Add --json output to read commands

[akshithg]

Adds a --json flag to the read/query commands so agents and scripts can consume untruncated, structured output instead of the formatted tables. Default table rendering is unchanged.

Commands

• list --json — droplets (id, name, status, ip, tailscale_ip, region, size, cost_monthly, in_ssh_config, ssh_hostname, tags), hibernated, total_monthly_cost
• info --json — the list fields plus created_at, vcpus, memory_mb, disk_gb, transfer_tb, image, features, full networks (incl. IPv6)
• list-ssh-keys --json — username and ssh_keys (name/id/fingerprint)
• version --json — version string

Notes

• Shared emit_json() and field-extraction builders are centralized; build_droplet_detail reuses build_droplet_record rather than duplicating extraction.
• Missing values are emitted as null, not placeholder strings.
• JSON mode suppresses decorative status lines.
• Follow-up planned: --json + non-interactive --yes for the mutating commands.

Tests

New tests/test_list.py and tests/test_json_output.py cover the builders and all four commands. Full suite: 291 passed; coverage 30.61%. (The 12 test_rename.py failures are pre-existing on main, unrelated to this change.)

🤖 Generated with Claude Code

[ret2libc]

Automated review — --json output for read commands

Solid refactor; the table path looks behavior-preserving (cost totals, status colors, placeholders, summary line all match the old inline logic). Findings below are mostly about the JSON contract being a bit leakier than its docstring/purpose promises.

1. dropkit/main.py:~2596 (and the except/not-found paths) — errors print non-JSON to stdout in --json mode.
console = Console() writes to stdout, and error/not-found messages aren't guarded by json_output. dropkit info missing --json | jq . emits Error: Droplet 'missing' not found... on stdout (exit 1) instead of structured output — the exact failure the flag exists to prevent. Consider routing errors to stderr (or emitting a JSON error object) in JSON mode.

2. dropkit/main.py:2262, 2283 — builder defaults contradict the documented "missing values are None" contract.
build_droplet_record's docstring says missing values are None "so JSON consumers can branch on them cleanly," but name defaults to "" and cost_monthly to 0.0. An agent can't distinguish a free/$0 resource from one whose price_monthly was absent (both serialize as 0). Either make these None or correct the docstring.

3. dropkit/main.py:2350 — render_droplet_list exceeds the CLAUDE.md hard limit "cyclomatic complexity ≤8".
Two table builds, two row loops, and the summary block stack well past 8 decision points. Splitting the droplet table and hibernated table into separate render helpers would bring it under the limit.

4. dropkit/main.py:2443 — list inlines its --json help instead of reusing JSON_HELP.
The other three commands use the shared JSON_HELP constant; list hardcodes a differently-worded string ("instead of a table" vs "instead of formatted output"). A future change to JSON_HELP will silently skip list.

5. dropkit/main.py:54 — emit_json double-encodes.
console.print_json(json.dumps(payload)) serializes to a string only for print_json to re-parse it. console.print_json(data=payload) does the same in one step.

6. dropkit/main.py:2320 — build_droplet_detail assumes nested objects are non-null.
size = droplet.get("size", {}) returns {} only when the key is absent; an explicit "size": null yields None, and size.get("vcpus") then raises AttributeError. Low likelihood (DO populates these for live droplets), so plausible rather than confirmed — droplet.get("size") or {} guards it.

Nothing blocking for correctness; #1 and #2 most directly affect the machine-readability this PR is adding.

🤖 Generated with Claude Code

[ret2libc]

The JSON error contract needs one more pass before approval. list --json, info --json, and list-ssh-keys --json all call load_config_and_api(), which still writes missing/invalid config errors to stdout and raises typer.Exit before these commands reach emit_error(). I reproduced this with an empty home directory: dropkit list --json exits 1, writes Error: Config not found. Run 'dropkit init' first. to stdout, and writes nothing to stderr. This breaks the documented clean-stdout/JSON-stderr behavior. info --json also reaches find_user_droplet(), whose username/listing API error branches still print plain text to stdout. Please route these helper failures through the JSON error path in JSON mode and cover them with CLI tests.

I ran the full suite on the PR head: 308 passed.

[akshithg]

Thanks for the reproduction. I updated load_config_and_api() and find_user_droplet() to honor --json. Missing or invalid config and username or droplet-listing API failures now exit with empty stdout and a JSON error on stderr. CLI tests cover both config failures across list, info, and list-ssh-keys, plus the info lookup failures.

I also moved the existing list table rendering back to its original path to keep the diff focused. Validation passed: 441 pytest tests, prek run, and the manual E2E lifecycle hook.

[ret2libc]

Re-reviewed the latest head. The JSON error paths and output contract are addressed; no remaining relevant findings. Verified with 441 passing tests locally.