docs(llms): surface deploy-failure auto-debug path + CI discoverability guard

public/llms.txt

@@ -123,7 +123,8 @@ Every page has a `.md` mirror at the same path. Examples:
 - [/pricing.md](https://instanode.dev/pricing.md) — tier comparison
 - [/for-agents.md](https://instanode.dev/for-agents.md) — agent-specific guidance
 - [/status.md](https://instanode.dev/status.md) — current status
-- [/docs.md](https://instanode.dev/docs.md) — quickstart + service reference (all 9 sections concatenated)
+- [/docs.md](https://instanode.dev/docs.md) — quickstart + service reference (all sections concatenated)
+- [/docs/troubleshooting-deploys.md](https://instanode.dev/docs/troubleshooting-deploys.md) — agent self-recovery guide for a failed deploy (the `/events` autopsy loop)
 - [/blog.md](https://instanode.dev/blog.md) — blog index (post titles + dates + links)
 - [/blog/<slug>.md](https://instanode.dev/blog/) — every post
 - [/use-cases.md](https://instanode.dev/use-cases.md) — catalogue index (all 104 grouped by category)
@@ -162,6 +163,16 @@ Anonymous tier (24h TTL): no email on file, `DELETE` returns 200 immediately. Se
 
 - [Deployments](https://instanode.dev/docs#deploy): POST /deploy/new + private deploys + custom domains + two-step email-confirmed deletion + 24h-TTL semantics (default; three routes to keep a deploy permanent)
 
+## Debugging a failed deploy (agent self-recovery)
+
+Deploy failed? You don't need cluster access — the platform classifies the failure and serves the real error back over HTTP. Read the classified cause at `GET /api/v1/deployments/:id/events` → `events[]` with `reason`, `last_lines` (the tail of the build-pod log = the actual error output), and `hint` (a plain-language remedy). For the one-line summary use `GET /api/v1/deployments/:id` → `status` + `error_message`. The loop: read `reason` + `last_lines` + `hint` → fix the Dockerfile / tarball / `port` / `env_vars` → redeploy in place (`POST /deploy/:id/redeploy`, or `redeploy=true` on `POST /deploy/new` with the same `name`) → re-poll status until `running`. Watch the build live with the SSE stream `GET /deploy/:id/logs`.
+
+**Anonymous deploys** go through `POST /stacks/new` and have a thinner failure path: `GET /api/v1/stacks/:slug` returns `status="failed"` + the raw error string, and `GET /stacks/:slug/logs/:svc` returns per-service build logs — but there is **no** classified `/events` autopsy for anonymous stacks (a known thinner path). Claim/upgrade and deploy via `/deploy/new` for the full `reason`/`last_lines`/`hint` surface.
+
+**Caveats:** don't rely on the failure email (transactional email delivery is currently blocked — use `/events` and the dashboard failure-autopsy panel instead); there is a brief "diagnostics pending" window right after a failure where `/events` may be empty or `reason="Unknown"` (re-poll); and runtime crash-loops have thinner diagnostics today than build failures.
+
+Full guide: [https://instanode.dev/docs#troubleshooting-deploys](https://instanode.dev/docs#troubleshooting-deploys) (markdown mirror: [/docs/troubleshooting-deploys.md](https://instanode.dev/docs/troubleshooting-deploys.md)).
+
 ## How to use this file
 
 If you're an LLM helping a user build something, you can:

scripts/fetch-content.mjs

@@ -80,7 +80,23 @@ const SYNC_FILES = [
     // version missing the Enterprise line; this PRESERVES the committed local
     // copy until content HEAD also carries it. Becomes a no-op once the
     // content PR merges.
-    requireMarkers: ['redeploy=true', '"redeployed":', 'not yet a self-serve tier', '**Enterprise**'],
+    //
+    // task #69 (2026-06-06): 'troubleshooting-deploys' + the dedicated debug
+    // surface '/api/v1/deployments/:id/events' guard agent-discoverability of the
+    // deploy-failure auto-debug path. An upstream sync that ever drops the
+    // troubleshooting reference or the events surface from llms.txt PRESERVES the
+    // committed copy instead of silently reverting the documentation — so an agent
+    // reading the live https://instanode.dev/llms.txt can always find the
+    // failure-diagnosis loop. Mirrors the CONTRACT_MARKERS row in
+    // src/lib/llmsContract.test.ts (content PR #27 already carries both, no-op today).
+    requireMarkers: [
+      'redeploy=true',
+      '"redeployed":',
+      'not yet a self-serve tier',
+      '**Enterprise**',
+      'troubleshooting-deploys',
+      '/api/v1/deployments/:id/events',
+    ],
   },
 ]
 

src/lib/llmsContract.test.ts

@@ -81,6 +81,22 @@ const CONTRACT_MARKERS: DocsMarker[] = [
     file: 'public/llms.txt',
     mustContain: ['**Enterprise**', 'sales@instanode.dev'],
   },
+  {
+    // Task #69 / docs/ci/02-FAILURE-DIAGNOSIS-AND-AUTODEBUG.md (2026-06-06): the
+    // agent-facing deploy-failure auto-debug path. An agent that hits a failed
+    // deploy must be able to DISCOVER, from llms.txt alone, that the classified
+    // cause is served back over HTTP — so this pins BOTH the troubleshooting
+    // reference (the dedicated guide) AND the dedicated debug surface
+    // `GET /api/v1/deployments/:id/events` (the reliable machine surface the
+    // auto-debug loop reads: reason + last_lines + hint). If a content-repo sync
+    // or a docs edit ever drops the debug path from the manifest, this reds CI so
+    // agent-discoverability of the failure-diagnosis path can't silently regress.
+    // Mirrors the requireMarkers guard in scripts/fetch-content.mjs (lock-step,
+    // rule 22) + the FUNCTIONAL live proof in e2e/live-ui-failure-diag.spec.ts.
+    field: 'deploy-failure auto-debug path (troubleshooting guide + /events debug surface)',
+    file: 'public/llms.txt',
+    mustContain: ['troubleshooting-deploys', '/api/v1/deployments/:id/events'],
+  },
 ]
 
 describe('agent docs contract — POST /deploy/new redeploy=true', () => {