Summary

Adds an output_format property to the cli_command_executed telemetry event and fixes a Go/TS divergence in how is_agent is computed. Together these make the JSON-vs-human output choice measurable and ensure both CLI shells agree on what counts as an agent. This is the measurement-and-parity groundwork for auto-switching agents to JSON output (next subissue under GROWTH-806).

Changes

• Record the resolved output format on every cli_command_executed. Added output_format to the Go catalog (events.go) and its TS mirror (event-catalog.ts), and emit it from the Go capture and both TS instrumentation paths. db query resolves its format into a command-local flag, so it now mirrors the resolved value onto the global that telemetry reads, keeping the value accurate for the highest-volume agent command. The property does not exist today, so current JSON usage is unmeasurable without it.
• Fix is_agent Go/TS parity. The TS legacy analytics layer folded CI into is_agent (is_agent = aiTool || isCi), tagging CI environments as agents while the Go binary does not. is_agent now reflects coding-agent detection only, matching Go on both shells. is_ci continues to be reported as its own separate property.

Notes for reviewers

• output_format values are shell-native (Go: pretty|json|yaml|toml|env; TS: text|json|stream-json), disambiguated by $lib; normalize in analysis.
• Deliberately out of scope: aligning the CI env-var lists across the two shells. runtime.isCi gates behavior in legacy-platform-api.layer.ts, so changing those lists would alter non-interactive behavior for some CI environments. That belongs in a separate change, not this telemetry-parity fix.
• The three instrumentation/platform-api unit tests now provide a mock Output because the instrumentation reads output.format.

Linear

• fixes GROWTH-912
• parent GROWTH-806