Agent Integration

Write Ghostable's maintained guidance into the repository's AGENTS.md:

$ ghostable agent init --dry-run
$ ghostable agent init
$ ghostable agent instructions
 

Setup can perform the same integration with --agent-instructions. Review generated guidance before committing it, especially when the repository already has agent rules.

agent capabilities emits the recommended allowlist for default agent behavior:

$ ghostable agent capabilities
$ ghostable agent capabilities --json
 

The list favors status, lists, redacted diffs, history, validation, review, hygiene, and dry-run pull or deploy planning. It is not the complete CLI inventory.

$ ghostable status --json
$ ghostable env list --json
$ ghostable env diff --from staging --to production --json
$ ghostable validate --env staging --json
$ ghostable review --secrets-only --json
$ ghostable access matrix --json
 

Agents should prefer structured output, state assumptions explicitly, and avoid extrapolating plaintext values from metadata.

Pushing, syncing, promoting, deleting, rotating, approving access, revoking devices, pulling to a file, and deploying all change state or expand plaintext exposure. An agent should perform them only when the user explicitly requests that operation and should dry-run or diff first where available.

$ ghostable env diff --env staging --file .env.staging --json
$ ghostable validate --env staging --json
$ ghostable env push --env staging --file .env.staging --reason "Requested configuration update" --json
 

Prefer file-based input for non-interactive var push, redacted JSON diffs, env run --mask-output, and scoped automation credentials. Treat terminal output and tool logs as possible disclosure channels.