Skip to main content

Wire Autopilot

Wire Autopilot runs Wire commands non-interactively — you give it a release folder and a target phase, and it executes the generate → validate → review cycle for every artifact in that phase, pausing only when it encounters a validation failure or a review decision that requires human input.

Autopilot is designed for the "known path" phase of an engagement — once the requirements artifacts are approved and the shape of the work is stable, you can hand the development and testing phases to Autopilot and return for the review gates.

Starting Autopilot​

/wire:autopilot <release-folder> --phase <phase-name>

Examples:

/wire:autopilot 20240115_barton_peveril_full_platform --phase development
/wire:autopilot 20240115_barton_peveril_full_platform --phase testing
/wire:autopilot 20240115_barton_peveril_full_platform --from requirements --to testing

--from and --to run all phases between two named phases, inclusive. Phases are ordered: requirements → design → development → testing → deployment → enablement.

How it works​

For each artifact in the phase, Autopilot runs:

  1. generate — creates the artifact
  2. validate — runs automated checks
  3. If validation passes: records the result and moves to the next artifact
  4. If validation fails: pauses, presents the failures, and waits for instruction

At review gates (-review commands), Autopilot presents the artifact and pauses for explicit human approval. You can:

  • Approve — Autopilot marks it Approved and continues
  • Request changes — provide feedback inline; Autopilot re-generates with the feedback incorporated and presents it again
  • Delegate — Autopilot emails the artifact link to the named stakeholder and monitors for a reply

Handling validation failures​

When a validation failure occurs mid-run, Autopilot presents a triage menu:

VALIDATION FAILURE: dbt-models-validate (students domain)
  ✗ not_null test on student_pk failing (14 rows)

What would you like to do?
  [1] Auto-fix — apply the recommended fix and re-validate
  [2] Skip and continue — mark as known issue, continue to next artifact
  [3] Pause here — stop Autopilot and return control
  [4] Edit fix — show the recommended fix before applying it

Option 1 is available when Wire has a high-confidence fix (single-cause failures with a clear remediation). Options 2–4 are always available.

Progress display​

While running, Autopilot shows a live progress board:

Wire Autopilot — 20240115_barton_peveril_full_platform / development
─────────────────────────────────────────────────────────────────────

  ✓  ingestion-generate          completed  12:34
  ✓  ingestion-validate          completed  12:36
  ✓  ingestion-review            approved   12:41 (Mark Rittman)

  ✓  dbt-models-generate (students)     completed  13:02
  ⚠  dbt-models-validate (students)     1 failure — awaiting decision
  …  dbt-models-generate (finance)      queued

  Time elapsed: 47 min   Artifacts remaining: 7

Resuming after a pause​

If Autopilot is interrupted — by a validation failure, a browser close, or a session timeout — it can resume from where it stopped:

/wire:autopilot 20240115_barton_peveril_full_platform --resume

Autopilot reads the execution log, identifies the last completed artifact, and picks up from the next one. Completed artifacts are not re-run unless you pass --force.

Review delegation​

For review artifacts, Autopilot can delegate to a named stakeholder rather than waiting for your approval:

/wire:autopilot 20240115_barton_peveril_full_platform --phase requirements \
  --delegate-reviews "[email protected]"

With this flag, each -review command emails the artifact to [email protected] with a review link. Autopilot monitors for a reply and continues once the artifact is marked Approved. If no reply arrives within the configured timeout (default 24 hours), Autopilot pauses and notifies you.

Dry run mode​

Preview what Autopilot would execute without running anything:

/wire:autopilot 20240115_barton_peveril_full_platform --phase development --dry-run

This prints the full execution plan — every command in order, with estimated duration — and exits without running anything.

Configuring Autopilot​

Autopilot settings live in .wire/releases/<release>/autopilot.yaml:

auto_fix: true                  # apply high-confidence fixes automatically
review_timeout_hours: 24        # how long to wait for delegated reviews
notify_on_failure: [email protected]
skip_known_issues: false        # re-run artifacts with previously recorded failures

What Autopilot cannot do​

  • Make judgment calls about stakeholder feedback — a stakeholder reply that contains conflicting instructions will cause Autopilot to pause and ask you to arbitrate
  • Execute destructive deployment steps (database drops, Fivetran connector deletions, cutover) — these always require manual confirmation regardless of Autopilot mode
  • Approve its own review artifacts — every -review command requires a human approval, even in fully automated runs