gtapps/forge-failed-deploys

Forge Failed Deploys

Scheduled-check skill: scans the org-wide site list and flags any sites whose latest deployment status is failed, failed-build, or cancelled.

Contract: idempotent, read-only, no self-scheduling, short-running. Returns findings or silence — never creates proposals itself.

Steps

1. Run the estate scan.

   php ${CLAUDE_PLUGIN_ROOT}/php/forge.php failed-deploys --json
   

   This pages through the org-wide site list via organizationSites()->lazy() and fetches the deployment_status field on each site. For sites in a failure state it fetches the latest deployments() detail. Rate limiting (429) is handled internally with conservative pacing.

2. Parse the JSON output. Each entry has: site_id, site_name, server_id, status, optionally deploy_id, deploy_status, commit. An entry whose detail fetch was rate-limited instead carries an error key (and no deploy_id/commit) — treat those as "failure, detail unavailable" and omit the commit line. Never assume all keys are present.

3. Output the findings block. Always output to stdout, regardless of outcome. reflect --scheduled-checks classifies the result.

   Failures found:

   forge-failed-deploys findings — <YYYY-MM-DD>
   Failed deployments: <N>
   - [deploy-failure] <site_name> (server: <server_id>): status <status><commit line if available>
   

   One bullet per failed site.

   No failures:

   forge-failed-deploys findings — <YYYY-MM-DD>
   No actionable findings.
   

4. Do not fetch deployment logs. This scan surfaces sites with failures; the forge-logs or forge-deploy skills handle remediation. Fetching logs per failure during the scan would hit rate limits on large estates.

Notes

• This skill writes no artifact. All output goes to stdout for reflect --scheduled-checks.
• Registered by /laravel-forge-hermit:hatch step 7 via a scheduled_checks config entry (interval_days: 1). The core daily scheduled-checks routine fires reflect --scheduled-checks, which picks it up once 1+ day has elapsed since last_run.
• No channel notifications — this is analysis-only. Findings surface as [reliability] proposals via the normal pipeline.
• scope: if organizationSites() does not carry deployment_status in your Forge plan/API version, the scan will report zero findings (not an error). Scope to watched_sites in .claude-code-hermit/config.json if org-wide scan is unavailable.
• Rate limit: Forge allows ~60 req/min. The scan paces conservatively; if a 429 is returned mid-scan it waits 30 seconds and the scan exits with an error — the check will retry on the next scheduled run.