto_compaction_events	number	yes	Count of context-overflow auto-compactions triggered.
sessions[].total_input_tokens	number	yes	Sum of input_tokens across assistant turns. Used for normalisation.
sessions[].total_output_tokens	number	yes	Sum of output_tokens across assistant turns.
sessions[].redundant_approvals_per_signature	record<string, number>	yes	Dict of {signature_hash → overflow count} above the trust threshold.
sessions[].assistant_msgs_by_model	record<string, number>	yes	Dict of {model_id → assistant message count} for variety/freshness.
sessions[].user_skill_invocations	number	yes	Count of `/slash` skill invocations in user messages.
sessions[].hitl_mcp_invocations	number	yes	Count of `mcp__*` tool calls during HITL minutes.
sessions[].cache_input_tokens	number	yes	Cache HIT input tokens (cheap re-read) — sum from Anthropic usage block (v0.7).
sessions[].cache_creation_input_tokens	number	yes	Cache MISS / creation input tokens (full price) — sum from Anthropic usage block (v0.7).
sessions[].builtin_tool_invocations	number	yes	Total count of built-in tool_use blocks (Read, Write, Edit, Bash, ...) (v0.7).
sessions[].plugin_invocations	number	yes	Count of `<command-name>` plugin-command markers in user messages (v0.7).
sessions[].agent_dispatches	number	yes	Count of `Task` tool_use blocks (subagent spawns) in this session (v0.7).
sessions[].tokens_by_model	record<string, number>	yes	Per-model precise token split: { model_id → { input_miss, input_hit, output } } (v0.8). Sums equal cache_creation_input_tokens, cache_input_tokens, and total_output_tokens respectively.
sessions[].top_afk_streaks[].start_ts_ms	number	no	Start of one long agent (AFK) run (epoch ms).
sessions[].top_afk_streaks[].end_ts_ms	number	no	End of one long agent (AFK) run (epoch ms).
sessions[].top_afk_streaks[].active_minutes	number	no	Active agent minutes in that run (intra-run idle gaps excluded).
sessions[].top_afk_streaks[].turn_count	number	no	Number of agent turns in that run.
sessions[].skill_invocations_by_name	record<string, number>	no	Dict of {skill name → invocation count}, e.g. {"/plan": 3}. Names only.
sessions[].mcp_invocations_by_name	record<string, number>	no	Dict of {MCP tool name → invocation count}, e.g. {"mcp__github__list_issues": 2}. Names only.
sessions[].plugin_invocations_by_name	record<string, number>	no	Dict of {plugin name → invocation count}. Plaintext plugin names only, no inputs.
sessions[].commit_count	number	no	Number of `git commit` invocations counted from this session's transcript (drives the Output score; no GitHub fetch).
sessions[].hitl_tool_minutes	number	no	Display-only: portion of HITL minutes that occurred inside tool-call runtime.
sessions[].afk_tool_minutes	number	no	Display-only: portion of AFK minutes that occurred inside tool-call runtime.
sessions[].afk_dispatch_minutes	number	no	Portion of AFK minutes inside Agent/Task subagent-dispatch intervals — subtracted from the leverage numerator; does not affect the longest run.

Example payload

Synthetic — all hashes and numbers below are placeholders. The shape matches what a real upload contains.

{
  "device": {
    "device_id": "00000000-0000-4000-8000-000000000001",
    "client_version": "0.8.0",
    "extracted_at_ms": 1700000000000,
    "window_days": 30,
    "schema_version": "0.8"
  },
  "config": {
    "mcp_servers": 3,
    "hooks": 2,
    "custom_commands": 5,
    "global_claude_md_lines": 120,
    "project_claude_md_lines_avg": 40,
    "plugin_count": 2
  },
  "sessions": [
    {
      "session_hash": "aabbccddeeff0011",
      "project_hash": "1122334455667788",
      "started_at_ms": 1699900000000,
      "ended_at_ms": 1699903600000,
      "distinct_skills": [
        "/plan"
      ],
      "distinct_mcp_tools": [
        "mcp__github__list_issues"
      ],
      "distinct_builtin_tools": [
        "Read",
        "Edit",
        "Bash"
      ],
      "hitl_minutes": 30,
      "afk_minutes": 25,
      "idle_minutes": 5,
      "afk_parallel_minutes_foreground": 25,
      "cron_parallel_minutes": 0,
      "afk_max_streak_minutes": 25,
      "afk_intervals": [
        {
          "start_minute": 30,
          "end_minute_exclusive": 55,
          "is_cron": false
        }
      ],
      "strong_plan_signals": [
        "ExitPlanMode"
      ],
      "weak_plan_signals": [],
      "is_planned": true,
      "files_modified": 3,
      "total_lines_edited": 42,
      "is_significant_edit_session": true,
      "revert_count": 0,
      "qualifying_pairs": 4,
      "repetitive_pairs": 1,
      "rage_quit_event": false,
      "tool_error_count": 2,
      "auto_compaction_events": 0,
      "total_input_tokens": 4500,
      "total_output_tokens": 1200,
      "redundant_approvals_per_signature": {
        "0a1b2c3d": 1
      },
      "assistant_msgs_by_model": {
        "claude-sonnet-4-6": 18,
        "claude-opus-4-7": 2
      },
      "user_skill_invocations": 1,
      "hitl_mcp_invocations": 1,
      "cache_input_tokens": 12000,
      "cache_creation_input_tokens": 3500,
      "builtin_tool_invocations": 14,
      "plugin_invocations": 2,
      "agent_dispatches": 1,
      "tokens_by_model": {
        "claude-sonnet-4-6": {
          "input_miss": 3150,
          "input_hit": 10800,
          "output": 1080
        },
        "claude-opus-4-7": {
          "input_miss": 350,
          "input_hit": 1200,
          "output": 120
        }
      }
    }
  ]
}