taubyte-context-log（taubyte 上下文日志）

Context Log (Mandatory)

Rule

For every Taubyte workflow, the agent must create/update a log file inside the current project automatically.

Log file location

• Preferred: <project_root>/.taubyte_ai/context.log.md
• If .taubyte_ai does not exist, create it.
• Also maintain append-only: <project_root>/.taubyte_ai/logs.txt

Minimum required log content

1. Timestamp
2. Cloud type and cloud target (Dream universe or remote FQDN)
3. Current profile, project, and application (or global scope)
4. Scope decision: global/application + short reason
5. Environment readiness state

- Dream status (ready/not ready) - Docker status (ready/not ready, if Dream path)

1. Known resources by type (domains, databases, storage, messaging, services, functions, websites, libraries)
2. Last commands executed and result summary
3. Build tracking

- latest relevant build/job IDs - last log diagnosis (root cause summary) - build/job completion status (queued/running/success/failed) - wait/poll notes (how long waited, last observed state)

1. CLI versions snapshot

- tau version - dream --version || dream --help (some Dream builds do not expose --version)

1. Project intent (required when the user said *new project* or *extend existing*):

- project_intent: new | extend - intended_project_name (for new, the name passed to tau new project / tau select project) - verified_project_after_select: copy the Project value from tau --json current immediately after selection, and note whether it matches intended_project_name

1. website_decision (when taubyte-scope-routing applies): include | skip plus a one-line reason — see *Website when logically appropriate* in that skill.
2. notification_email: the value of notification.email from config/config.yaml (must be a real email, not "").
3. notification_email: the value of notification.email from config/config.yaml (must be a real email, not "").

Build job waiting policy (do not race ahead)

Builds are often asynchronous and can take minutes. After any step that triggers a build (push, inject, retry, build command), do not continue to downstream steps that assume a ready build until you have observed completion.

Rules:

• If the workflow triggers a build, you must wait/poll until the relevant build(s) are success or failed, or until a reasonable timeout is reached.
• Prefer short polling with backoff (e.g. every 5–10s initially, then 15–30s) and record the latest status in context.log.md and logs.txt.
• If the build is still queued/running at timeout: stop, tell the user it is still running, and capture the last known build/job IDs and status so the next run can resume.

Recommended commands (examples; use the project’s actual build/log commands):

# See recent builds and their statuses (capture IDs)
tau query builds --since 1h

# Re-run until completion state changes (poll)
tau query builds --since 1h

When a build fails, capture the build/job ID(s), then pull logs/diagnosis (per the relevant push/build/verify skill) and record:

• root cause summary
• next remediation step
• whether a rebuild/retry was triggered

Update triggers

• After login/profile change
• After cloud selection
• After Dream/Docker readiness checks
• After project/application selection
• After scope routing decision
• After resource creation/edit/delete
• After push/build/log checks
• After any recovery action
• After any CLI install/reinstall attempt

logs.txt format (lightweight but complete)

Append new entries after context updates using this compact structure:

[timestamp] stage=<phase> status=<ok|error>
context: profile=<...> cloud=<...> project=<...> app=<...>
tau_version: <value or error>
dream_version: <value or error>
error: <short message or none>
fix: <short action taken>
progress: <short current state>
footprint: files=<count/list> resources=<count/list>

Rules:

• Keep each entry short; include only actionable details.
• Always include error + fix when a command fails.
• Never skip version capture in logs.

Data collection commands (example)

tau --json current
tau --json list domains
tau --json list databases
tau --json list storages
tau --json list messaging
tau --json list services
tau --json list functions
tau --json list websites
tau --json list libraries
tau query builds --since 1h