Scriptify

scriptify

Freeze a flow the agent already worked out into a committed, repo-native .sh runner, so the maintainer can re-run it with no agent and no token cost. This is the mechanism the other skills call in their "offer a runner" step.

Phase 0 — Discover (reuse preflight, read-only)

Run/reuse preflight to get the exact launch, seed, and capture commands and the output location. A runner must be a recording of a verified flow — either one a prior approved bugsweep/a11ysweep/perfsweep run already executed, or one this skill verifies after the gate (Phase 3). Do not run the flow during discovery.

Phase 1 — Plan + approval gate (MANDATORY)

Present what the script will do (targets, steps, output path, where the file lands) and whether verifying it will execute the flow, then stop per the shared gate in ../beagle-shared/approval-gate.md.

Phase 2 — Emit

Generate the script per ../beagle-shared/emit-runner.md:

• Reuse the repo's existing script style/location if it has one; else portable bash (set -euo pipefail, env-parameterized, idempotent, reads bundle id/package from build output, echoes every artifact path).
• Carry the hard rules into the script so they survive without the agent: Android am start -n (never monkey), uninstall-before-install; iOS verify BUILD SUCCEEDED in the log.
• Write it to the repo's scripts/ (or wherever its current scripts live), chmod +x.

Phase 3 — Verify, then hand over

Run the emitted script once and confirm it produces the expected artifacts. A runner you didn't run is unverified — say so; don't claim it works. Then offer the script + a one-line usage note; let the maintainer commit it (don't auto-commit).