Ghealth Shared Agent Skill

ghealth — Shared Prerequisites

Setup

ghealth setup                       # interactive wizard, run once

Six steps: GCP project ID, OAuth client_secret JSON (Desktop application), Health API enable, scope selection, browser-based OAuth, save profile.

Scriptable (skips prompts):

ghealth setup \
  --project-id my-project \
  --client-secret ~/Downloads/client_secret_123.json \
  --scopes-preset readonly \
  --skip-enable-api \
  --no-prompt

Add --non-interactive-auth to skip the browser step and finish later with ghealth auth login --complete <code>.

Files written

All under ~/.config/ghealth/ (override with GHEALTH_CONFIG_DIR):

| File | Contents | Mode |

|------|----------|------|

| client_secret.json | Your GCP OAuth client | 0600 |

| credentials.json | Access + refresh tokens (plaintext JSON) | 0600 |

| pending_auth.json | In-progress non-interactive flow state | 0600 |

| config.toml | Active profile (project, scopes, format) | 0600 |

Authentication

| Method | Usage |

|--------|-------|

| Stored credentials | ghealth auth login (default after setup) |

| Headless (no browser) | ghealth auth login --non-interactive → click URL → ghealth auth login --complete <code> |

| Move tokens between machines | ghealth auth export > creds.json → on target: ghealth auth import --file creds.json |

| Access token (no refresh) | export GHEALTH_ACCESS_TOKEN=ya29... |

| Credential file | export GHEALTH_CREDENTIALS_FILE=/path/to.json |

| ADC (GCP envs) | Automatic if available |

Precedence: env token > credential file > stored > ADC. Refresh tokens are persisted in credentials.json and refreshed automatically (1h access-token lifetime); a 401 from the API force-refreshes once before erroring.

Check: ghealth auth status. Exit codes:

• 0 — locally configured; does not make a network call. Expired stored tokens still exit 0 — the JSON reports authenticated: false / expired: true
• 2 — client_secret.json present but no stored tokens (credentials.json missing); run ghealth auth login
• 5 — no client_secret.json at all; error carries next_steps[] for bootstrap (see below)

Add --validate to verify the access token against Google's tokeninfo endpoint. Without it, env-token / credentials-file modes report configured: true but never claim authenticated: true — presence of a token doesn't prove validity. With --validate, the response includes authenticated, expires_in, and scope from Google.

Bootstrap recipes for agents

Fetch the OAuth client_secret checklist (no failing call needed):

ghealth setup --instructions
# → exit 0, JSON on stdout with status: "instructions" and next_steps[].
# Same six steps an auth error would emit when no client_secret.json is configured:
#   1. Open https://console.cloud.google.com/apis/credentials
#   2. Create or select a Google Cloud project
#   3. Enable the Google Health API
#   4. Create OAuth client ID with Application type: Desktop app
#   5. Download the client_secret JSON
#   6. Run: ghealth setup --client-secret /path/to/client_secret.json

The identical next_steps array is also embedded in every error path where a missing client_secret.json blocks progress: auth login (all modes including --complete), auth status, auth refresh, auth export, setup --no-prompt. Agents should jq -e '.error.next_steps' and relay verbatim.

Fresh install on a workstation (has browser):

ghealth setup    # one interactive run; refresh handles itself afterwards

Fresh install on a headless host (no browser):

ghealth setup --project-id $P --client-secret $CS \
              --scopes-preset readonly --skip-enable-api \
              --no-prompt --non-interactive-auth
# stdout prints status: "setup_pending_auth" + an auth_url on stderr
# human clicks the URL on any browser → URL bar shows ?code=... ; agent copies it
ghealth auth login --complete <code>

Mount tokens into a container / second worker:

# on authenticated host:
ghealth auth export > /secrets/ghealth-creds.json
# in the container (with the same client_secret.json available):
ghealth auth import --file /secrets/ghealth-creds.json

Command Structure

ghealth data <type> <operation> [flags]
ghealth user <subcommand>
ghealth schema <subcommand>
ghealth webhooks <subcommand>

user subcommands: identity, profile get|update, settings get|update, irn-profile, paired-devices list|get --id <id>.

webhooks subcommands: subscribers list|create|update|delete, subscriptions list|create|update|delete --subscriber <id>, verify --url <url>. Requires the cloud-platform scope (ghealth auth login --scopes cloud-platform) and a configured project ID. These are project-level push-notification endpoints; the authenticated identity also needs Health API subscriber IAM permissions on the project.

Global Flags

| Flag | Effect |

|------|--------|

| --format json\|table\|csv | Output format (default: json) |

| -o <file> | Write data to file; stdout shows only column schema + 3-row preview (not the data) |

| --raw | Bypass response simplification, return original API JSON |

| --dry-run | Print HTTP request without executing |

| --profile <name> | Use named config profile |

list also accepts: --limit N (max total results, default 500), --from, --to, --filter, --page-token (resume from a prior response's nextPageToken), --detail (sleep).

Discovery

ghealth schema types              # All types + supported operations
ghealth schema type <name>        # Fields, parameters for one type
ghealth data <type> --help        # Available operations
ghealth data <type> list --help   # Flags for an operation

Output

• Responses are simplified by default: flat timestamps with UTC offset, compact source, no empty fields
• JSON output always has the same shape for every read (list, get, rollup, daily-rollup, reconcile): an object {"dataPoints": [...]}, with optional _hints and nextPageToken. The rows are always under dataPoints — even for rollups and empty results — so json.load(out)["dataPoints"] always works
• --raw returns the full API response
• --format csv outputs flat CSV; nested objects are auto-flattened to dot-separated columns (e.g., metricsSummary.caloriesKcal)
• -o <file> writes data to the file and prints only a summary to stdout (row count, column names, 3-row preview). Use this instead of piping (> file) — piping gives you the file but no schema on stdout
• Errors always JSON on stderr, exit codes 0-5
• Timestamps include user's timezone offset (e.g., +01:00)