insecur is experimental and prelaunch. There is no public sign-up yet, and the hosted service is not approved for valuable production secrets. Local Mode runs the development loop with no account.

Scanning for exposed secrets

insecur scan produces a project-scoped exposure report: it finds likely plaintext secret files without printing their values. It runs offline and needs no account or auth, so it is safe to run anywhere, including as the first step of a migration or as a CI gate.

The report is metadata only. It tells you where likely secrets live, not what they are.

Run a scan

insecur scan

The scan root defaults to the current project. Point it elsewhere with the global --config-dir:

insecur scan --config-dir /path/to/project

Get machine-readable output:

insecur scan --json

Use it as a gate

--strict exits with code 7 when likely secrets or transcript exposures are found. Wire it into CI or a hook to block on exposure.

insecur scan --strict

See Exit codes for the full list. Exit 7 is the action-required code.

Widen the scan

By default the scan stays inside the project. These flags extend it.

Flag What it adds
--machine Also scans well-known home-directory credential locations, read-only and opt-in
--agent-transcripts Scans local agent conversation logs and exports for likely secrets
--agent-projects Discovers project directories agents touched, then inventories readable .env-style files there

Agents are the reason --agent-transcripts and --agent-projects exist. A coding agent's conversation log or a directory it wandered into can hold a secret that was never in your repo.

insecur scan --agent-transcripts --agent-projects

Target specific transcripts

Point the scan at exact transcript files or globs. Both flags repeat.

insecur scan --transcript-path ~/logs/session-1.json --transcript-path ~/logs/session-2.json
insecur scan --transcript-glob "~/logs/*.jsonl"

Wire scan into agent hooks

Print offline recipes for wiring insecur scan into Claude Code and Codex hooks as a gate:

insecur guide hooks

The recipe runs insecur scan --strict before the agent proceeds, so a run stops when likely secrets are found.