Rifts (Workgraph Autopilot) Skill

Rifts (Workgraph Autopilot)

Rifts is the umbrella workflow: Speedrift (drift + contracts) plus optional Specrift (spec/doc drift) and optional UXrift (UX checks), integrated into Workgraph.

Start / Resume (Repo Bootstrap)

Run these from the repo root:

# Ensure wrappers + executor guidance exist (idempotent)
/Users/braydon/projects/experiments/driftdriver/bin/driftdriver install

# Ensure every open/in-progress task has a `wg-contract` block (idempotent)
./.workgraph/speedrift ensure-contracts --apply

# Optional: snapshot drift into wg logs + spawn follow-ups
./.workgraph/speedrift scan --write-log --create-followups

Optional (only if you plan to do UX work in this repo):

/Users/braydon/projects/experiments/driftdriver/bin/driftdriver install --with-uxrift

Specrift is lightweight and is installed automatically when found; it runs only for tasks that declare a specrift block.

Then start “hands-off” execution:

# Workgraph daemon: auto-spawn agents for ready tasks (uses .workgraph/config.toml executor/model)
wg service start

Optional pit-wall telemetry (continuous drift monitor+redirect):

./.workgraph/rifts orchestrate --write-log --create-followups

Planning: Turn “Do The Project” Into Workgraph Tasks

When asked to “create an end-to-end plan with dependencies in Workgraph”:

Create a short root task (the why/what).
Decompose into executable leaf tasks with stable IDs.
Encode dependencies with wg add --blocked-by ....
Put a wg-contract block at the top of every task description.
Add a uxrift block only for tasks that need UX checks (so rifts will run UXrift automatically).
Add a specrift block for tasks that must keep docs/specs in sync (so rifts will run Specrift automatically).

Useful commands:

wg add "Kickoff: plan + execute" --id kickoff -d "..."
wg add "Implement X" --id implement-x --blocked-by kickoff -d "..."
wg add "Test X" --id test-x --blocked-by implement-x -d "..."
wg add "UX check flow" --id ux-check --blocked-by implement-x -d $'```uxrift\nschema=1\nurl=\"...\"\n...\n```'
wg ready

Per-Task Protocol (What Agents Must Do)

Always pass a task id (avoids ambiguity when multiple tasks are in progress):

./.workgraph/rifts check --task <task_id> --write-log --create-followups

Use it:

Once at task start (before edits)
Once just before marking done/submitting

Interpretation:

Exit 0: clean
Exit 3: findings exist (non-blocking; act via follow-ups / contract edits)

Handling Drift Without Slowing Down

Speedrift is advisory. Preferred responses:

If scope changed: update the contract touch set (don’t silently expand scope).
```
./.workgraph/speedrift contract set-touch --task <task_id> src/** tests/**
```
If hardening_in_core: don’t add retries/fallbacks/guardrails in the core task. Do the harden: follow-up task instead (or create one if missing).
Let follow-up tasks carry the drift (keeps the main task focused, preserves velocity).

Choosing Contract Mode

mode = "core": strict drift checks (scope/deps/churn + “hardening in core” signals). Default.
mode = "harden": use when the objective is explicitly guardrails/retries/timeouts (expect to add “hardening”).
mode = "explore": use for research/data analysis spikes where broad exploration is expected.

Agent Skills: Rifts (Workgraph Autopilot)

Install this agent skill to your local

Skill Files