everettjf/copyapp/copyapp-start

Start Copy APP with a coding agent

Help the user run Copy APP from their request through the requested deliverable. Work in the Copy APP checkout and read AGENTS.md. This skill is a repository instruction file: the agent reads it directly, so the user does not need to install a provider-specific extension or type every CLI command.

Prepare the Mac and identify the target

If the repository has not been checked out, clone https://github.com/everettjf/copyapp into a new directory and work there. Reuse an existing checkout when the user has selected it; preserve its local changes. Do not overwrite another project to create the checkout.

Use the repository's .venv/bin/copyapp. If it is missing, run sh scripts/bootstrap-macos.sh. Read setup (中文) for additional prerequisites required by the requested phase. Complete available Mac-side setup yourself; do not hand the user a list of commands the agent can execute.

Run .venv/bin/copyapp onboard --stage survey to inventory USB phones and basic readiness. Identify the exact phone, then use .venv/bin/copyapp apps --udid <UDID> --all to resolve the user's app name to its Bundle ID. Use an available foreground-app observation to disambiguate when possible. Ask only when the target or device remains ambiguous; do not arbitrarily choose a phone or substitute a sample app.

For an ordinary iPhone, an inactive jailbreak, or an explicit request to avoid jailbreak services, read standard mode (中文) and use onboard --mode standard. Its UI-first analysis/report/delivery contract is separate from the historical deep-analysis contract below. Set up a valid development-signed iCTRL runner with explicit USB transport, then verify actual XCTest actions and exact-device screenshots. Do not repair SSH, Frida, TrollStore or tweaks for that route. A rebooted phone with residual jailbreak files is an inactive-jailbreak test device, not a pristine ordinary-phone acceptance sample. Named-device limitations and unfinished real-credential/second-device gates remain explicit.

Use analyze or prepare-app --mode standard --script ... for selected UI evidence and review its TASK.md into ios_behavior_report plus analysis-delivery; continue through independent development-signing installation and paired review. For one current-session parent use copy --mode standard --agent-provider current, with supplied scripts, accepted standard-mode discovery or --auto-discover. Complete each handoff yourself and resume the exact job. The strict parent still requires its full device delivery; a fidelity milestone alone does not complete it. Package inputs and offline credentials are optional host evidence and cannot establish phone runtime identity from version/build alone.

Follow the survey skill for the initial verified evidence. If only USB capture is available, have the user open the target app when needed and collect survey --ui-backend usb without --launch; a USB screenshot probe alone does not establish SSH, Frida or UI-control readiness.

Some prerequisites require the user on the phone: unlocking it, trusting the Mac, confirming its SSH fingerprint through the documented first-use process, approving permissions, or supplying signing access. State the concrete missing action and continue independent Mac-side work. Follow the pinned-key checks in the setup guide. Do not claim that installing the Mac CLI installs the phone services or jailbreaks the device.

Continue to the requested outcome

• First check only: deliver the verified survey, device/app identity, screenshot and remaining setup needs. Stop here when the user requested only a connection or installation check.
• Analyze: follow copyapp-analyze, using onboard --stage analyze and the guide's ordered repairs where necessary. Finish the reviewed, cited deep report and feature-spec.json; analyze/analysis.md alone is a collection index.
• Copy / analyze and rebuild: follow copyapp-copy. Discover representative flows when none were supplied, review the analysis, freeze the known-capability target, implement an independent app, then build, install and compare it on the exact phone. Keep deferred, blocked and unknown behavior explicit.
• Rebuild from an existing report: follow copyapp-rebuild, preserving the report's source identity and the user's selected scope.

Inside Codex or Claude Code, do the agent work in the current session. Use copy --agent-provider current when that workflow applies and complete its handoff tasks yourself; do not launch a nested coding-agent CLI. Commands that pause at a handoff have not finished the user's task.

Save device evidence and credentials only in ignored local cases/, verify every collected or composed case, and show the resulting report paths. Report what was actually observed and tested. A passing selected-flow comparison is not a claim of whole-app equivalence. Read current status for named-device limits rather than assuming compatibility.