api-health

Today is ${today}. Read `memory/MEMORY.md` and `memory/issues/INDEX.md` for context. ## Why this skill exists Skills that depend on a paid API key fail *silently* when credits run out or the key is revoked — the failure surfaces only days later in a self-review, after every dependent skill has degraded. This skill closes that gap. It runs ~30 minutes before the morning batch, reads the result of `scripts/prefetch-api-probe.sh` (which probes each provider with a configured key), and notifies immediately if a provider is down — so the operator has a window to top up credits or rotate the key before the batch starts. Currently probed: **xAI** (`XAI_API_KEY` — used by refresh-x, tweet-roundup, list-digest, narrative-tracker, remix-tweets, content-performance). To probe another provider, add a guarded block to `scripts/prefetch-api-probe.sh`. ## Steps ### 1. Read probe results Read `.api-probe/status.json`. Written by `scripts/prefetch-api-probe.sh` during this workflow's prefetch phase. Format: one entry per provider: ```json { "xai": { "http_code": 200, "checked_at": "...", "error_msg": "" } } ``` If the file is missing: - Log `API_PROBE_SKIP: prefetch-api-probe.sh did not write status — likely skill misconfiguration` - No notification. Stop. ### 2. Interpret HTTP status — per provider For each provider entry, parse `http_code`: | Code | Meaning | Action | |------|---------|--------| | 200 | Healthy | Log OK, skip notification | | 0 | No key / network failure | Log...