# Diagnosing Dream Builds > Diagnoses Dream local-cloud builds when `tau list/query builds` is empty or unreliable, by hitting the jobs HTTP endpoint directly (`GET /jobs/`, `GET /job/`) using the GitHub token from `~/tau.yaml`, then downloading logs with `tau query logs --jid`. Use when Dream builds appear silent, the build table is empty after `dream inject`, or you need raw job ids and logs for a failing build. - Skill: `taubyte/diagnosing-dream-builds` (Agent Skill) - Install (CLI): `npx skillmds@latest add taubyte/diagnosing-dream-builds` - Raw SKILL.md: https://api.skillmd.com/api/skills/taubyte/diagnosing-dream-builds/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: taubyte (https://skillmd.com/u/taubyte) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/taubyte/diagnosing-dream-builds --- # Diagnosing Dream Builds ## When to use - Just ran `dream inject push-all` / `push-specific`, but `tau query builds --since 1h` shows nothing. - Need raw job ids that `tau` isn't surfacing. - A build looks failed but `tau` won't print its logs. - Verifying that a `push-specific` actually enqueued a job before chasing other failure causes. ## Mental model `tau list/query builds` is a convenience view. On Dream local clouds it isn't always wired to local jobs. The **source of truth for Dream jobs is the local jobs HTTP endpoint** exposed by Dream's auth/seer service. Once you have a job id from there, `tau query logs --jid ` can download the logs. **Token health:** if **`GET /jobs/`** (or related Patrick HTTP) returns **`401`** or **`invalid Github token`**, refresh the GitHub PAT in **`~/tau.yaml`** / `tau login …` first — see [authenticating-taubyte-cli](../authenticating-taubyte-cli/SKILL.md). A stale token looks like a "Dream is broken" problem but is only auth. ## Workflow ``` Progress: - [ ] Step 1: Get the project id from config/config.yaml - [ ] Step 2: Discover the jobs endpoint port - [ ] Step 3: GET /jobs/ for the JobIds[] list - [ ] Step 4: GET /job/ for an individual job object - [ ] Step 5: tau query logs --jid to fetch logs ``` ## Step 1: Project id ```bash awk '/^id:/{print $2; exit}' config/config.yaml ``` The project id is the value of `id:` at the root of `config/config.yaml`. You can also get it from `tau query project --json`. ## Step 2: Jobs endpoint port ```bash dream status universe ``` Find the service HTTP port that answers **`GET /jobs/`**. Docs often say **auth/seer**; on some multiverse layouts **`patrick@` HTTP** responds to `/jobs/...` while another service returns `404`. If the first port you try does not serve `/jobs`, probe other **`dream status universe `** HTTP ports until you get `{"JobIds":[...]}`. **Ports are per-run** — rediscover every session. Optional CORS preflight (mirrors what the Console does): ```bash curl -i -X OPTIONS "http://localhost:/jobs/" \ -H "Access-Control-Request-Method: GET" \ -H "Access-Control-Request-Headers: authorization" \ -H "Origin: https://console.taubyte.com" ``` ## Step 3: List job ids for the project ```bash TOKEN=$(awk '$1=="token:"{print $2; exit}' "$HOME/tau.yaml") curl -sS "http://localhost:/jobs/" \ -H "Accept: application/json" \ -H "Origin: https://console.taubyte.com" \ -H "Authorization: github $TOKEN" ``` Expected JSON shape: ```json { "JobIds": ["", "", ...] } ``` If the array is empty, the inject didn't enqueue a build — go back to [triggering-dream-builds](../triggering-dream-builds/SKILL.md) and confirm bootstrap (`push-all`) and any required website/library [registration](../registering-dream-repositories/SKILL.md). ## Step 4: Inspect a single job ```bash TOKEN=$(awk '$1=="token:"{print $2; exit}' "$HOME/tau.yaml") curl -sS "http://localhost:/job/" \ -H "Authorization: github $TOKEN" ``` Useful fields in the response: - `meta.repository` — which repo triggered the build (helps spot wrong-repo injects). - `Logs` — a map of CIDs keyed by timestamp / job id; usually consumed via `tau query logs` rather than directly. - Status fields — confirm whether the job succeeded, failed, or is still running. ## Step 5: Download logs via `tau` ```bash tau --defaults --yes query logs --jid --output "$HOME/projects/logs_" ``` This writes log files into the output directory. Read them like any text file. ## Reusable helper snippet ```bash dream_job_ids () { local jobs_port="$1" local project_id="$2" local token token=$(awk '$1=="token:"{print $2; exit}' "$HOME/tau.yaml") curl -sS "http://localhost:${jobs_port}/jobs/${project_id}" \ -H "Accept: application/json" \ -H "Authorization: github ${token}" \ | python3 -c 'import sys,json; print("\n".join(json.load(sys.stdin).get("JobIds",[])))' } ``` ## Decision tree: build is "missing" ``` tau query builds --since 1h └── empty → curl /jobs/ ├── empty JobIds → inject didn't fire │ └── re-run dream inject push-all (bootstrap) — see triggering-dream-builds │ └── for website/library: registering-dream-repositories first └── non-empty → pick newest jid → tau query logs --jid ├── log shows compile/build error → fix code/config and push again └── no logs / cancelled → re-run inject and re-check ``` ## Gotchas - **Don't trust `dream inject` exit codes.** They can be `0` even when the inject didn't enqueue anything. The jobs endpoint is the truth. - **Dynamic ports.** The jobs endpoint port changes per Dream run; always rediscover via `dream status universe `. - **Token from `~/tau.yaml`.** Use `awk '$1=="token:"{print $2; exit}'` instead of pasting the token into shell history. - **`tau query build`** (singular) and `tau cancel build` / `tau retry build` may print `No help topic for ...` even when `tau query job` works in the same session — don't rely on those subcommands; use `tau query job --jid` and the HTTP fallback. ## Related skills - `triggering-dream-builds` - `registering-dream-repositories` - `inspecting-dream-status` — find the jobs port - `authenticating-taubyte-cli` — where the token comes from