/aitask-run-gates

Run a task’s declared gates — the conversational front of the gate orchestrator

/aitask-run-gates <task-id> runs a task’s declared gates and explains the result in prose. It is the conversational front of the same orchestrator engine that ait gates run drives — it never reimplements the decision logic, so the CLI and the skill can never disagree about what should run.

Reach for the skill when you want the outcome interpreted — which gate is blocking, what the next action is, whether the task is ready to archive. Reach for the CLI when you want the raw result.

Usage:

/aitask-run-gates 42                          # Run every unlocked gate
/aitask-run-gates 42_2                        # Child task
/aitask-run-gates 42 --gate risk_evaluated    # Force-run a single gate
/aitask-run-gates 42 --dry-run                # Report the plan; run nothing

A task ID is required — a parent (42) or a child (42_2).

Note: Must be run from the project root directory. See Skills overview for details.

Arguments

ArgumentEffect
--gate <name>Run one named gate, even if it already passed and even if its retry budget is spent. Its predecessors must still be satisfied. Parallel fan-out is skipped.
--dry-runReport the decision tree — what would run, what is locked, what is exhausted — without invoking a verifier or writing to the ledger.

What it reports

  • All gates satisfied — it says so and suggests archiving. It never sets status: Done itself.
  • A human gate is pending — it explains the action a person now has to take. It never creates the sign-off signal on your behalf; an agent signing a human gate would defeat the point of having one.
  • Retries exhausted — it points you at ait gate log for the failure history.
  • A verifier errored or the registry is malformed — it surfaces the problem rather than treating it as a failed gate.

What it will not do

The skill is advisory and orchestration only. It does not edit task frontmatter, merge branches, or archive tasks — those belong to the task workflow. Gates whose verifier is a procedure an agent must carry out (such as docs_updated) are not run here either: the orchestrator reports them as needing an agent, and the attended workflow runs them at review time. See /aitask-gate-docs-updated.